Would you want a car that protected just you - or as many lives as possible? Who do you think the car should hit? What about the legal issues like sleeping and drinking while riding?
Another legal issue: Say you have a self-driving car and it runs over and kills a pedestrian due solely to imperfection in the software. Who is at fault? It could be the person in the driver's seat, but they have no control over the accident happening, attentive or not. It could be the manufacturers, but blaming them for deaths may penalize progress that would ultimately save more lives than it costs. It could be nobody, but the driver and manufacturer offered a consent to the inherent risk that the victim didn't. Furthermore, if a car had to choose between killing two drivers and it could access information on who they are... should it consider factors like age, criminal record, impact on society, number of loved ones, and so on... or should it make no effort to compare the value of two human lives and decide by the software equivalent of a coin flip?
The car should act like a moral human being would. Do its best to save both. Morals can be taught. Legally, manual cars will have to be phased out over about 20-25 years and it will have to be illegal to drive one, unless on special courses built for leisure driving.
I had considered adding things like age, criminality, societal impact, occupation, and even race to the video. There are many versions of the trolley problem that include these things. But I figured I wouldn't muddy the water too much.
I'd want the car to save as many lives as possible. I think. Otherwise you'd survive the crash but you would have to live with knowing that you survived because a computer chose your life over another person. Then again during the situation my brain/instincts would obviously do what ever it could to save my life. Hopefully Jesus would just take the wheel.
One huge problems with a dog driving a car, I imagine, is that a dog thinks it is just a game. Once the dog grows bored of it, does the dog realize that once the dog no longer steers the car, that the car will hit the wall or something, and that that will not be fun anymore?
It seems like an "easy" way to alleviate the intersection issue is to replace them with roundabouts and other road designs that don't have the vehicles turning across oncoming traffic.
Yes. Also I'd like to see all self driving cars default to NEVER hitting a pedestrian. Even if it slows traffic down. We should prioritize walking. It causes minimal environmental damage, is much less expensive to provide as an infrastructure, and is much healthier, which also reduces societal costs.
@@brysonwest93 First thing to do in regards to climate change? Make cruise's illegal under international law. Those things pump out the equivalent pollution of 5mil cars.
Just a small point here. I'm actually a mathematical programmer and work primarily in computer vision. You actually did very well in this video,. You hit all the main points with out dumbing them down. Although it is still research level, we DO have a full mathematical theory for object recognition which works well. The problem is, computing these "shape spaces" as they are called is currently too slow for real time applications, but we're working on that. In fact, most "problems" with self driving cars are solved theoretically. It is just not cost effective to have a super computer in every car.... yet. :) As always great video!
Again, as always it's a little scary when someone starts off a comment with their credentials of how they're demonstrably more of an expert than I am. Glad to know I got most of it right - the last time I learned anything vision related was only a year or two ago so... I hope nothing I said was too out of date. I hadn't seen you comment in a while I was worried I scared you off somehow, glad you're still around :)
A Tx2 by Nvidia cost 500$ and can do trillions of calculations per second. AI doesn’t rely on calculating shape spaces. It can learn what objects are by just seeing enough examples of the object
Jacob Richards Deep learning is not a mathematical theory of shapes. It is an engineering technique without any formal mathematical theory (i.e., actual theorem proof style justification). I actually work daily with the TX2 and trust me, there is a reason self driving cars do NOT run off the TX2. Yes nvidia has papers the run a car off the TX2, these are toy examples and once again there is a reason no one uses a single TX2 for full autonomy. I was responding directly responding to KnowingBetter's point regarding object recognition. You are right, no one uses shape spaces in self driving cars, because the inference time for some deep neural net is much much faster than optimizing over a shape space.
Cloud computing is exactly what I think programmers and corporations would resist and screw everything up because stand-alone robotics is much more expensive but much more practical for personal robotics like self driving cars.
11:47 "There is no scenario where a computer couldn't make a decision." As an avid video game player who has seen countless AI characters stand there and stare as I'm getting my ass kicked, I have to dispute this.
@@dontgankmebruv7273 well said, Though it's very interesting to think about how an ai would react to actual phsyical limits as opposed to literally what you said which is again ...very true
Checkers is only weakly solved, meaning that a strategy exists to achieve the perfect-play result of the game (see Schaeffer et al, "Checkers is Solved"), and this was not do by memorization of every game state as you say. I think you are underestimating the size of 5*10^20. Squeezing by with just one byte per game state, that would still require 500 exabytes, to put that in perspective Randall Monroe estimated that all of the data in Google's datacenters only numbers 15 exabytes. Similarly, chess is not solved *at all*. Even if modern AIs are superior to the human grandmasters, that does not mean the game is "solved" in a mathematical sense. No one knows if white can force a win, or black, or if perfect play will result in a tie.
Thank god someone else pointed this out. I basically gave up on the video at this point because it's just a really bad look when an "informational" video comes out of the gate giving you background info that's completely factually wrong. I mean I'm not a tenured professor of game theory or anything but I feel like that's such a crazy thing to think that chess had been "solved" a few years ago. The maximum number of potential chess games isn't even a value that's been precisely calculated, it's estimated to be greater than the amount of atoms in the observable universe.
A minor point Chris, but chess games can end in "draws," not "ties." (A tie is a decision in which each players receive an equal score, such as can happen in Reversi or golf, while a draw is an ending that is not decisive, such as can happen in chess or cricket.) However, two chess players at the same tournament could tie in points. For instance, if I scored two wins, one loss, and two draws, while you scored one win, no losses, and four draws, we would be tied with 2½ points each (1 for each win and ½ for each draw).
This is a misleading sentiment and is a nitpick that misses the comparison he was obviously going for. Checkers, Chess, Go, etc are all effectively simple games with fixed and relatively simple rules that a computer can use to competently play the game as well or better than humans (which is his definition of "solved" here). Whether or not the games are literally "solved" is not actually the point. Self driving cars are a much more complicated problem for computers to "solve" better than humans
@@TheMysteryDriver Horses do not step on people on purpose. You have to hide the human under a blanket or similar to be able to get the horse to step on the person.
I had my first driving lesson after I got my learner's permit yesterday. It was the first time I had ever been on the highway and it shocked me how easy it was
Self-drive cars will communicate, so why would a rogue self-drive car be skating around on the ice? The other self-drive cars would have told it of the hazard and to slow down. "Who to hit?" becomes less of an ethical issue when crashes become much more rare. But I do think it is a huge question that needs some answers, "Who will the self-drive car be most loyal to?" The auto manufacturer? Stupid government mandates? The owner? The safety of people in general? Will it rank human lives as far more important than pets or wild animals? "Could you please drive a little faster? My wife is about to give birth." "Sorry, I am mandated to not go even 1 mph over the speed limit. Regulations."
Thank you I'm a chauffeur I've been told that in five to ten years all cars will be driverless Drive through Manhattan then tell me how a self driving car is going to get though all those situations People don't use walk ans don't walk they just go If a person knows a self driving car won't move with them in front of it People will just walk in front and the car will never move
Agree with this but also people erroneously think car automation is like automation in other industries where jobs are replaced. Uber and Lyft drivers are NOT employees, they are independent contractors, so there will be no automation savings for replacing a person with an automated car. (see accounting 101). Prices for such vehicles will definitely be higher; manufacturers will need to protect themselves from liability, not to mention recouping R&D costs. Aside from that, currently every vehicle has a driver who is responsible for loss and also provides other essential services like vehicle maintenance and attention to passenger comfort. This is true for mass transit, limo services and private chauffeurs. I just can't picture any wealthy person giving up personalized service to mess around programming their destinations in their automated dashboards and being libel for all losses whatever the root cause.
@@Val_Emrys what? Uber and Lyft still has to pay their independent contractor drivers. We don't have enough information to know whether self-driving cars would save them money, but Uber obviously thinks so.
@@chaklee435 As I said, they are paying independent contract drivers now, but switching to automated cars will force THEM to be the owner operators who will have to provide maintenance, programming, and insurance. I am not even factoring in the purchase of cars. Do you really think independent contractors are going to be the ones to buy autonomous vehicles, carry all the liability and upkeep and lease them to Uber of Lyft for the same rate as today? PS. Uber and Lyft know full well there is no benefit in forcing a conversion to self-driving cars, but they are getting lots of tax breaks for 'researching' it.
I might be a little late, but I want to add something. Radar can detect the radial velocity of moving objects by using the doppler effect (redshift/blueshift). You only need one radar puls for that. It is used for primary air radar and can detect the speed, altitude, and velocity of planes with disabled transponders.
Russia has one of the worst roads in the world and it will stay that way forever given the density of population and harsh weather. Besides, Russia nas no highways, Russia has highways to hell.
russia: the only country where you can spot the sober guy because he drives slalom and buy alcohol in a gun shaped bottle (seriously its called vodka 47 ande is delivered in a ak47 shaped bottle).
OK, OK, hear me out. The answer to all of self driving cars problems is: Ejector seats. Your car going to crash? No problem, ejector seat you to safety. Car needs to decided between killing two people or ramming itself into a ditch, just ejector seat you first. Car needs to be used as a shield, just ejector seat you first. Assholes tries to override the autopilot to break traffic laws? NP, ejector seats. I await my 5 million dollar check from Tesla in the mail.
When I was a kid in the 80's everyone told me we'd have flying cars by the year 2000, boy did their prediction suck! Even "educational" programs were on the team for this...what a terrible idea that would be! Why? Because 2 dimensions is the max for most drivers, and a fender bender at 100 feet would be at list a tiny bit more deadly!!
No one would accept the training requirements and maintenance of the flying car. We already have it today and it is called aircraft and helicopters and they are not directly common.
I mean- flying cars are relatively simple, but, like you pointed out, a lot of future predictions don't let feasibility get in the way of dreams. Now maybe, if we ever implemented some measure of VTOL that didn't require venting a turbofan engine, then something as useful as the kind of flying cars in BTTF might at least be plausible. But, that still doesn't solve enforcing lanes in 3 dimensions, or mechanical issues. Every car on the side of the road now, would turn into a 1 ton plummeting object, that would land on whatever or whoever is under it. Now, maybe some kind of emergency parachute, warning system, and directional thrusters could land the vehicle in a safe location, while allowing people on the ground to keep clear, but even that could fail.
@@digitalutopia1 pfff ez you just call up atc request your altitude and done...lol rly tho imagine the shitshow of any type of attempted regulation...not to mention the livid pilots everywhere
@@petrovstalingrad3994 1) Are you talking about dual-clutch? That's basically paddles. Some of us have considered it but it's regarded as an empty feeling. Sorry but manuals just offer more of a feel.
@@juanmanuelpenaloza9264Yeah I know what you are saying. I used to drive a 1990 Ford Taurus. I love shifting but my style choice of cars don't really have manual transmissions.
Barring the moral questions, one thing people tend to forget is that self-driving cars do not need to be perfect. They only need to be better than humans. That is a much easier goal to reach.
It's a fair point, one that I have often made myself, however if we can do better than humans then why can't we do better still? That pressure will always exist. The doing better than humans 'metric' is really the theshold for allowing them on the road in the first place. It shouldnt be the end of the safety improvement drive.
AI fan self driving cars already beat humans in safety today. The companies testing self driving cars have to put out monthly safety reports in the US and they are safer than people already.
Society as we know it is based on a liability system...who will eventually be liable for any incidents that occur with self driving technology? The inventors? The developers? The programmers? The company selling the tech? The human who turns the self driving car from off/to on and presses "ACCEPT LIABILITY" terms of use button? :-\
But self driving cars killing select people based on software choices is a lot harder so swallow than blaming a human for inattentiveness: "Sorry mrs., the Ford software installed in the car that killed your husband, calculated that running him over was less liability than flying over the embankment into the wetland protection zone."
zach, as an AI researcher working in self driving technology research, I have read some of those reports. You'll find that they are operating under a narrow set of conditions, and yes various companies are productionising those systems right now. You can expect to see level 4 systems for sale in the next 4 years, though probably limited to highways initially.
I like this guy already. Beside the fact that he is smart and knowledgeable, I like the fact that he recorded this video inside a car when he was NOT driving, which means he puts safety first.
I'm glad someone else made this video for me, because self-driving cars are virtually impossible to get right. You said it best when highway driving proves nothing - intersections are where the vast majority of car accidents occur. Have a self-driving car make it down a street in Manhattan. Then I'll be impressed.
This is not confusing. The traffic light takes precedence. The stop sign is covered, meaning ignore it. The detour sign is not the beginning of a detour, it is the middle of a detour, and is relevant only if you’re following the detour.
But the car is not mortal!?! But that probably doesn't count. Though I suppose it doesn't sound quite as grand to say can only be killed by an immortal. But then can they NOT be killed by another immortal? If they can, then they really are not immortal. Confusing stuff, magic and fantasy.
rationalmartian If the Lich King who says he can be killed by no man is immediately decapitated by a woman, then a self driving car can definitively kill Celty. The devil is the details as they say, and arguing semantics odesn't matter when dealing with ancient mystic powers.
@@kyriss12 Ackchyually, it was Merry killed him with a magic Dagger of Westernesse. It was enchanted with the power to harm the Witch-king of Angmar himself by a weaponsmith of Arthedain long before. When he stabbed the Witch-king in the knee with it, it distracted the Nazgûl and broke the spell binding his undead flesh to his will, allowing Eowyn to kill him by driving her sword into his unseen head, thus fulfilling the prophesy of Glorfindel that "not by the hand of man shall he fall." Literally anyone trying would have only cut the cloth of the wraith's hood before dying to said chuckling wraith if not for the relic Arthedain blade from a country destroyed thousands of years earlier. This said, I think a Dullahan can only be killed by destroying the head, which Celty keeps very safe in the darkest reaches of Area 51. The Headless Hessian Horseman of the Hollow however already had his head smashed apart by cannon-fire, but you can chock that up to differences in regional lore and myths.
Although any good chess program can beat any human, the game of chess was never solved, not even weakly. Prove that you solved the game would require that you provide the entire possible trees of all playable games. For chess this would be hundreds of Petabytes of data. Nobody ever did that. There are some allegation on Internet, but never a reference was provided (just like you, that provided an apocryphal date, but no reference whatsoever). Some types of endgames are know to produce some results, but that is the nearest that anyone reached. Checkers was solved by Jonathan Schaeffer, after a 16 years work.
Petabytes is still a huge, huge underestimate.To memorize every possible game state will take so much space that we couldn't even get close to doing it if we combined every single bit of memory ever. The amount of Chess positions is such an absurd number that it mights as well be infinite, human minds just can't comprehend that big of a number. Technology is advancing at a terrifying rate, but I doubt we will even get close in the next couple of decades.
@@josephhanicak7922 Yes it is. Johanthan Schafer that I mentioned said: "The possible number of chess games is so huge that no one will invest the effort to calculate the exact number." It is estimated to be hundreds of orders of magnitudes higher than the atoms of the Universe.
@@agranero6 Oh yeah. There are estimated to be 10 to the power of 187 atoms on the universe. Thousands of times that is such an absurd number that you could reasonably use infinity instead
01:34 chess wasn’t solved in 2015 and it’s speculated that it’ll never be solved. Despite being easy for a computer to calculate game states, we do not have the computing power, the storage or the time to fully solve it.
The point is not to make cars that won't make mistakes, but to make diverless cars better than the average driver. Accidents will occur just like with manually driven cars. Many situations stated in this video are examples of accidents that happens due to human error when human drives as well. But yes we're not there yet with driverless cars but It's just around the corner I think.
And sadly "average driver" is a such a low bar that driverless cars are pretty much already better. They may popularize FAST because : -Cheaper to produce, cheaper insurance, cheaper (and easier !) license. -LOTS of people can't drive ! -Commute time becomes free time. -Governments incentive. Reduced accidents, les wear, less infrastructure needed.
I mean, I feel like that doesn't really address most of the points he made in the video though. Most of the problems he talks about are reasons why a human driver will be better for a decent time longer now.
@@telcharthegreatsmithofthef7585 yes but self drive doesn't have to be perfect to replace human drivers. Just removing tired, drunk, high angry, shit human drivers. It's just a matter of time, then nobody will own cars as what's the point, u just have taxis.
If I'm in a self driving car, and it for some reason decides to strike a person or vehicle, would your insurance policy be affected, or will your vehicle's manufacturer pay for both damaged vehicles/medical bills/funeral bills?
But what about responsibility? In a human controlled accident it is almost always clear who is at fault but what about a driver less accident? Is it the car company? The person behind the wheel? The other driver? What if no one is in the car?? Who is at fault in a self driving vehicle with no one inside
I would disagree, alphazero has solved chess. I'm sorry to burst your bubble. In just 4hrs of playing itself it beat stockfish in a 100 game match never lost a single game.
Not losing a single game doesn't mean the game has been SOLVED. To "solve" chess means to have determined the outcome of *every single possible* chess game. At last count, doing this would require more bits of data storage than the number of atoms in the observable universe.
@@rogermwilcox inaccurate, I don't need to process every possible 1st or 2nd move in tic tac toe to understand to force a draw, mark the center. (I am using a very basic concept for this) It is well known there are irrelevant moves in chess in the first 2 moves (1w1b) that don't even need to looked down. significantly reducing the branching possibilities. also, Alphazero has millions of years worth of chess experience.
@@MTDurkee Sorry but being good or great or unbeatable like Alpha does not mean the game has been solved. All moves (as he pointed out in the video about checkers) have not been cataloged. Same with go. Just take the 'L' Matthew.
I have to say I giggled so hard at the outro this time, that was too perfect, especially since I had to explain why I went nuts at the addition to the intro making me VERY happy to my roommate!
Thanks for reviewing. We are developing the self drive out of the air. In contrasts, I think it much better to make it in rail for the real safety, but there must not be an actual rail on the road. As a developer, I don’t trust only the sensors much, unless car can talks each others during commuting.
Ivailo Yanev won't.be intersections, just flyovers, that's happening now.in my city a lot if major.cross road intersections with light as being replaced with continuous over passed.
Honestly, with all the automatic assists new, expensive cars have nowadays, they are almost self-driving. I remember some months ago a story about a guy who drove halfway through Poland while being so drunk he was unconscious more often than not, but with all the things that nudge your steering wheel and brakes according to cameras seeing road stripes, how close you are to other cars and stuff, he was totally fine. Until there was a damaged piece of road with road cones on it, and the automation didn't know how to deal, then he crashed. The car doesn't really need to 'see' all around to drive semi-automatically. It needs a clearly marked road and other cars to behave predictably and a human that can occasionally intervene when something other happens.
Another big problem with driverless cars becoming the norm is cultural. You kinda touched on this in the video. The United States for example is a car culture (from what I understand Germany is the same way) that enjoys driving and it will be a tuff sell to say "hey let this car drive you for you!" The average American (and I assume German) would say "Nope, no thank you. I prefer to drive myself." For driverless cars to work it would require a cultural shift or government intervention or both. And if we can't take guns away good luck with cars. Not to mention we'd have to spend trillions of dollars to get the all the roads, cities and high ways updated (mentioned in video) and we cannot get the funds today to just maintain our current roads. So forget driverless cars becoming the norm in 30 years, try 100 years assuming a lot of luck is involved.
people used trains and the main reasons many used horses (mainly in the country side) was because they were cheap and cars weren't affordable.Your analogy is stupid
Another reason Haden Wong's analogy is stupid is because the cars replacing the horses still required the rider/driver to control them, whereas driverless cars require no control from anyone seated inside them, nor can they be controlled by their occupants.
simple, gen 1 it can drive itself if you want it to, gen 2 well a steering wheel is a optional extra, gen 3 no one considers it, each generation is about 15-20 years.
dstbij 52 okay so 60 years, 40 years shy of my prediction. That is still more than half of the average human life span. So currently we won't see full automation.
Excellent video, making points you don't see made often. And concise too. I would add that computers lack our intuitive psychological capacity to intuit the intentions of others. A car may calculate a pedestrian's distance exactly, but they don't read pedestrian's body language.. you know like hesitation to make sure the care is slowing down, or the gestures we make that tell other drivers what we're doing. And think of all the info you gather from seeing the other driver. You can't see the car execute a program. Sure you can't climb into the other driver's brain. But you can do the next best thing.
*I will echo a comment I saw on another video that also covered ethical concerns of driverless cars: We can and do already kill thousands of people every year with human-driven vehicles and we consider it normal. While sorting out these hypothetical ethical questions is important, it's also important to note that implementing ANY rational solution will result in saving thousands of lives compared to the status quo, and unnecessary delay leaves us with very real, non-hypothetical blood on our hands. Don't let the perfect be the enemy of the good.*
This is true but it's also worth noting that blame is pretty easy in the current situation. Someone fucked up in a car wreck or accident but who fucked up in a self-driving car? The "driver" who wasn't paying attention or the AI not doing it's job properly? If it's the AI who gets blamed? There were probably literally hundreds of people involved in programming such a thing and it's not going to be easy to say who screwed up when code gets that complex. Or was it the streetmaps on google (or whatever company) incorrectly placing a sign or road conditions? Was it malicious people putting bags over signs or malicious drivers manually overriding due to poor decision making? I actually agree that almost any decent implementation would probably save tens of thousands of lives even in just a few years but on some level I'd like to be able to decide that ruining someone's life isn't something people can just walk away from without consequence. If the motorcyclist situation in the video happens, I want to know where the prosecution lies. Honestly, though we won't have any of this solved until we just get around to implementing self-driving cars and wait for the law to catch up.
The self driving car is less likely to put an owner in the dangerous situation. As more vehicles enter the system I forsee a mandatory vehicle to vehicle communication system that adjusts a single vehicle in concert with those around it. If "your" car is approaching a junction, it will already know that another autonomus vehicle is approaching from another direction and possibly (?) that a human driver is approaching from a third(!), reported by an exiting "AV". Maybe, even parked vehicles will broadcast the position of human drivers as they pass?
Don't let the wolf in sheep's clothes cloud your views. It is dreadful how many people are victims of car accidents, and to hear the actual numbers is stunning. The morality in our hearts quickly jumps to support any offers that helps reduce tragedies, and don't really think about other factors we have no knowledge of. Then again do we really think this Technology will actually save lives? These driverless car have a number of accident already in their testing, and most if not all have lead to a fatality. So if anything as of now it has saved no lives, and has actually contributed in the death totals for car accidents. We must then ask ourselves how much more testing is needed? And how many more deaths will be added to the totals before they can perfected this technology? Wnd that's assuming it could even be perfected. What if it can't? Then the lives it took during its failed testing would all be in vain. Let's say your hopeful, and believe eventually it will work. Then you say the lives of a sudden few now is worth it when it saves millions later. Okay well that's your prerogative. But are you then really trying to make others believe that you actually care about people, and just want to end the suffering caused by car accidents? So what about the countless number of people that will be put out of work when drivers are no longer needed? I'm sure there will be suffering there. What about the growth in homelessness that we will reach when AI and robotnics takes over all labor jobs. The blue collar worker will be left jobless for lack of education. Where's your morals there? Plus do you really think this will be good for the Economy? What will this do to us all as a country? Will our nation be able to overcome the poverty while taking care of all of its citizens? The wolf in sheep's clothes is playing our hearts. It doesn't care that people are killed in car accidents every year. They only care about driving every competitor in transportation out, and being the only source of transportation so they can do what they want with the market. This is far more harmful than good, if any good at that. This in my opinion would end our nation as the world power, we won't be able to fund our military, we will become vulnerable and unable to defend ourselves, and will inevitably be invaded and destroyed. All because you allowed the wolf to play with your marality and good heart, and supported their actions.
@@Kobolds_in_a_trenchcoat Who do we blame when a rock breaks off a cliff and kills someone? Nobody. Shit just happens. And in the time the driverless car crash investigation, the first 5 hours, 10 people died from human-related crashes. Current standards for the good driverless cars are already better than humans. It's only a matter of time.
Gradual introduction of self driving technology is the best approach, modern cars already have things such as parking assist, blind spot detection, and sensors for when a car is too close. Eventually there may be 'smart roads' that are designed to support fully automated vehicles and manual overrides are restricted on such roads. Yes there will be bumps and legal quandaries along the way, but if the end goal is the complete eradication of road accidents and fatalities, it would definitely be worth it.
You should know that human drivers can never be banned, not until self-drive penetration at least hits "critical mass" at which the public is ready to do away with human drivers. And the transition over to self-drive cars, is far too "gradual" already. The more time that they dilly-dally around, trying to get it just right, horrible human drivers are still killing people in crashes that could have perhaps been avoidable, had we moved more quickly over to self-drive. Why can't they get more serious about self-drive, and put those computers into the airplane pilots' flight simulators, and throw a bunch of scenarios at them, and get them to handle them correctly? Put the self-drive cars onto the track and throw scenarios at them. Hose down the track in freezing weather, and see how well self-drive computers can drive on an ice skating rink.
11:14 Program it to shut off the motor and shift into 1st. If it's an electric car, maximize regenerative braking until it can safely reverse motor rotation. Or just hit both guys. Why not?
Solution to the "WHO HITS WHO?!" situation: the owner of the vehicle performs a quick 25 questionnaire about moral dilemmas. Stop bringing up this "this will happen in this fake scenario and the programming masterminds have to put their code in which will elect their own choices" and whatnot because it's really not that hard. It's literally as simple as "SCHOOL CHILDREN ARE CROSSING THE ROAD AND A SELF DRIVING CAR CANNOT STOP OR TURN AND YOUR SELF DRIVING CAR CAN SAVE THEM WHAT DO?! A) HAVE YOUR VEHICLE DRIVE IN FRONT OF THE ROGUE SELf_DRIVING CAR AND STOP THE CAR OR B) DO NOTHING"
Pedestrian bridges typically have some sort of protection on them to prevent people from falling over. In Vegas they have glass panels and one nearby me has a metal cage around it. No ones going to fall in those.
It is easy to see if an object is moving towards you, away from you or is doing neither. It is called the Doppler effect. We use it all the time in when measuring the speed of a vehicle. Bats use it hear if something is moving trough echolocation.
One big problem with radar is if you have 40 automobiles in a segment of Street how many different frequencies of radar are possible? At any one time you could have a few cars canceling each other's radar out. It's like how when I was helping to move my sister from one apartment to a different one we had a big green station wagon and I would lock the car and bring something out unlock the car and put it in. At one point I went out to the car to get something that was next to the driver's seat so I came around that side unlocked it, went to reach in, and realized it was not our car. Two spaces up there was an identical station wagon to ours parked there. Neither station wagon was even remotely new. One thing that was different is that we had come from a neighboring state to where my sister was to move her. Different cars, identical keys...
@@mypenisisunbelievablysmall5258 the biker in black suit and with cat ears on the helmet is a character from anime Durarara. I totally recommend watching the first season!
I have seen a lot of good videos on this channel, but this one is good enough to get me to subscribe. As a techie for more than 50 years and deeply involved in electric distribution automation from 2010 until recently, I am astounded at the public perception of it being simple. If there were algorithms for everything a driver has to do we would already teach drivers that way and would not allow anybody behind the wheel until they could demonstrate familiarity with each one. Instead, we are constantly having to figure things out, which a computer simply can't do.
There are some who support roundabouts and some who don't. Edmonton, Alberta, Canada has a lot of them, but where traffic volume has increased, they're being replaced by conventional signal-controlled intersections.
I hate this US argument... I lived on cape cod most of my life and we had roundabouts everywhere. Not sure why people say they aren't here in America .-.
"You'll never overcome" = we haven't figured out how to do so yet. It's been true throughout history so far. The more we can get the entire human population connected to all the ideas and theories of humanity and contributing their own talents and theories to challenge what we assume is possible, the more likely things we "can't overcome" become footnotes in history of how backwards our own generation was in its thinking.
Just want to correct one thing - radar can detect object speed, by using frequency modulation or simply measuring frequency shift of incoming signal due to Doppler effect.
It can only detect speed parallel to the direction vector of the photons. If an object is travelling perpendicularly to said vector, there won’t be a doppler shift caused by said object (although there can still be a doppler shift caused by movement of the radar, as would be the case with a self‑driving car). *TL;DR:* The measured velocity will be the actual velocity multiplied by the cosine of the angle between the radar beam and the object’s velocity vector minus your own velocity (technically addition, but one of the velocities will be negative to the other when travelling in opposite directions, which is probably the case in this example). *TL;DR;DR:* vₒ = vₐ × cos(θ) - v₀ Also, this is simplifying things quite a bit.
The hypothetical situation I love is the kid running out into the street, and if you swerve to avoid the kid, you go over a cliff. How is this kid allowed to play with a ball near a busy street that's on the edge of a cliff?
Hey man. True fan here. Love your videos. But need to sort of hip you to some possibilities that people don't consider: 1. Early iterations of self-driving cars have to work basically like people do. Future iterations may not. 2. For instance, self-driving cars can be networked and be required to have transponder capability. (Much like airplanes). Your car won't have to SEE other cars to avoid hitting them. It will just know where all cars are, nearby, how fast they're moving and where. 3. Visual navigation is fine and self-driving cars can do it. In the future, though, roads can be constructed with inexpensive optical and magnetic and other types of sensors that let cars navigate even better. For instance, your car could be reading mile marker bar codes every 1/10th of a mile. Combined with their data of information (think of wayze) they can know that a traffic slowdown is going to occur in 1/4 of a mile or whatnot. Your car will be aware of the environment to a degree that's impossible for a human. 4. Visual recognition of NON CAR objects? Man, we're not even in the 1st inning of a 9 inning game, there. As I tell my students "once a computer can do something, even badly, the game is over, because the computer will get relentlessly better, 24x7, forever." Don't judge future capabiltiies by today's technology. Our technology will improve. Insanely. They will exceed human capability. That's not in some distant future. That's in the next 5 years. 5. Keep in mind that we kill, what, 40,000 people a year on our highways? Something like that. And how many of those are due to drunk driving? Inattentive driving? Sleep-deprived driving? Self driving cars won't suffer from any of that. Once this technology is fully fleshed out, we could literally reduce our fatalities to something like a few hundred people a year. Maybe a few dozen. That's basically like curing a form of cancer. So, yeah, don't think of cars in terms of how well they imitate people. A wrench doesn't work because it's exactly like a person's fingers. And technology works because it does things humans can't.
Self driverless cars aren't a special messiah that will solve the problems of roads. Like already mentioned, you can't have everybody own a driverless car, specially the low-income, antique, and manual drivers. The annual death count will merely be reduced by a bit, even if we have superhuman cars with ultra fast reaction times, it isn't enough to sway a 1 ton metal machine at 60 km/h to avoid hitting a child, while preventing the cars behind from cashing where ALL of them are self driving. And unless you can get cars to weigh like 100kg, roads with optical sensors and constant QR codes, while being maintained for an immense amount of cars going by, won't be feasible economically in the tiniest bit. Just spend money on public transport than trying to figure out how to find an iron needle in a haystack without a magnet
I will never sit inside a vehicle i cannot get to the controls of in case of emergency, i need a compromise where i can still drive but when the car detects danger or rogue traffic then it takes over, that i think is the future as it does not take away peoples ability to drive there own car but also has the safety of the faster computer mind.
8:30 yeh that's not a problem here in the UK. The only hexagonal sign on the road is a stop sign. Even when the stop sign is completely obscured by snow you can still tell just by the shape.
I think the main thing that people want from self driving is not safety or speed but the fact that driving requires literally zero effort. Self driving cars will never be significantly faster than regular cars. We will always have intersections. Traffic will get worse. But none of it matters, because you do not have to deal with it.
Governments will not need to intervene to make driverless cars a thing; insurance companies will as soon as these are demonstrably shown to be safer than human drivers. The driverless discount on auto insurance will help drive the driverless market. As far as the strawman scenario "who should the car hit?" the car will do it's best to avoid *any* collision and even if a collision is imminent will take steps to protect you from more serious injury. Cars are *already* being marketed with these crash protection and avoidance features where the car operates semi-autonomously. www.iihs.org/iihs/ratings/crash-avoidance-features
If it's fully autonomous, liability actually seems fairly simple. If it was a manufacturer or electronics error, it's the manufacturers fault. If it was a failure caused by improper upkeep, the driver's fault. And the fault of the person hit if they endangered themselves.
Daza Mistwalker yes but car manufacturers will counter this heavily, especially those who produce cars that are used for moding, muscle cars, sport cars, trucks, and off roads
I only see from one eye, and although there's truth just on how I get depth cues as explained in this video, one thing is often forgot is that no one see in static images, one moves around a get to model a 3d world even if only has access to 2d input, perhaps not as easy as once with two eyes but the brain is kept busy trying to infer from the data it collects
I would just like to remind ppl of the recent update that an appliance company sent out to their high-end microwaves that bricked every microwave that received that update. The update was sent out on accident and they have to send a technician to everyone's house individually in order to do a reset on the microwaves to get them to work again. It's not that rare of an occurrence either. The Internet of Things (including cars) is just not a good idea.
Just test drove a model Y yesterday. It drove for me the whole time. A lot can happen in 6 years. They’ll all be driving themselves pretty soon. But these philosophical issues are still pressing.
Yes, and computers have had no problem at all with depth perception for over two decades now. Its very easy to achieve using infrared sensor technology. This guy is a moron
He was talking about inferring depth from a 2D image, which humans are surprisingly good at (he mentions some of the clues and stipulates the computers can make use of some of them, so he doesn't say it can't do that at all, only that it's a hard problem which will at minimum take time to solve). He also notes that computers can use echos instead (tho he should have made more explicit that LIDAR is the most relevant one, using Time of Flight of light, rather than audio or radar).
Actually, depth perception from binocular vision only works for short distances -- threading needles, not figuring out which of two objects across the street is closer.
Deep blue had a hard time wining. It lost a couple of times and was specifically tweaked for one special opponent. Even if afterwards it won against said opponent, it is assumed that it would be in disadvantage against other players because it was very much tweaked.
I know this is late, but that is a very myopic view. What individual, when competing at the highest level in chess, wouldn't observe the common strategies of their opponent? You leave out that Deep Blue's opponents also spent a significant amount of time studying tactics that exploit the weaknesses of computers.
All the "moral issues" of the driverless cars are moral issues of the driving itself. What's the difference between a car making decision or a human making decision? Decision to not act and cower your eyes is still a decision, even if cowardly or panicky one. But we have this extremely weird cognitive bias that it is way more acceptable for human to chose the wrong thing, than a machine. For driverless cars to start replacing humans, they don't need to be perfect in every situation. They just need to get in car accidents less often than meatbags, statistically.
you're out of your mind if you think im gonna let a machine decide if i live or die. in every one of those examples he brought up its an easy choice for me but maybe not for a computer. i will always choose self-preservation over trying to save others. I'm not gonna get a car that's gonna try to be a hero with me inside as a sacrifice. two bikers coming? welp one of them is gonna die cause it sure as hell wont be me. kids crossing the road and my brakes are out? well sorry parents but i got a future too. I would never put my life in the hands of 1s and 0s.
in the hypothetical situation, the brakes are out and there's no stopping the car. Im not gonna throw my damn life. Thats called self preservation. Im not gonna let a damn car decide to sacrafice me for someonelse's life.
A week ago or so there was an accident in Poland, when one famous journalist drove his car over 300 kilometers, on a freeway... while being completely, utterly wasted. Turned out the reason he managed to drive so far, being that drunk, without any trouble, was that his car did most of the driving for him. It adjusted the right speed and remained stable in the lane. However, he reachead a point where there were some road works, and the signs and road lines were replaced with a temporary ones, organizing the traffic for the time of the construction. The car apparently couldn't tell the difference and got confused which lines to follow, and started to drive through traffic cones. As a result, the guy got caught.
This is a great summation of the issues involved with autonomous cars. The key to self-driving cars would be achieving the right balance between autonomy and centralized control. Given a strong enough up/down link to a central computer, all the cars within the zone of control could be controlled to such a degree that they will avoid hitting each other because they're all being given instructions that by design and definition do not conflict. Balance that with the ability of cars to sense and respond to their surroundings when outside central control, along with exactly the right hand-off process, and you have it licked. But as you noted, that ignores real-world events that can't be controlled by the computer. In short, well thought-out and well presented.
Another issue is going to be getting lower income people to buy into this technology. People in this class already choose older used vehicles just to have a way back and forth to work. Getting them to invest in a new driver-less car is going to be difficult. Even when driver-less cars have been around long enough for used ones to be on the market, they will be prone to having software that is either out of date, going out of date or obsolete and unsupported, including the hardware needed to runs these processes. Lower income people have a tendency to hold on to a car till the wheels fall off before buying a new one or fix the current one themselves because the budget just isn't there to support buying the newest, latest model every time one comes out. There has to be a huge change to work economy to make full adoption of this technology feasible because the cost to go into this kind of structure and maintain it is going to be massive. Who is going to foot that bill?
Yes, I was thinking about that too. Not to mention repair costs. I think the most likely approach would be, as others have stated, a lease model where some company owns the car and provides maintenance and insurance as part of the monthly fee; it would also get rights to the collected data of course. If some particular model has problems, they would be removed from the road (see clause 45 in your contract). As for realizing the advantages of all automatic (all self driving)... Dedicated lanes on some highways could be one of the early approaches, likely first replacing HOV and toll lanes which are already segregated to some degree. IF all automated cars can go zipping by with high speed, minimal separation between cars, and high safety, that would be a massive advertisement to the other lanes. As more cars to automatic, more lanes would become dedicated (massive highway infrastructure changes, but over time). I really can't see city or rural roads going all-automatic for a very long time. The automated intersection seems far off.
That's my plan for maintaining my vintage accord and F150. Get a motor for both, rebuild them and keep them on standby just in case the current one goes. Drop the rebuilt unit in and rebuild the old one if possible or replace it. Keep that cycle going until either they get wrecked or I can't drive anymore.
The ability to summon a car from an app on a smartphone will eventually lead to people sharing a car with family and friends, eventually this will lead to car sharing organizations which you simply join and summon a vehicle from the organization's pool of cars. This would will allow people to avoid car ownership and the associated costs completely hence enabling the less affluent. If we really want to press the technology to support less affluent people, we could implement public transit variations such as PRT which has actually been possible on guided through fares for at least a generation. However, the sharing of vehicles will likely lead to fewer vehicles per family so what will the car manufacturers do to stem the loss of market? I suspect that they will emphasize and market various stages of semi-autonomous vehicles to both reduce the market for fully autonomous vehicles and deflect the public from wanting full autonomous vehicles. I think we'll already seeing this in effect with the SAE categories of autonomous vehicles and the major manufacturers releasing assisted driving features while only companies without a base in automobile manufacturing such as Google and Waymo focusing on fully autonomous vehicles.
heres a tip. if your breaks ever go out in your car while youre in rout to somewhere, the only thing you have to do is slam the gear in reverse or park. there was a terrible accident that happened a few years ago where a family was traveling and the gas got stuck somehow and the car was accelerating. the breaks werent working. they were coming up fast on a 4 way stop. everyone in the car that day, perished. they tried to call 911 and were actually talking to the dispatcher for about a minute before disaster struck. it was such a tragic situation because nobody knew how to help them. they were so afraid. there is no way i would allow a computer to drive my car when stuff like this is still happening with cars. we rely on computers far too often
Hey, wouldn't you agree that trying to code in preference in complex moral dilema would cause more death through malfunction (due to increased complexity) than just coding the car to avoid obstacle and follow the law? By the way, you're one of my favorite youtuber, your videos are thoughtful and interesting, on various topics, I love the calm and rational approach you seem to have, and you should have way more suscriber. Maybe it's because this is still a small channel, but the comment section also seems friendlier than elsewhere... You guys are also awesome.
The problem with saying "just leave it simple and have it follow the law" is in the end, legality. When YOU freeze up and end up hitting someone, most people - including the injured, the family of those who died, or a jury - are able to put themselves in your shoes and at least see how they could have done the same thing. If a computer-driven car does it... someone has to be at fault. Which means likely, the car company. People can forgive other people. People rarely forgive corporations. The first kid who gets run over by a driverless car (which let's be honest, it will happen eventually, there's no way it won't) the lawsuits and hysteria will be so real it's doubtful many other car companies or people will be pro-driverless. That's why this is so complex and difficult and can't simply be solved by "as long as it's only slightly better than people drivers, it's better for everyone." Also, yeah, my comments tend to be pretty nice. I'm not sure if it's because I present my content from a middle-perspective or because I'm just lucky... but I hope I continue to grow with this crowd. :)
Why cant cars just do what most people would do now while driving? If I saw a car speeding down about to strike some kids i probably wouldn't pull forward. If there were two motorcyclists and I couldn't stop, i'd hit the idiot without a helmet/protective gear that's driving on the wrong side of the road basically asking for a darwin award. They guy with the key going to open his car, i'd keep driving and so should the car. If the guy was about to walk out onto the road then the car should stop if there's enough time, but if there's not enough time then there certainly wouldn't have been enough time for a person to stop so the dude who walks out on the road without looking both ways gets hit obviously. The only problem i can forsee is that if everyone starts using self driving cars and the definition of "do what most people would do now while driving" becomes meaningless.
Pretty much my attitude. Human drivers suck, just need to suck as much, or be slightly better and I'll be happy. I don't imagine a driverless car will sit 2-3m from my rear bumper on a dangerous windy road beeping at me for not going above speed limit! Human driver, all the fucking time! Maybe that's what is needed for driverless acceptance - fully automated vehicles that sit on your arse and honk at you if you're not going 10-20kmh over the limit. Even in the situations you prob shouldn't.
public transport is also way easier to organize and can transport people much more efficiently. Either by rail or drivers, even all of your human drivers can be extra eductaed for the job
As a Truck diver this was very interesting to watch. One large gap that I rarely hear people discus is changing road conditions. It is not always easy to detect slippery roads before you have an issue. Vision and radar are not going to be able to tell a car what a safe speed to drive is when conditions get bad.
difficult for a person to identify* As per the beginning of the video, things people did hard, computers can find easy, and vice versa. the car is also going to have very precise measurements on how it's driving affects how it moves and could probably be testing constantly to check how slippery the road is now, which a person would get bored/distracted of, or not have enough precision to notice small differences
I understand that the cars will probably be constantly measuring acceleration forces and compare them to the inputs it is giving. I have a hard time beliving that whatever test it runs it is going to be able to tell a difference until traction is lost. A classic scenario is you are driving along and the temperature is in the mid 30s and there is some lite rain. the temperature slowly drops below freezing how is it identified that the road surface is icy and needs to slow down to a safe speed. What is the safe speed if other cars on the road are unaware of how slippery the road is and continue driving at the speed limit or even above it? I am sure the cars will be programmed to drive under a variety of weather conditions but how quickly does the car decide when it has changed from one category of raining to freezing rain or snowing to sleet.Another one is it starts raining on a hot day. when it first starts raining the road can be slipperier because of a slight film of grease and oil from the surface of the blacktop that gradually washed away.
sorry to tell this, but detecting the road-tire grip is quite easy and (some) cars are doing this already in realtime. these constant, millisecond scale measurements and instantaneous torque modifications available in an electric powertrain are reasons for tesla to be able to beat lambos in 1/4 mile while driving with much narrower tires.
As someone from Canada, I'm happy with just the tesla system. Driving in the city is whatever, but if I'm going from Vancouver to Prince George or Calgary to ft. McMurray, staying properly attentive while essentially doing nothing is really hard. I would like it to reach the point where I could sleep or eat while it's doing this, would hugely increase mobility, routes like Vancouver to Winnipeg are routes that either need two drivers on shift (and that is exhausting and a lot) or a night somewhere on the way, (and that's still pushing the trip a lot). Being able to sleep in the car and eat would make everyone in the country a lot closer.
Yes, riding is so much easier than driving. And I have long ago noticed that I can notice and appreciate so much more scenery, riding with somebody, than having to keep my eyes glued to the boring road. I think it is crazy, the very idea of driving, being always but a few inattentive seconds away from crashing. Humans should not have to go through that. And why do I feel like my brain is running a complicated computer program when I am driving, having to constantly scan for hazards and to remain constantly alert for up to many hours? The human brain was not designed to run computer programs, but rather to write them. Computers should run computer programs. Computers should do our driving for us. Perhaps soon, someday, when people finally get to ride in a competent self-drive car, their attitudes will change. It may feel that it is on some fixed invisible track on the road, as the car should not need to wander around in the lane, except to avoid obstacles such as potholes or tree limbs or whatever. People may feel even more secure with the computer driving? Somebody some years ago, typed to me a reply that the government will program your self-drive car to kill you. I replied back, "You have been riding on self-drive elevators for decades." Of course the elevator only has to make simple decisions, rides on a track, and does not know who you are. Still, I think that self-drive will prove to be far safer than human drivers, and then no longer will be have to worry about becoming drowsy, and we humans will be able to go to more places and our useful traveling range will become farther. How many activities have I cancelled, simply because I felt a little too tired to commit to driving that far? How wonderful would it be to be able to curl up on the back seat, sleep soundly for hours, to have our car finally wake us up, "You will be arriving at your destination in 5 minutes. Wake up sleepy head." Imagine all the people will will actually be able to go visit family a bit more often, without having to deal with that grueling long drive. And your car could also check the traffic and weather conditions, all along the route, and plan accordingly, something that human drivers very often do not bother to do. Maybe reroute to avoid steep roads, if there could be snow.
@@briancrawford8751 For a short 30-minute drive, getting on an airplane would take even longer, well except in Alaska in which airplanes are about as common as cars. Someday, cars will fly and be fully-autonomous self-fly.
@@yosefmacgruber1920 Uh huh. Yeah, okay. You can't stay awake and attentive for a 30 minute drive, so you need a self-driving car? That's messed up. I have to disagree that self-driving cars will succeed in the Eastern US. All it takes is the stroke of a Governor's pen on a bill passed by a state legislature to outlaw them, and given how powerful the AFL-CIO or "Teamsters" are, I predict that they will be outlawed before they hit the road in many states. In my state of West Virginia, there's no way a self-driving truck of any kind could make a journey from one town to another without some kind of catastrophe occurring. I couldn't see a self-driving car doing it except in well planned neighborhoods while all the kids are at school. We have roads that run between a rock wall and a cliff with no way to pull over, single lane roads on which people travel either way, and the occasional rockfall.
@@briancrawford8751 I did not say me necessarily. But do you really think that I want to work a long shift, and then have to drive for yet another hour before I can rest or sit down and relax? Maybe that is why so many of us are loath to look for a job in another city other than where we live. Can you not see how this horrible manual-drive concept is not only so yester-century but is seriously limiting our options and making the roads less safe? The problem of tired driving is so bad, that even the lousy government has put in a lot of rumble strips in the shoulders of the freeway, to keep drivers from dozing off while driving and crashing. Think about it, feeling that rumble would seem to say, "Stop driving and find a place to rest" and not just "Try to stay awake." It should say, "Just push the auto-drive button" already, well if there was one that is. Ironically, one of the very first episodes of _Knight Rider_ depicted that same scenario, of Michael Knight falling asleep while driving. At least K.I.T.T. had the fully autonomous self-driving (AI-based) feature, and was able to assume control, and then helped him to come up with an excuse to tell the cop who pulled him over. Michael Knight had taken off in a hurry, without allowing time for the many amazing technology-advanced features of his "company car" to be explained to him, I think for a dramatic excuse to introduce the audience to what that 1980s TV series was going to be about. Self-driving would make those treacherous drives a lot safer. Computers can actually rehearse what they are going to do, and do mathematical calculations of what speeds are safe to take a curve at. And you seem to forget that self-driving cars may perhaps learn from the experiences of other self-driving cars. Humans can not be trained anything like that, well unless they believe every textbook claim (that they didn't even bother to read) with no proof. Textbook says, "You can not safely drive while drunk or stoned." "Ha! Ha! I am the exception. I am special." Yeah, especially stupid. Textbook says, "You should follow the 2 or 3-second safe following distance rule." "I have been driving for decades, and I am an over-confident good driver, as you can see by my torn-up front bumper, so I should be allowed to draft right on somebody's bumper, like some sort of hemorrhoid, because I haven't been in all that many accidents." Self-drive is the future. What stupid public is going to allow for the future to be outlawed? Anti-progress DemocRATS? I think it will come to be seen as a safety feature, as self-drive improves and proves itself. And what union is really going to be against paying truck "drivers" to sleep on the job? If they are in the back sleeping while the truck drives itself, surely they have to be paid for that? We pay drivers to drive while they have the cruise control enabled, right? It is just a driving aid or tool. We pay drivers to drive, even while they run the heater or air conditioner. Having such comforts are just part of improving the working conditions, they are not financial compensation for doing the job. The human driver is still the driver. Or we may have to revise a few terms. Instead of "truck drivers" how about "truck riders"? Hey, I would love to have one of those modern self-driving trucks if I get to keep my job, and don't have to so much "go to work" until reaching the destination or having to refuel or whatever. You do know that at some point, self-driving vehicles will be talking to one anther, right? So in your one-lane bridges or narrow passages, they can time their arrival and alternate their passage so that they can share the 1 lane. Have you seen the YT video computer animation of how self-drive cars will just sail through the intersection without stopping, because self-drive cars will request a reservation to go through the intersection, and then time their passage so that they do not collide? Wouldn't it be nice to largely eliminate stop lights, well except for pedestrian crossings? Also, a self-driving vehicle can navigate through difficult driving conditions, without feeling stress, because stress is a human emotion. Imagine having less stress, as your very-competent self-driving vehicle gets you home in a difficult blizzard, without getting stuck, when you might have found a place to pull over and give up until the weather improves. It would know of which slick spots to be extra-careful about, as the other self-drive vehicles would know all about the current conditions and share that info. Imagine having to take fewer pit-stops and rests on a long journey, because computers just do not get tired after 10 grueling hours of driving. They remain just as alert as ever. A computer is not a biological organism that needs a rest cycle. Because nobody really trusts the self-drive technology, it thus improves the software engineers' dedication to make sure that it is as safe as possible. It obviously is not a fair competition between human and computer drivers. Computers are expected to do the job perfectly, and why not, if it is possible? Computers would do exhaustive scans of their surroundings, hopefully seeing early a child or deer approaching the road, not being subject to "freeway tunnel vision", nor being already tired after a long work shift. Also, in Driver's Ed, they teach you that a ball rolling out into the road could soon be followed by a child retrieving the ball. Self-drive cars could relay that info to the several self-drive cars behind them as well, to slow down, due to a temporary hazard. When humans wreck a car, no big deal, right? That's human drivers for you. Why we have to have liability insurance supposedly. When a computer wrecks a car, there is so much more evidence. Log files will be picked apart. The dash cam videos will be studied. Much like when airplanes wreck, they will find out why, and release a software update to fix the problem. But that is increasingly becoming a problem. Today's high-tech devices are only as good as their crappy software updates, and just how many free software updates are we entitled to or do we expect to get? Will your self-drive car be "outdated" and get no more software updates after just 8 years? Smart-TVs seem to only get updates for what? Maybe 2 or 3 years? So I am supposed to go buy a new one, in just 3 years? Think again. I usually use things, until they just do not work anymore and I can't somehow patch it together anymore. So will the greedy insurance companies, complain that self-drive cars are just way too safe, and find some way to sabotage them, so as to jack up the price of their largely-a-scam socialism share-the-risks insurance?
I actually put this video on to keep me company while I was working but it was so interesting that I stopped working and focused on the video. Great work! Oh you also introduced me to a new song. Thanks!
Driverless cars are more hype right now… sure they do well in certain situations but they don’t in others. There are many edge cases that will hiccup any plan for Level 5 autonomy where we can get rid of steering wheels and pedals. Just the other day a “geofenced” taxi drove behind a car stopped where the owner forgot to turn off the lights and the taxi stayed there overnight.. it interpreted the car as being stopped so it just waited.
Oh come on...Until 10minutes I was: "ok sounds reasonable, but solvable with enough time and computers"...But then the dilemma situations came again...This is just scare mongering imo. If the car is designed so badly, that it drives so reckless that it cant stop and not hit one of the two bikers then it shouldnt be on the road in the first place. Also simple physics, the best thing to do in those situation is to emergency brake, it gives the bikers more time and reduces the total energy of the crash, if there is no sensible way to just avoid them.And the school class example is just soooooo unlikely, that I dont even want to start talking about it but ok. Even a selfdriving car will have an E-brake, as has every other vehicle on this planet, that drives on a road or rail. This will not happen! Its just sooooo unlikely that having less reckless or old drivers causing crashes outweighs these freak accidents by a mile. And in the end it just boils down to lawmakers. If they make the car manufacturers responsible in those situations, then they should also be responsible for all the deaths, that occur if you stop self driving cars...Because it then is Their decision to let people die because of too old or stupid drivers
Yet, we now have at least a few examples of humans that have died at the hands of automated cars. We had a recent situation in Arizona where someone died due to an automated car that WAS being attended by a driver. We also have a report of a Tesla that had a sensor failure causing the death of the occupant of the car. This doesn't include the large number of potential non-life threatening problems that automated cars can still be a part of, and the fact that they are not immune to new problems we don't even know yet. Just ask someone who relies heavily on computers to conduct business when the business network was down -- and that's dealing with a fairly static set of variables (i.e. computers sitting on a desk plugged into a wall or connected to wireless). Personally, I think the problem that needs to be solved more than anything is how to we make existing cars safer NOW. How do we reduce distractions and educate all drivers of the horrible dangers of distracted driving? Even if we get automated cars on the roads within the next 10 years, we still have a large number of needless injuries and deaths that could be prevented. Since one of the BIGGEST contributors to accidents is distracted driving, what are we doing NOW to reduce distractions in motor vehicles? I believe these are far easier and cost effective technologies to implement. Things like lane departure, crash detection, cross-traffic warnings are all great technologies that can help reduce driver error. By all means, we should continue the research for automated cars and continue to work on making all cars safer with these technologies. However, I just don't see driverless cars as the Utopian solution to this problem.
westdccs the other side of this is that that car, and all other cars can learn from the experience and better avoid it later. I'm not saying it is perfect. But I think (if I am thinking correctly this was the woman on bike at night scenario) that a human wouldn't have had better odds at avoiding that in the dark. It is possible. But that late and that middle of no where, no one is really paying attention.
You seem to have forgotten that lawyers exist. You can bet good money that the first time someone is injured in even the most unlikely and freakish scenario, a lawyer will be in contact with the injured party offering to earn a fortune while clogging up the courts, which in the US means that Joe Numnuts and his barely achieved high school diploma will be sitting on a jury trying to decide what millions of lines of code should really say. The only reason we have not seen this already is that Uber settled out of court.
Like for the Celty reference btw. Of all the biker chicks in fiction, i name her Best Grill. I'm blind in my right eye. Plasticity of the brain and the growth provided by life allows me to have excellent monocular depth perception. I have no problem driving, shooting hoops, riding my motorcycle at break neck speeds around twisties, or catching something someone tosses me (unless they have a bad arm). It should be (and one day will be) perfectly feasible to teach an AI to do what i do for fun with only one camera. These self driving cars better not randomly veer into my lane when i'm passing... Hard to sue a car...
I worked on a self driving car project last year and the ethical dilemmas everyone is freaking out about are a non issue. The car is just programmed to plan to stop. That's it. No swerving, no ethical choices. Just apply brakes and stop. Let the ABS and ESP handle any traction loss from the brakes.
Good to see people who know commenting on this kind of video. I am curious though, I get the impression that you guys work more on the technical side of things and in the example in the video where they are driving side by side I would be surprised if this was an issue. Having said that I still think the idea is valid - what happens when something unforeseen means that some kind of accident is inevitable? From a technical standpoint "slam on the anchors" and minimise the damage is fine, my question would be, what happens when the lawyers get involved? You can bet that there will be at least one who throws a spanner in the works. Are there legal representatives on development teams to look out for these issues? I would guess yes, but the world often doesn't work the way we think it does so I'm genuinely curious to find out.
Fernando, I was actually in an accident many years ago where my mental heuristic operated exactly like that. The end result: I did wreak into the parked cars that the a-hole wanted me to, but I took him out with me by literally driving down the shrinking opening I had until I hit both!
> Sometimes stopping isn't the best answer. Sometimes acceleration is the best answer. The car doesn't have to pick a best answer, it has to pick a valid answer. And breaking is a valid answer in case of some unforseen weird thing happening.
"A problem that will never be solved though..." Is a really bold statement. I can imagine this has been said about all the technological advances ever made.
"If humans were meant to fly, God would have given us wings." Well that one was wrong also. In fact, God's creation of the birds, gave us some clues that heavier-than-air flight was actually possible, and how we might go about doing it.
If I had to sit in a car that doesn't even have my input on where it goes I would just take a bus. I would see a lot of people starting to ride motorcycles or bicycles once self driving cars are the norm. People will always want to have input on a vehicle. If I had to sit in a car that I paid for that I can't even control I would hate my life.
@@prospero4183 well people still want individualized transport. 2 wheeled vehicles are stored and parked very efficiently so I bet they will become popular again once roads become safer.
I’ll make my own self driving car so I can turn off A.I mode and do things myself. But then I’d be a problem. So what’s the point of a self driving car in the first place?!
@@streamtrollmike5348 it's actually mostly that humans will make dumb decisions and good self driving will just be safer. By driving, you're risking other people's lives
@@streamtrollmike5348 That's a key point -- can self driving cars follow not only the written rules, but the unwritten rules of the road. For example, will they take 10 minutes to pass a semi, because they're both going almost exactly the speed limit? Or will it speed up to get around, so that traffic can flow in the left lane?
I once drove on a dirt road with the lane departure correction turened on in the brand new work car for like 100m. It randomly decided there was a line and started steering me across the road whilst the traction control was taking drive off my wheels for every corrugation i hit. It was like driving on glass, whilst fighting some pirate for the wheel. Pulled over as soon as I could and turned all that nonsense off. NEVER AGAiN.
Would you want a car that protected just you - or as many lives as possible? Who do you think the car should hit? What about the legal issues like sleeping and drinking while riding?
Another legal issue:
Say you have a self-driving car and it runs over and kills a pedestrian due solely to imperfection in the software. Who is at fault?
It could be the person in the driver's seat, but they have no control over the accident happening, attentive or not.
It could be the manufacturers, but blaming them for deaths may penalize progress that would ultimately save more lives than it costs.
It could be nobody, but the driver and manufacturer offered a consent to the inherent risk that the victim didn't.
Furthermore, if a car had to choose between killing two drivers and it could access information on who they are...
should it consider factors like age, criminal record, impact on society, number of loved ones, and so on...
or should it make no effort to compare the value of two human lives and decide by the software equivalent of a coin flip?
The car should act like a moral human being would. Do its best to save both. Morals can be taught. Legally, manual cars will have to be phased out over about 20-25 years and it will have to be illegal to drive one, unless on special courses built for leisure driving.
I had considered adding things like age, criminality, societal impact, occupation, and even race to the video. There are many versions of the trolley problem that include these things. But I figured I wouldn't muddy the water too much.
"The car should act like a moral human being would"
How a moral human would act is debatable.
I'd want the car to save as many lives as possible. I think. Otherwise you'd survive the crash but you would have to live with knowing that you survived because a computer chose your life over another person. Then again during the situation my brain/instincts would obviously do what ever it could to save my life. Hopefully Jesus would just take the wheel.
Just had to spend the last half hour watching dogs driving cars. Thanks Jackass. (No seriously, good stuff as always.)
exurb1a shouldn't you be busy questioning our existence?
What brings you here, mighty turtle of existencial crises?
One huge problems with a dog driving a car, I imagine, is that a dog thinks it is just a game. Once the dog grows bored of it, does the dog realize that once the dog no longer steers the car, that the car will hit the wall or something, and that that will not be fun anymore?
Make a video with this guy. I'm tired of pseudo-science channels getting most views on UA-cam.
Does voicing a ferret count? ua-cam.com/video/fDoAs0qN7lU/v-deo.html
“So easy a dog could do it”
*cries in visually impaired*
Your seeing eye dog can do it
It seems like an "easy" way to alleviate the intersection issue is to replace them with roundabouts and other road designs that don't have the vehicles turning across oncoming traffic.
Yes. Also I'd like to see all self driving cars default to NEVER hitting a pedestrian. Even if it slows traffic down. We should prioritize walking. It causes minimal environmental damage, is much less expensive to provide as an infrastructure, and is much healthier, which also reduces societal costs.
@@brysonwest93 First thing to do in regards to climate change? Make cruise's illegal under international law. Those things pump out the equivalent pollution of 5mil cars.
Well that relies on the human drivers not being dumbfucks.
That song at the end was like a train wreck... I couldn't look away even though I really wanted to
Great thanks again
Stuck in my head now :(
I loved it, best thing he has produced by far :)
You're right, that's a chair.
But it's also the most happy chair I've ever seen.
Just a small point here. I'm actually a mathematical programmer and work primarily in computer vision. You actually did very well in this video,. You hit all the main points with out dumbing them down.
Although it is still research level, we DO have a full mathematical theory for object recognition which works well. The problem is, computing these "shape spaces" as they are called is currently too slow for real time applications, but we're working on that. In fact, most "problems" with self driving cars are solved theoretically. It is just not cost effective to have a super computer in every car.... yet. :)
As always great video!
Again, as always it's a little scary when someone starts off a comment with their credentials of how they're demonstrably more of an expert than I am. Glad to know I got most of it right - the last time I learned anything vision related was only a year or two ago so... I hope nothing I said was too out of date.
I hadn't seen you comment in a while I was worried I scared you off somehow, glad you're still around :)
Scared off? Not a chance! I'm a huge fan. Just been busy with exams lately. I'm glad to see your videos keep getting better.
A Tx2 by Nvidia cost 500$ and can do trillions of calculations per second. AI doesn’t rely on calculating shape spaces. It can learn what objects are by just seeing enough examples of the object
Jacob Richards Deep learning is not a mathematical theory of shapes. It is an engineering technique without any formal mathematical theory (i.e., actual theorem proof style justification).
I actually work daily with the TX2 and trust me, there is a reason self driving cars do NOT run off the TX2. Yes nvidia has papers the run a car off the TX2, these are toy examples and once again there is a reason no one uses a single TX2 for full autonomy.
I was responding directly responding to KnowingBetter's point regarding object recognition. You are right, no one uses shape spaces in self driving cars, because the inference time for some deep neural net is much much faster than optimizing over a shape space.
Cloud computing is exactly what I think programmers and corporations would resist and screw everything up because stand-alone robotics is much more expensive but much more practical for personal robotics like self driving cars.
11:47 "There is no scenario where a computer couldn't make a decision." As an avid video game player who has seen countless AI characters stand there and stare as I'm getting my ass kicked, I have to dispute this.
It’s code decided it couldn’t be god for that moment.
@@dontgankmebruv7273 well said, Though it's very interesting to think about how an ai would react to actual phsyical limits as opposed to literally what you said which is again ...very true
Couldn't and won't are two different things... just because the AI _can_ make the decision to help you doesn't mean it _will._
@@binaryglitch64 Yeah. This one goes out to you two, Donald Duck and Aerith.
That Durarara reference was 👌
So true OP. Go Celty Go
Just great
Checkers is only weakly solved, meaning that a strategy exists to achieve the perfect-play result of the game (see Schaeffer et al, "Checkers is Solved"), and this was not do by memorization of every game state as you say. I think you are underestimating the size of 5*10^20. Squeezing by with just one byte per game state, that would still require 500 exabytes, to put that in perspective Randall Monroe estimated that all of the data in Google's datacenters only numbers 15 exabytes.
Similarly, chess is not solved *at all*. Even if modern AIs are superior to the human grandmasters, that does not mean the game is "solved" in a mathematical sense. No one knows if white can force a win, or black, or if perfect play will result in a tie.
Thank god someone else pointed this out. I basically gave up on the video at this point because it's just a really bad look when an "informational" video comes out of the gate giving you background info that's completely factually wrong. I mean I'm not a tenured professor of game theory or anything but I feel like that's such a crazy thing to think that chess had been "solved" a few years ago. The maximum number of potential chess games isn't even a value that's been precisely calculated, it's estimated to be greater than the amount of atoms in the observable universe.
This comment should be higher.
Chris Harding That's why it hasn't died out
A minor point Chris, but chess games can end in "draws," not "ties." (A tie is a decision in which each players receive an equal score, such as can happen in Reversi or golf, while a draw is an ending that is not decisive, such as can happen in chess or cricket.) However, two chess players at the same tournament could tie in points. For instance, if I scored two wins, one loss, and two draws, while you scored one win, no losses, and four draws, we would be tied with 2½ points each (1 for each win and ½ for each draw).
This is a misleading sentiment and is a nitpick that misses the comparison he was obviously going for. Checkers, Chess, Go, etc are all effectively simple games with fixed and relatively simple rules that a computer can use to competently play the game as well or better than humans (which is his definition of "solved" here). Whether or not the games are literally "solved" is not actually the point.
Self driving cars are a much more complicated problem for computers to "solve" better than humans
We all ready have all most perfect self driving vehicles
Horses
My Pretty little Ponny
Horses trample people though, they don't give a shit.
@@TheMysteryDriver
Horses do not step on people on purpose. You have to hide the human under a blanket or similar to be able to get the horse to step on the person.
I have an idea that probably wouldn’t work. What if we make cars bigger with a horse that can control the car in the front.
@@theperfectmix2
It would be great for weddings and parades.
I had my first driving lesson after I got my learner's permit yesterday. It was the first time I had ever been on the highway and it shocked me how easy it was
"You'll never see the end of your therapy bills..." I love your genius humor!
Self-drive cars will communicate, so why would a rogue self-drive car be skating around on the ice? The other self-drive cars would have told it of the hazard and to slow down. "Who to hit?" becomes less of an ethical issue when crashes become much more rare.
But I do think it is a huge question that needs some answers, "Who will the self-drive car be most loyal to?" The auto manufacturer? Stupid government mandates? The owner? The safety of people in general? Will it rank human lives as far more important than pets or wild animals? "Could you please drive a little faster? My wife is about to give birth." "Sorry, I am mandated to not go even 1 mph over the speed limit. Regulations."
Well spoken Walt
Dude has straight up made me fall out of my chair laughing so hard. Easily one of my favorite youtubers
Thank you
I'm a chauffeur
I've been told that in five to ten years all cars will be driverless
Drive through Manhattan then tell me how a self driving car is going to get though all those situations
People don't use walk ans don't walk they just go
If a person knows a self driving car won't move with them in front of it
People will just walk in front and the car will never move
Agree with this but also people erroneously think car automation is like automation in other industries where jobs are replaced. Uber and Lyft drivers are NOT employees, they are independent contractors, so there will be no automation savings for replacing a person with an automated car. (see accounting 101). Prices for such vehicles will definitely be higher; manufacturers will need to protect themselves from liability, not to mention recouping R&D costs. Aside from that, currently every vehicle has a driver who is responsible for loss and also provides other essential services like vehicle maintenance and attention to passenger comfort. This is true for mass transit, limo services and private chauffeurs. I just can't picture any wealthy person giving up personalized service to mess around programming their destinations in their automated dashboards and being libel for all losses whatever the root cause.
@@Val_Emrys what? Uber and Lyft still has to pay their independent contractor drivers. We don't have enough information to know whether self-driving cars would save them money, but Uber obviously thinks so.
@@chaklee435 As I said, they are paying independent contract drivers now, but switching to automated cars will force THEM to be the owner operators who will have to provide maintenance, programming, and insurance. I am not even factoring in the purchase of cars. Do you really think independent contractors are going to be the ones to buy autonomous vehicles, carry all the liability and upkeep and lease them to Uber of Lyft for the same rate as today? PS. Uber and Lyft know full well there is no benefit in forcing a conversion to self-driving cars, but they are getting lots of tax breaks for 'researching' it.
Am I a bad person for laughing at the bowling pin sound effect?
I love how the title can be interpreted in two different ways and still make sense.
I might be a little late, but I want to add something. Radar can detect the radial velocity of moving objects by using the doppler effect (redshift/blueshift). You only need one radar puls for that. It is used for primary air radar and can detect the speed, altitude, and velocity of planes with disabled transponders.
About the highways - you clearly haven't seen many russian videos ;)
yeah Russians only need two wheels on their cars
russians only need a wheel a stronk boy to push it for their cars.
We should have those cameras in the United States.
Russia has one of the worst roads in the world and it will stay that way forever given the density of population and harsh weather.
Besides, Russia nas no highways, Russia has highways to hell.
russia: the only country where you can spot the sober guy because he drives slalom and buy alcohol in a gun shaped bottle (seriously its called vodka 47 ande is delivered in a ak47 shaped bottle).
OK, OK, hear me out. The answer to all of self driving cars problems is: Ejector seats.
Your car going to crash? No problem, ejector seat you to safety.
Car needs to decided between killing two people or ramming itself into a ditch, just ejector seat you first.
Car needs to be used as a shield, just ejector seat you first.
Assholes tries to override the autopilot to break traffic laws? NP, ejector seats.
I await my 5 million dollar check from Tesla in the mail.
Ejecto seato cuz!!
Unsafe situation in a tunnel? Ejector seat!
Unsafe situation on a bridge? Ejector seat!
Unsafe situation under some power lines? Ejector seat!
There should be at a thousand comments behind this comment with "ejector seats" at the end of them
sir kamyk clearly we just need floating devices, parachutes and 3 dimensional launch on the ejector seat. This is the perfect solution.
Well that should have been installed on the McDonald Douglas DC 10
When I was a kid in the 80's everyone told me we'd have flying cars by the year 2000, boy did their prediction suck! Even "educational" programs were on the team for this...what a terrible idea that would be! Why? Because 2 dimensions is the max for most drivers, and a fender bender at 100 feet would be at list a tiny bit more deadly!!
No one would accept the training requirements and maintenance of the flying car. We already have it today and it is called aircraft and helicopters and they are not directly common.
I think the noise pollution would be the worst concern for me. Flying is very noisy!
I mean- flying cars are relatively simple, but, like you pointed out, a lot of future predictions don't let feasibility get in the way of dreams.
Now maybe, if we ever implemented some measure of VTOL that didn't require venting a turbofan engine, then something as useful as the kind of flying cars in BTTF might at least be plausible.
But, that still doesn't solve enforcing lanes in 3 dimensions, or mechanical issues. Every car on the side of the road now, would turn into a 1 ton plummeting object, that would land on whatever or whoever is under it.
Now, maybe some kind of emergency parachute, warning system, and directional thrusters could land the vehicle in a safe location, while allowing people on the ground to keep clear, but even that could fail.
@@digitalutopia1 pfff ez you just call up atc request your altitude and done...lol rly tho imagine the shitshow of any type of attempted regulation...not to mention the livid pilots everywhere
Yeah. But flying cars have the COOL factor.
One of the biggest challenges is selling them to people like me who won't even touch an automatic transmission.
Save the Manuals!!
Want to control your own vehicle? Ok, straight to the gulag you go!
Some new cars have the option to switch into manual or stay in auto matic. Would you be OK with that or would you stay with stick with no gimmicks?
@@petrovstalingrad3994 1) Are you talking about dual-clutch? That's basically paddles. Some of us have considered it but it's regarded as an empty feeling. Sorry but manuals just offer more of a feel.
@@juanmanuelpenaloza9264Yeah I know what you are saying. I used to drive a 1990 Ford Taurus. I love shifting but my style choice of cars don't really have manual transmissions.
Barring the moral questions, one thing people tend to forget is that self-driving cars do not need to be perfect. They only need to be better than humans. That is a much easier goal to reach.
It's a fair point, one that I have often made myself, however if we can do better than humans then why can't we do better still? That pressure will always exist. The doing better than humans 'metric' is really the theshold for allowing them on the road in the first place. It shouldnt be the end of the safety improvement drive.
AI fan self driving cars already beat humans in safety today.
The companies testing self driving cars have to put out monthly safety reports in the US and they are safer than people already.
Society as we know it is based on a liability system...who will eventually be liable for any incidents that occur with self driving technology? The inventors? The developers? The programmers? The company selling the tech? The human who turns the self driving car from off/to on and presses "ACCEPT LIABILITY" terms of use button? :-\
But self driving cars killing select people based on software choices is a lot harder so swallow than blaming a human for inattentiveness: "Sorry mrs., the Ford software installed in the car that killed your husband, calculated that running him over was less liability than flying over the embankment into the wetland protection zone."
zach, as an AI researcher working in self driving technology research, I have read some of those reports. You'll find that they are operating under a narrow set of conditions, and yes various companies are productionising those systems right now. You can expect to see level 4 systems for sale in the next 4 years, though probably limited to highways initially.
I like this guy already. Beside the fact that he is smart and knowledgeable, I like the fact that he recorded this video inside a car when he was NOT driving, which means he puts safety first.
17:09 great singing can you make a music video
I'm glad someone else made this video for me, because self-driving cars are virtually impossible to get right. You said it best when highway driving proves nothing - intersections are where the vast majority of car accidents occur. Have a self-driving car make it down a street in Manhattan. Then I'll be impressed.
8:55 the number of human drivers that would screw that up cannot be understated
True for all of his stupid examples...
yep
*overstated
Oh yes, looking at that example I was trying to figure out what I would do at the intersection and was very confused for awhile.
This is not confusing. The traffic light takes precedence. The stop sign is covered, meaning ignore it. The detour sign is not the beginning of a detour, it is the middle of a detour, and is relevant only if you’re following the detour.
hit Celty - she can't be killed. not by mere mortals.
But the car is not mortal!?! But that probably doesn't count.
Though I suppose it doesn't sound quite as grand to say can only be killed by an immortal. But then can they NOT be killed by another immortal? If they can, then they really are not immortal.
Confusing stuff, magic and fantasy.
rationalmartian you don't understand magic - you feel it.
rationalmartian
If the Lich King who says he can be killed by no man is immediately decapitated by a woman, then a self driving car can definitively kill Celty.
The devil is the details as they say, and arguing semantics odesn't matter when dealing with ancient mystic powers.
@@kyriss12 Ackchyually, it was Merry killed him with a magic Dagger of Westernesse. It was enchanted with the power to harm the Witch-king of Angmar himself by a weaponsmith of Arthedain long before. When he stabbed the Witch-king in the knee with it, it distracted the Nazgûl and broke the spell binding his undead flesh to his will, allowing Eowyn to kill him by driving her sword into his unseen head, thus fulfilling the prophesy of Glorfindel that "not by the hand of man shall he fall." Literally anyone trying would have only cut the cloth of the wraith's hood before dying to said chuckling wraith if not for the relic Arthedain blade from a country destroyed thousands of years earlier.
This said, I think a Dullahan can only be killed by destroying the head, which Celty keeps very safe in the darkest reaches of Area 51.
The Headless Hessian Horseman of the Hollow however already had his head smashed apart by cannon-fire, but you can chock that up to differences in regional lore and myths.
Although any good chess program can beat any human, the game of chess was never solved, not even weakly. Prove that you solved the game would require that you provide the entire possible trees of all playable games. For chess this would be hundreds of Petabytes of data. Nobody ever did that. There are some allegation on Internet, but never a reference was provided (just like you, that provided an apocryphal date, but no reference whatsoever). Some types of endgames are know to produce some results, but that is the nearest that anyone reached. Checkers was solved by Jonathan Schaeffer, after a 16 years work.
Airton Granero he said checkers lol
@@MrDawgon I heard chess..lol
Petabytes is still a huge, huge underestimate.To memorize every possible game state will take so much space that we couldn't even get close to doing it if we combined every single bit of memory ever. The amount of Chess positions is such an absurd number that it mights as well be infinite, human minds just can't comprehend that big of a number. Technology is advancing at a terrifying rate, but I doubt we will even get close in the next couple of decades.
@@josephhanicak7922 Yes it is. Johanthan Schafer that I mentioned said: "The possible number of chess games is so huge that no one will invest the effort to calculate the exact number." It is estimated to be hundreds of orders of magnitudes higher than the atoms of the Universe.
@@agranero6 Oh yeah. There are estimated to be 10 to the power of 187 atoms on the universe. Thousands of times that is such an absurd number that you could reasonably use infinity instead
01:34 chess wasn’t solved in 2015 and it’s speculated that it’ll never be solved. Despite being easy for a computer to calculate game states, we do not have the computing power, the storage or the time to fully solve it.
The point is not to make cars that won't make mistakes, but to make diverless cars better than the average driver. Accidents will occur just like with manually driven cars. Many situations stated in this video are examples of accidents that happens due to human error when human drives as well. But yes we're not there yet with driverless cars but It's just around the corner I think.
And sadly "average driver" is a such a low bar that driverless cars are pretty much already better. They may popularize FAST because :
-Cheaper to produce, cheaper insurance, cheaper (and easier !) license.
-LOTS of people can't drive !
-Commute time becomes free time.
-Governments incentive. Reduced accidents, les wear, less infrastructure needed.
I mean, I feel like that doesn't really address most of the points he made in the video though. Most of the problems he talks about are reasons why a human driver will be better for a decent time longer now.
@@telcharthegreatsmithofthef7585 yes but self drive doesn't have to be perfect to replace human drivers. Just removing tired, drunk, high angry, shit human drivers. It's just a matter of time, then nobody will own cars as what's the point, u just have taxis.
If I'm in a self driving car, and it for some reason decides to strike a person or vehicle, would your insurance policy be affected, or will your vehicle's manufacturer pay for both damaged vehicles/medical bills/funeral bills?
But what about responsibility? In a human controlled accident it is almost always clear who is at fault but what about a driver less accident? Is it the car company? The person behind the wheel? The other driver?
What if no one is in the car?? Who is at fault in a self driving vehicle with no one inside
Chess has never been solved.
I would disagree, alphazero has solved chess. I'm sorry to burst your bubble. In just 4hrs of playing itself it beat stockfish in a 100 game match never lost a single game.
Not losing a single game doesn't mean the game has been SOLVED. To "solve" chess means to have determined the outcome of *every single possible* chess game. At last count, doing this would require more bits of data storage than the number of atoms in the observable universe.
@@rogermwilcox inaccurate, I don't need to process every possible 1st or 2nd move in tic tac toe to understand to force a draw, mark the center. (I am using a very basic concept for this)
It is well known there are irrelevant moves in chess in the first 2 moves (1w1b) that don't even need to looked down. significantly reducing the branching possibilities. also, Alphazero has millions of years worth of chess experience.
@@MTDurkee Sorry but being good or great or unbeatable like Alpha does not mean the game has been solved. All moves (as he pointed out in the video about checkers) have not been cataloged. Same with go.
Just take the 'L' Matthew.
@@MTDurkee Chess isn't solved until we've determined perfect play for both players
0:56 Real talk, what are those chess pieces made from and where can I get some of those?
They are sexy as all hell.
I got the durarara reference.
Praise me.
praiseth be the bismark, who recognized the helmet of best girl celty
"You're helping our future machine overlords recognize objects."
I have to say I giggled so hard at the outro this time, that was too perfect, especially since I had to explain why I went nuts at the addition to the intro making me VERY happy to my roommate!
Thanks for reviewing. We are developing the self drive out of the air. In contrasts, I think it much better to make it in rail for the real safety, but there must not be an actual rail on the road. As a developer, I don’t trust only the sensors much, unless car can talks each others during commuting.
Intersection with no trafic lights would be like a whole horror movie squished in 3 seconds :)
Ivailo Yanev won't.be intersections, just flyovers, that's happening now.in my city a lot if major.cross road intersections with light as being replaced with continuous over passed.
This video is going to be very useful for outsmarting Skynet.
Honestly, with all the automatic assists new, expensive cars have nowadays, they are almost self-driving. I remember some months ago a story about a guy who drove halfway through Poland while being so drunk he was unconscious more often than not, but with all the things that nudge your steering wheel and brakes according to cameras seeing road stripes, how close you are to other cars and stuff, he was totally fine. Until there was a damaged piece of road with road cones on it, and the automation didn't know how to deal, then he crashed.
The car doesn't really need to 'see' all around to drive semi-automatically. It needs a clearly marked road and other cars to behave predictably and a human that can occasionally intervene when something other happens.
God... of course one of the motorcyclists is Celty
Does that mean one should aim for her, as she's functionally immortal?
TheGameFilmGuruMan Yeah, that makes sense.
Another big problem with driverless cars becoming the norm is cultural. You kinda touched on this in the video. The United States for example is a car culture (from what I understand Germany is the same way) that enjoys driving and it will be a tuff sell to say "hey let this car drive you for you!" The average American (and I assume German) would say "Nope, no thank you. I prefer to drive myself." For driverless cars to work it would require a cultural shift or government intervention or both. And if we can't take guns away good luck with cars. Not to mention we'd have to spend trillions of dollars to get the all the roads, cities and high ways updated (mentioned in video) and we cannot get the funds today to just maintain our current roads. So forget driverless cars becoming the norm in 30 years, try 100 years assuming a lot of luck is involved.
DeadMarine1980 If that was true people riding horses to work have ended not to long ago.
people used trains and the main reasons many used horses (mainly in the country side) was because they were cheap and cars weren't affordable.Your analogy is stupid
Another reason Haden Wong's analogy is stupid is because the cars replacing the horses still required the rider/driver to control them, whereas driverless cars require no control from anyone seated inside them, nor can they be controlled by their occupants.
simple, gen 1 it can drive itself if you want it to, gen 2 well a steering wheel is a optional extra, gen 3 no one considers it, each generation is about 15-20 years.
dstbij 52 okay so 60 years, 40 years shy of my prediction. That is still more than half of the average human life span. So currently we won't see full automation.
Excellent video, making points you don't see made often. And concise too. I would add that computers lack our intuitive psychological capacity to intuit the intentions of others. A car may calculate a pedestrian's distance exactly, but they don't read pedestrian's body language.. you know like hesitation to make sure the care is slowing down, or the gestures we make that tell other drivers what we're doing. And think of all the info you gather from seeing the other driver. You can't see the car execute a program. Sure you can't climb into the other driver's brain. But you can do the next best thing.
*I will echo a comment I saw on another video that also covered ethical concerns of driverless cars: We can and do already kill thousands of people every year with human-driven vehicles and we consider it normal. While sorting out these hypothetical ethical questions is important, it's also important to note that implementing ANY rational solution will result in saving thousands of lives compared to the status quo, and unnecessary delay leaves us with very real, non-hypothetical blood on our hands. Don't let the perfect be the enemy of the good.*
You speak Truth.
This is true but it's also worth noting that blame is pretty easy in the current situation. Someone fucked up in a car wreck or accident but who fucked up in a self-driving car? The "driver" who wasn't paying attention or the AI not doing it's job properly? If it's the AI who gets blamed? There were probably literally hundreds of people involved in programming such a thing and it's not going to be easy to say who screwed up when code gets that complex. Or was it the streetmaps on google (or whatever company) incorrectly placing a sign or road conditions? Was it malicious people putting bags over signs or malicious drivers manually overriding due to poor decision making?
I actually agree that almost any decent implementation would probably save tens of thousands of lives even in just a few years but on some level I'd like to be able to decide that ruining someone's life isn't something people can just walk away from without consequence. If the motorcyclist situation in the video happens, I want to know where the prosecution lies. Honestly, though we won't have any of this solved until we just get around to implementing self-driving cars and wait for the law to catch up.
The self driving car is less likely to put an owner in the dangerous situation.
As more vehicles enter the system I forsee a mandatory vehicle to vehicle communication system that adjusts a single vehicle in concert with those around it.
If "your" car is approaching a junction, it will already know that another autonomus vehicle is approaching from another direction and possibly (?) that a human driver is approaching from a third(!), reported by an exiting "AV".
Maybe, even parked vehicles will broadcast the position of human drivers as they pass?
Don't let the wolf in sheep's clothes cloud your views. It is dreadful how many people are victims of car accidents, and to hear the actual numbers is stunning. The morality in our hearts quickly jumps to support any offers that helps reduce tragedies, and don't really think about other factors we have no knowledge of.
Then again do we really think this Technology will actually save lives?
These driverless car have a number of accident already in their testing, and most if not all have lead to a fatality. So if anything as of now it has saved no lives, and has actually contributed in the death totals for car accidents.
We must then ask ourselves how much more testing is needed? And how many more deaths will be added to the totals before they can perfected this technology?
Wnd that's assuming it could even be perfected.
What if it can't?
Then the lives it took during its failed testing would all be in vain.
Let's say your hopeful, and believe eventually it will work. Then you say the lives of a sudden few now is worth it when it saves millions later.
Okay well that's your prerogative. But are you then really trying to make others believe that you actually care about people, and just want to end the suffering caused by car accidents?
So what about the countless number of people that will be put out of work when drivers are no longer needed?
I'm sure there will be suffering there.
What about the growth in homelessness that we will reach when AI and robotnics takes over all labor jobs. The blue collar worker will be left jobless for lack of education.
Where's your morals there?
Plus do you really think this will be good for the Economy?
What will this do to us all as a country? Will our nation be able to overcome the poverty while taking care of all of its citizens?
The wolf in sheep's clothes is playing our hearts. It doesn't care that people are killed in car accidents every year.
They only care about driving every competitor in transportation out, and being the only source of transportation so they can do what they want with the market.
This is far more harmful than good, if any good at that.
This in my opinion would end our nation as the world power, we won't be able to fund our military, we will become vulnerable and unable to defend ourselves, and will inevitably be invaded and destroyed. All because you allowed the wolf to play with your marality and good heart, and supported their actions.
@@Kobolds_in_a_trenchcoat Who do we blame when a rock breaks off a cliff and kills someone?
Nobody. Shit just happens.
And in the time the driverless car crash investigation, the first 5 hours, 10 people died from human-related crashes.
Current standards for the good driverless cars are already better than humans.
It's only a matter of time.
Gradual introduction of self driving technology is the best approach, modern cars already have things such as parking assist, blind spot detection, and sensors for when a car is too close. Eventually there may be 'smart roads' that are designed to support fully automated vehicles and manual overrides are restricted on such roads.
Yes there will be bumps and legal quandaries along the way, but if the end goal is the complete eradication of road accidents and fatalities, it would definitely be worth it.
You should know that human drivers can never be banned, not until self-drive penetration at least hits "critical mass" at which the public is ready to do away with human drivers.
And the transition over to self-drive cars, is far too "gradual" already. The more time that they dilly-dally around, trying to get it just right, horrible human drivers are still killing people in crashes that could have perhaps been avoidable, had we moved more quickly over to self-drive. Why can't they get more serious about self-drive, and put those computers into the airplane pilots' flight simulators, and throw a bunch of scenarios at them, and get them to handle them correctly? Put the self-drive cars onto the track and throw scenarios at them. Hose down the track in freezing weather, and see how well self-drive computers can drive on an ice skating rink.
11:14
Program it to shut off the motor and shift into 1st.
If it's an electric car, maximize regenerative braking until it can safely reverse motor rotation.
Or just hit both guys. Why not?
I found that stupid too, it's a computer, it doesn't need to use the brake necessarily to shut off the car.
@@aidancollins1591 Also, most of the scenarios he mentions are incredibly rare.
How am I such a car nerd that I know you're filming from inside a Toyota Rav 4 haha
Finally, someone who isn't a tech nerd who knows nothing about cars!
Solution to the "WHO HITS WHO?!" situation: the owner of the vehicle performs a quick 25 questionnaire about moral dilemmas. Stop bringing up this "this will happen in this fake scenario and the programming masterminds have to put their code in which will elect their own choices" and whatnot because it's really not that hard.
It's literally as simple as "SCHOOL CHILDREN ARE CROSSING THE ROAD AND A SELF DRIVING CAR CANNOT STOP OR TURN AND YOUR SELF DRIVING CAR CAN SAVE THEM WHAT DO?! A) HAVE YOUR VEHICLE DRIVE IN FRONT OF THE ROGUE SELf_DRIVING CAR AND STOP THE CAR OR B) DO NOTHING"
Pedestrian bridges typically have some sort of protection on them to prevent people from falling over. In Vegas they have glass panels and one nearby me has a metal cage around it. No ones going to fall in those.
Celty doesn't ride a motorcycle, just a _headless_ horse.
Love the durarara reference!
VECTORS!
It is easy to see if an object is moving towards you, away from you or is doing neither. It is called the Doppler effect. We use it all the time in when measuring the speed of a vehicle. Bats use it hear if something is moving trough echolocation.
One big problem with radar is if you have 40 automobiles in a segment of Street how many different frequencies of radar are possible? At any one time you could have a few cars canceling each other's radar out.
It's like how when I was helping to move my sister from one apartment to a different one we had a big green station wagon and I would lock the car and bring something out unlock the car and put it in. At one point I went out to the car to get something that was next to the driver's seat so I came around that side unlocked it, went to reach in, and realized it was not our car. Two spaces up there was an identical station wagon to ours parked there. Neither station wagon was even remotely new. One thing that was different is that we had come from a neighboring state to where my sister was to move her.
Different cars, identical keys...
Dogs driving cars?! I'm shook! This is now my new UA-cam obsession.
11:23 Durarara!!!!!!!
@@mypenisisunbelievablysmall5258 the biker in black suit and with cat ears on the helmet is a character from anime Durarara. I totally recommend watching the first season!
Savvasaam you have fallen into a trap 😂
I have seen a lot of good videos on this channel, but this one is good enough to get me to subscribe. As a techie for more than 50 years and deeply involved in electric distribution automation from 2010 until recently, I am astounded at the public perception of it being simple. If there were algorithms for everything a driver has to do we would already teach drivers that way and would not allow anybody behind the wheel until they could demonstrate familiarity with each one. Instead, we are constantly having to figure things out, which a computer simply can't do.
17:09 (for people who just came here for the singing.)
4 way intersections are horrible roundabouts are a lot safer, except for some reason nobody in us knows how to handle them
4 way stop signs are one of humanity's most awkward inventions ever.
There are some who support roundabouts and some who don't. Edmonton, Alberta, Canada has a lot of them, but where traffic volume has increased, they're being replaced by conventional signal-controlled intersections.
@@heronimousbrapson863 stoplights > roundabouts > stop signs.
As someone who works with self driving cars we can do 6 way intersections.
I hate this US argument... I lived on cape cod most of my life and we had roundabouts everywhere. Not sure why people say they aren't here in America .-.
"You'll never overcome" = we haven't figured out how to do so yet. It's been true throughout history so far. The more we can get the entire human population connected to all the ideas and theories of humanity and contributing their own talents and theories to challenge what we assume is possible, the more likely things we "can't overcome" become footnotes in history of how backwards our own generation was in its thinking.
Pseudoscientific nonsense
What about death?;-)
@@MRW2276 we'll overcome that too.
Even today people are working on ways to defeat the heat death of the universe
Just want to correct one thing - radar can detect object speed, by using frequency modulation or simply measuring frequency shift of incoming signal due to Doppler effect.
It can only detect speed parallel to the direction vector of the photons.
If an object is travelling perpendicularly to said vector, there won’t be a doppler shift caused by said object (although there can still be a doppler shift caused by movement of the radar, as would be the case with a self‑driving car).
*TL;DR:* The measured velocity will be the actual velocity multiplied by the cosine of the angle between the radar beam and the object’s velocity vector minus your own velocity (technically addition, but one of the velocities will be negative to the other when travelling in opposite directions, which is probably the case in this example).
*TL;DR;DR:* vₒ = vₐ × cos(θ) - v₀
Also, this is simplifying things quite a bit.
The hypothetical situation I love is the kid running out into the street, and if you swerve to avoid the kid, you go over a cliff. How is this kid allowed to play with a ball near a busy street that's on the edge of a cliff?
Good I really don't want them to come. Getting rid of the ability to drive rids us of a certain independence that is just to satisfying to give up.
Hey man. True fan here. Love your videos. But need to sort of hip you to some possibilities that people don't consider:
1. Early iterations of self-driving cars have to work basically like people do. Future iterations may not.
2. For instance, self-driving cars can be networked and be required to have transponder capability. (Much like airplanes). Your car won't have to SEE other cars to avoid hitting them. It will just know where all cars are, nearby, how fast they're moving and where.
3. Visual navigation is fine and self-driving cars can do it. In the future, though, roads can be constructed with inexpensive optical and magnetic and other types of sensors that let cars navigate even better. For instance, your car could be reading mile marker bar codes every 1/10th of a mile. Combined with their data of information (think of wayze) they can know that a traffic slowdown is going to occur in 1/4 of a mile or whatnot. Your car will be aware of the environment to a degree that's impossible for a human.
4. Visual recognition of NON CAR objects? Man, we're not even in the 1st inning of a 9 inning game, there. As I tell my students "once a computer can do something, even badly, the game is over, because the computer will get relentlessly better, 24x7, forever." Don't judge future capabiltiies by today's technology. Our technology will improve. Insanely. They will exceed human capability. That's not in some distant future. That's in the next 5 years.
5. Keep in mind that we kill, what, 40,000 people a year on our highways? Something like that. And how many of those are due to drunk driving? Inattentive driving? Sleep-deprived driving? Self driving cars won't suffer from any of that. Once this technology is fully fleshed out, we could literally reduce our fatalities to something like a few hundred people a year. Maybe a few dozen. That's basically like curing a form of cancer.
So, yeah, don't think of cars in terms of how well they imitate people. A wrench doesn't work because it's exactly like a person's fingers. And technology works because it does things humans can't.
Self driverless cars aren't a special messiah that will solve the problems of roads. Like already mentioned, you can't have everybody own a driverless car, specially the low-income, antique, and manual drivers.
The annual death count will merely be reduced by a bit, even if we have superhuman cars with ultra fast reaction times, it isn't enough to sway a 1 ton metal machine at 60 km/h to avoid hitting a child, while preventing the cars behind from cashing where ALL of them are self driving. And unless you can get cars to weigh like 100kg, roads with optical sensors and constant QR codes, while being maintained for an immense amount of cars going by, won't be feasible economically in the tiniest bit. Just spend money on public transport than trying to figure out how to find an iron needle in a haystack without a magnet
I will never sit inside a vehicle i cannot get to the controls of in case of emergency, i need a compromise where i can still drive but when the car detects danger or rogue traffic then it takes over, that i think is the future as it does not take away peoples ability to drive there own car but also has the safety of the faster computer mind.
@@Historyfan476AD A lot of cars already have that with automatic emergency braking and lane detection.
@@MichaelG485 I know, But my point stands.
8:30 yeh that's not a problem here in the UK. The only hexagonal sign on the road is a stop sign. Even when the stop sign is completely obscured by snow you can still tell just by the shape.
Pssssst. Octogon, hot hexagon. :)
I think the main thing that people want from self driving is not safety or speed but the fact that driving requires literally zero effort. Self driving cars will never be significantly faster than regular cars. We will always have intersections. Traffic will get worse. But none of it matters, because you do not have to deal with it.
Governments will not need to intervene to make driverless cars a thing; insurance companies will as soon as these are demonstrably shown to be safer than human drivers. The driverless discount on auto insurance will help drive the driverless market. As far as the strawman scenario "who should the car hit?" the car will do it's best to avoid *any* collision and even if a collision is imminent will take steps to protect you from more serious injury. Cars are *already* being marketed with these crash protection and avoidance features where the car operates semi-autonomously. www.iihs.org/iihs/ratings/crash-avoidance-features
At least we'll be able to drive about stoned or fucked up now.
If it's fully autonomous, liability actually seems fairly simple. If it was a manufacturer or electronics error, it's the manufacturers fault. If it was a failure caused by improper upkeep, the driver's fault. And the fault of the person hit if they endangered themselves.
'If it's fully autonomous, liability actually seems fairly simple'
No it doesn't.
Sheepdog Wrangler Gosh, the word "no" without any reasoning at all sure does change my mind!
Daza Mistwalker yes but car manufacturers will counter this heavily, especially those who produce cars that are used for moding, muscle cars, sport cars, trucks, and off roads
I only see from one eye, and although there's truth just on how I get depth cues as explained in this video, one thing is often forgot is that no one see in static images, one moves around a get to model a 3d world even if only has access to 2d input, perhaps not as easy as once with two eyes but the brain is kept busy trying to infer from the data it collects
I would just like to remind ppl of the recent update that an appliance company sent out to their high-end microwaves that bricked every microwave that received that update. The update was sent out on accident and they have to send a technician to everyone's house individually in order to do a reset on the microwaves to get them to work again. It's not that rare of an occurrence either. The Internet of Things (including cars) is just not a good idea.
Just test drove a model Y yesterday. It drove for me the whole time. A lot can happen in 6 years. They’ll all be driving themselves pretty soon. But these philosophical issues are still pressing.
Wait 'no way to see depth'?
Isn't there a camera that uses Paralax just like we do using the Kinect camera to do exactly that?
Yes, and computers have had no problem at all with depth perception for over two decades now. Its very easy to achieve using infrared sensor technology. This guy is a moron
He was talking about inferring depth from a 2D image, which humans are surprisingly good at (he mentions some of the clues and stipulates the computers can make use of some of them, so he doesn't say it can't do that at all, only that it's a hard problem which will at minimum take time to solve). He also notes that computers can use echos instead (tho he should have made more explicit that LIDAR is the most relevant one, using Time of Flight of light, rather than audio or radar).
YEAH you sing those tunes KB!!
Actually, depth perception from binocular vision only works for short distances -- threading needles, not figuring out which of two objects across the street is closer.
As a chess player I must point out, chess is not solved, at all. Engines still play against eachother all the time to see who is the best.
Deep blue had a hard time wining. It lost a couple of times and was specifically tweaked for one special opponent. Even if afterwards it won against said opponent, it is assumed that it would be in disadvantage against other players because it was very much tweaked.
I know this is late, but that is a very myopic view. What individual, when competing at the highest level in chess, wouldn't observe the common strategies of their opponent? You leave out that Deep Blue's opponents also spent a significant amount of time studying tactics that exploit the weaknesses of computers.
@@DiomedesStrosMkai Not everybody played anti-computer chess against it. And it wouldn't have mattered because Deep Blue was really good in it.
All the "moral issues" of the driverless cars are moral issues of the driving itself. What's the difference between a car making decision or a human making decision? Decision to not act and cower your eyes is still a decision, even if cowardly or panicky one. But we have this extremely weird cognitive bias that it is way more acceptable for human to chose the wrong thing, than a machine.
For driverless cars to start replacing humans, they don't need to be perfect in every situation. They just need to get in car accidents less often than meatbags, statistically.
Gabriel Anderson Doesn't change the fact its safer, and hence objectively better to use self driving cars.
you're out of your mind if you think im gonna let a machine decide if i live or die. in every one of those examples he brought up its an easy choice for me but maybe not for a computer. i will always choose self-preservation over trying to save others. I'm not gonna get a car that's gonna try to be a hero with me inside as a sacrifice. two bikers coming? welp one of them is gonna die cause it sure as hell wont be me. kids crossing the road and my brakes are out? well sorry parents but i got a future too. I would never put my life in the hands of 1s and 0s.
Darksypor you have some wacky ethics. You would allow yourself to run over a bunch of kids and wouldn't make any effort to prevent this?
in the hypothetical situation, the brakes are out and there's no stopping the car. Im not gonna throw my damn life. Thats called self preservation. Im not gonna let a damn car decide to sacrafice me for someonelse's life.
So, basically, you are that scumbag who betrays everybody in horror movies to save his own skin?
A week ago or so there was an accident in Poland, when one famous journalist drove his car over 300 kilometers, on a freeway... while being completely, utterly wasted. Turned out the reason he managed to drive so far, being that drunk, without any trouble, was that his car did most of the driving for him. It adjusted the right speed and remained stable in the lane. However, he reachead a point where there were some road works, and the signs and road lines were replaced with a temporary ones, organizing the traffic for the time of the construction. The car apparently couldn't tell the difference and got confused which lines to follow, and started to drive through traffic cones. As a result, the guy got caught.
This is a great summation of the issues involved with autonomous cars. The key to self-driving cars would be achieving the right balance between autonomy and centralized control. Given a strong enough up/down link to a central computer, all the cars within the zone of control could be controlled to such a degree that they will avoid hitting each other because they're all being given instructions that by design and definition do not conflict. Balance that with the ability of cars to sense and respond to their surroundings when outside central control, along with exactly the right hand-off process, and you have it licked. But as you noted, that ignores real-world events that can't be controlled by the computer. In short, well thought-out and well presented.
You're my new favorite channel! Hope you get a good upswing of subscribers soon!
Another issue is going to be getting lower income people to buy into this technology. People in this class already choose older used vehicles just to have a way back and forth to work. Getting them to invest in a new driver-less car is going to be difficult. Even when driver-less cars have been around long enough for used ones to be on the market, they will be prone to having software that is either out of date, going out of date or obsolete and unsupported, including the hardware needed to runs these processes. Lower income people have a tendency to hold on to a car till the wheels fall off before buying a new one or fix the current one themselves because the budget just isn't there to support buying the newest, latest model every time one comes out. There has to be a huge change to work economy to make full adoption of this technology feasible because the cost to go into this kind of structure and maintain it is going to be massive. Who is going to foot that bill?
Yes, I was thinking about that too. Not to mention repair costs.
I think the most likely approach would be, as others have stated, a lease model where some company owns the car and provides maintenance and insurance as part of the monthly fee; it would also get rights to the collected data of course. If some particular model has problems, they would be removed from the road (see clause 45 in your contract).
As for realizing the advantages of all automatic (all self driving)...
Dedicated lanes on some highways could be one of the early approaches, likely first replacing HOV and toll lanes which are already segregated to some degree. IF all automated cars can go zipping by with high speed, minimal separation between cars, and high safety, that would be a massive advertisement to the other lanes. As more cars to automatic, more lanes would become dedicated (massive highway infrastructure changes, but over time).
I really can't see city or rural roads going all-automatic for a very long time. The automated intersection seems far off.
dekoldrick i will never own a new car. vintage vw for life. rebuild the engine n get another 300,000 if not 500,000 miles. 🍻
That's my plan for maintaining my vintage accord and F150. Get a motor for both, rebuild them and keep them on standby just in case the current one goes. Drop the rebuilt unit in and rebuild the old one if possible or replace it. Keep that cycle going until either they get wrecked or I can't drive anymore.
The ability to summon a car from an app on a smartphone will eventually lead to people sharing a car with family and friends, eventually this will lead to car sharing organizations which you simply join and summon a vehicle from the organization's pool of cars. This would will allow people to avoid car ownership and the associated costs completely hence enabling the less affluent. If we really want to press the technology to support less affluent people, we could implement public transit variations such as PRT which has actually been possible on guided through fares for at least a generation.
However, the sharing of vehicles will likely lead to fewer vehicles per family so what will the car manufacturers do to stem the loss of market? I suspect that they will emphasize and market various stages of semi-autonomous vehicles to both reduce the market for fully autonomous vehicles and deflect the public from wanting full autonomous vehicles. I think we'll already seeing this in effect with the SAE categories of autonomous vehicles and the major manufacturers releasing assisted driving features while only companies without a base in automobile manufacturing such as Google and Waymo focusing on fully autonomous vehicles.
heres a tip. if your breaks ever go out in your car while youre in rout to somewhere, the only thing you have to do is slam the gear in reverse or park. there was a terrible accident that happened a few years ago where a family was traveling and the gas got stuck somehow and the car was accelerating. the breaks werent working. they were coming up fast on a 4 way stop. everyone in the car that day, perished. they tried to call 911 and were actually talking to the dispatcher for about a minute before disaster struck. it was such a tragic situation because nobody knew how to help them. they were so afraid. there is no way i would allow a computer to drive my car when stuff like this is still happening with cars. we rely on computers far too often
Hey, wouldn't you agree that trying to code in preference in complex moral dilema would cause more death through malfunction (due to increased complexity) than just coding the car to avoid obstacle and follow the law?
By the way, you're one of my favorite youtuber, your videos are thoughtful and interesting, on various topics, I love the calm and rational approach you seem to have, and you should have way more suscriber.
Maybe it's because this is still a small channel, but the comment section also seems friendlier than elsewhere... You guys are also awesome.
The problem with saying "just leave it simple and have it follow the law" is in the end, legality. When YOU freeze up and end up hitting someone, most people - including the injured, the family of those who died, or a jury - are able to put themselves in your shoes and at least see how they could have done the same thing. If a computer-driven car does it... someone has to be at fault. Which means likely, the car company. People can forgive other people. People rarely forgive corporations. The first kid who gets run over by a driverless car (which let's be honest, it will happen eventually, there's no way it won't) the lawsuits and hysteria will be so real it's doubtful many other car companies or people will be pro-driverless. That's why this is so complex and difficult and can't simply be solved by "as long as it's only slightly better than people drivers, it's better for everyone."
Also, yeah, my comments tend to be pretty nice. I'm not sure if it's because I present my content from a middle-perspective or because I'm just lucky... but I hope I continue to grow with this crowd. :)
Why cant cars just do what most people would do now while driving? If I saw a car speeding down about to strike some kids i probably wouldn't pull forward. If there were two motorcyclists and I couldn't stop, i'd hit the idiot without a helmet/protective gear that's driving on the wrong side of the road basically asking for a darwin award. They guy with the key going to open his car, i'd keep driving and so should the car. If the guy was about to walk out onto the road then the car should stop if there's enough time, but if there's not enough time then there certainly wouldn't have been enough time for a person to stop so the dude who walks out on the road without looking both ways gets hit obviously.
The only problem i can forsee is that if everyone starts using self driving cars and the definition of "do what most people would do now while driving" becomes meaningless.
NthMetal watch a movie, read a book, enjoy the scenery, have a beer, sounds good to me in peak traffic
Pretty much my attitude. Human drivers suck, just need to suck as much, or be slightly better and I'll be happy. I don't imagine a driverless car will sit 2-3m from my rear bumper on a dangerous windy road beeping at me for not going above speed limit! Human driver, all the fucking time! Maybe that's what is needed for driverless acceptance - fully automated vehicles that sit on your arse and honk at you if you're not going 10-20kmh over the limit. Even in the situations you prob shouldn't.
i cant believe this man burns his hand everytime he makes a video, so much dedication.
A big risk of self driving cars is, they might destroy public transport in a lot of areas. So it causes more emission.
public transport is also way easier to organize and can transport people much more efficiently. Either by rail or drivers, even all of your human drivers can be extra eductaed for the job
As a Truck diver this was very interesting to watch. One large gap that I rarely hear people discus is changing road conditions. It is not always easy to detect slippery roads before you have an issue. Vision and radar are not going to be able to tell a car what a safe speed to drive is when conditions get bad.
difficult for a person to identify* As per the beginning of the video, things people did hard, computers can find easy, and vice versa. the car is also going to have very precise measurements on how it's driving affects how it moves and could probably be testing constantly to check how slippery the road is now, which a person would get bored/distracted of, or not have enough precision to notice small differences
I understand that the cars will probably be constantly measuring acceleration forces and compare them to the inputs it is giving. I have a hard time beliving that whatever test it runs it is going to be able to tell a difference until traction is lost. A classic scenario is you are driving along and the temperature is in the mid 30s and there is some lite rain. the temperature slowly drops below freezing how is it identified that the road surface is icy and needs to slow down to a safe speed. What is the safe speed if other cars on the road are unaware of how slippery the road is and continue driving at the speed limit or even above it? I am sure the cars will be programmed to drive under a variety of weather conditions but how quickly does the car decide when it has changed from one category of raining to freezing rain or snowing to sleet.Another one is it starts raining on a hot day. when it first starts raining the road can be slipperier because of a slight film of grease and oil from the surface of the blacktop that gradually washed away.
sorry to tell this, but detecting the road-tire grip is quite easy and (some) cars are doing this already in realtime. these constant, millisecond scale measurements and instantaneous torque modifications available in an electric powertrain are reasons for tesla to be able to beat lambos in 1/4 mile while driving with much narrower tires.
As someone from Canada, I'm happy with just the tesla system. Driving in the city is whatever, but if I'm going from Vancouver to Prince George or Calgary to ft. McMurray, staying properly attentive while essentially doing nothing is really hard. I would like it to reach the point where I could sleep or eat while it's doing this, would hugely increase mobility, routes like Vancouver to Winnipeg are routes that either need two drivers on shift (and that is exhausting and a lot) or a night somewhere on the way, (and that's still pushing the trip a lot). Being able to sleep in the car and eat would make everyone in the country a lot closer.
Yes, riding is so much easier than driving. And I have long ago noticed that I can notice and appreciate so much more scenery, riding with somebody, than having to keep my eyes glued to the boring road. I think it is crazy, the very idea of driving, being always but a few inattentive seconds away from crashing. Humans should not have to go through that. And why do I feel like my brain is running a complicated computer program when I am driving, having to constantly scan for hazards and to remain constantly alert for up to many hours? The human brain was not designed to run computer programs, but rather to write them. Computers should run computer programs. Computers should do our driving for us.
Perhaps soon, someday, when people finally get to ride in a competent self-drive car, their attitudes will change. It may feel that it is on some fixed invisible track on the road, as the car should not need to wander around in the lane, except to avoid obstacles such as potholes or tree limbs or whatever. People may feel even more secure with the computer driving?
Somebody some years ago, typed to me a reply that the government will program your self-drive car to kill you. I replied back, "You have been riding on self-drive elevators for decades." Of course the elevator only has to make simple decisions, rides on a track, and does not know who you are. Still, I think that self-drive will prove to be far safer than human drivers, and then no longer will be have to worry about becoming drowsy, and we humans will be able to go to more places and our useful traveling range will become farther. How many activities have I cancelled, simply because I felt a little too tired to commit to driving that far?
How wonderful would it be to be able to curl up on the back seat, sleep soundly for hours, to have our car finally wake us up, "You will be arriving at your destination in 5 minutes. Wake up sleepy head." Imagine all the people will will actually be able to go visit family a bit more often, without having to deal with that grueling long drive. And your car could also check the traffic and weather conditions, all along the route, and plan accordingly, something that human drivers very often do not bother to do. Maybe reroute to avoid steep roads, if there could be snow.
@@yosefmacgruber1920 Just get on an airplane.
@@briancrawford8751
For a short 30-minute drive, getting on an airplane would take even longer, well except in Alaska in which airplanes are about as common as cars.
Someday, cars will fly and be fully-autonomous self-fly.
@@yosefmacgruber1920 Uh huh. Yeah, okay. You can't stay awake and attentive for a 30 minute drive, so you need a self-driving car? That's messed up. I have to disagree that self-driving cars will succeed in the Eastern US. All it takes is the stroke of a Governor's pen on a bill passed by a state legislature to outlaw them, and given how powerful the AFL-CIO or "Teamsters" are, I predict that they will be outlawed before they hit the road in many states. In my state of West Virginia, there's no way a self-driving truck of any kind could make a journey from one town to another without some kind of catastrophe occurring. I couldn't see a self-driving car doing it except in well planned neighborhoods while all the kids are at school. We have roads that run between a rock wall and a cliff with no way to pull over, single lane roads on which people travel either way, and the occasional rockfall.
@@briancrawford8751
I did not say me necessarily. But do you really think that I want to work a long shift, and then have to drive for yet another hour before I can rest or sit down and relax? Maybe that is why so many of us are loath to look for a job in another city other than where we live. Can you not see how this horrible manual-drive concept is not only so yester-century but is seriously limiting our options and making the roads less safe? The problem of tired driving is so bad, that even the lousy government has put in a lot of rumble strips in the shoulders of the freeway, to keep drivers from dozing off while driving and crashing. Think about it, feeling that rumble would seem to say, "Stop driving and find a place to rest" and not just "Try to stay awake." It should say, "Just push the auto-drive button" already, well if there was one that is.
Ironically, one of the very first episodes of _Knight Rider_ depicted that same scenario, of Michael Knight falling asleep while driving. At least K.I.T.T. had the fully autonomous self-driving (AI-based) feature, and was able to assume control, and then helped him to come up with an excuse to tell the cop who pulled him over. Michael Knight had taken off in a hurry, without allowing time for the many amazing technology-advanced features of his "company car" to be explained to him, I think for a dramatic excuse to introduce the audience to what that 1980s TV series was going to be about.
Self-driving would make those treacherous drives a lot safer. Computers can actually rehearse what they are going to do, and do mathematical calculations of what speeds are safe to take a curve at. And you seem to forget that self-driving cars may perhaps learn from the experiences of other self-driving cars. Humans can not be trained anything like that, well unless they believe every textbook claim (that they didn't even bother to read) with no proof. Textbook says, "You can not safely drive while drunk or stoned." "Ha! Ha! I am the exception. I am special." Yeah, especially stupid. Textbook says, "You should follow the 2 or 3-second safe following distance rule." "I have been driving for decades, and I am an over-confident good driver, as you can see by my torn-up front bumper, so I should be allowed to draft right on somebody's bumper, like some sort of hemorrhoid, because I haven't been in all that many accidents."
Self-drive is the future. What stupid public is going to allow for the future to be outlawed? Anti-progress DemocRATS? I think it will come to be seen as a safety feature, as self-drive improves and proves itself. And what union is really going to be against paying truck "drivers" to sleep on the job? If they are in the back sleeping while the truck drives itself, surely they have to be paid for that? We pay drivers to drive while they have the cruise control enabled, right? It is just a driving aid or tool. We pay drivers to drive, even while they run the heater or air conditioner. Having such comforts are just part of improving the working conditions, they are not financial compensation for doing the job. The human driver is still the driver. Or we may have to revise a few terms. Instead of "truck drivers" how about "truck riders"? Hey, I would love to have one of those modern self-driving trucks if I get to keep my job, and don't have to so much "go to work" until reaching the destination or having to refuel or whatever.
You do know that at some point, self-driving vehicles will be talking to one anther, right? So in your one-lane bridges or narrow passages, they can time their arrival and alternate their passage so that they can share the 1 lane. Have you seen the YT video computer animation of how self-drive cars will just sail through the intersection without stopping, because self-drive cars will request a reservation to go through the intersection, and then time their passage so that they do not collide? Wouldn't it be nice to largely eliminate stop lights, well except for pedestrian crossings?
Also, a self-driving vehicle can navigate through difficult driving conditions, without feeling stress, because stress is a human emotion. Imagine having less stress, as your very-competent self-driving vehicle gets you home in a difficult blizzard, without getting stuck, when you might have found a place to pull over and give up until the weather improves. It would know of which slick spots to be extra-careful about, as the other self-drive vehicles would know all about the current conditions and share that info. Imagine having to take fewer pit-stops and rests on a long journey, because computers just do not get tired after 10 grueling hours of driving. They remain just as alert as ever. A computer is not a biological organism that needs a rest cycle.
Because nobody really trusts the self-drive technology, it thus improves the software engineers' dedication to make sure that it is as safe as possible. It obviously is not a fair competition between human and computer drivers. Computers are expected to do the job perfectly, and why not, if it is possible? Computers would do exhaustive scans of their surroundings, hopefully seeing early a child or deer approaching the road, not being subject to "freeway tunnel vision", nor being already tired after a long work shift. Also, in Driver's Ed, they teach you that a ball rolling out into the road could soon be followed by a child retrieving the ball. Self-drive cars could relay that info to the several self-drive cars behind them as well, to slow down, due to a temporary hazard.
When humans wreck a car, no big deal, right? That's human drivers for you. Why we have to have liability insurance supposedly. When a computer wrecks a car, there is so much more evidence. Log files will be picked apart. The dash cam videos will be studied. Much like when airplanes wreck, they will find out why, and release a software update to fix the problem.
But that is increasingly becoming a problem. Today's high-tech devices are only as good as their crappy software updates, and just how many free software updates are we entitled to or do we expect to get? Will your self-drive car be "outdated" and get no more software updates after just 8 years? Smart-TVs seem to only get updates for what? Maybe 2 or 3 years? So I am supposed to go buy a new one, in just 3 years? Think again. I usually use things, until they just do not work anymore and I can't somehow patch it together anymore.
So will the greedy insurance companies, complain that self-drive cars are just way too safe, and find some way to sabotage them, so as to jack up the price of their largely-a-scam socialism share-the-risks insurance?
I actually put this video on to keep me company while I was working but it was so interesting that I stopped working and focused on the video. Great work!
Oh you also introduced me to a new song. Thanks!
Driverless cars are more hype right now… sure they do well in certain situations but they don’t in others. There are many edge cases that will hiccup any plan for Level 5 autonomy where we can get rid of steering wheels and pedals. Just the other day a “geofenced” taxi drove behind a car stopped where the owner forgot to turn off the lights and the taxi stayed there overnight.. it interpreted the car as being stopped so it just waited.
Oh come on...Until 10minutes I was: "ok sounds reasonable, but solvable with enough time and computers"...But then the dilemma situations came again...This is just scare mongering imo. If the car is designed so badly, that it drives so reckless that it cant stop and not hit one of the two bikers then it shouldnt be on the road in the first place. Also simple physics, the best thing to do in those situation is to emergency brake, it gives the bikers more time and reduces the total energy of the crash, if there is no sensible way to just avoid them.And the school class example is just soooooo unlikely, that I dont even want to start talking about it but ok. Even a selfdriving car will have an E-brake, as has every other vehicle on this planet, that drives on a road or rail. This will not happen! Its just sooooo unlikely that having less reckless or old drivers causing crashes outweighs these freak accidents by a mile. And in the end it just boils down to lawmakers. If they make the car manufacturers responsible in those situations, then they should also be responsible for all the deaths, that occur if you stop self driving cars...Because it then is Their decision to let people die because of too old or stupid drivers
KOrbiid
Agreed. First half of his video was very informative and resonable, but then became a strawman argument video toward the end.
What if car apply emergency brake and the one behind it crashes into yours?
Yet, we now have at least a few examples of humans that have died at the hands of automated cars. We had a recent situation in Arizona where someone died due to an automated car that WAS being attended by a driver. We also have a report of a Tesla that had a sensor failure causing the death of the occupant of the car. This doesn't include the large number of potential non-life threatening problems that automated cars can still be a part of, and the fact that they are not immune to new problems we don't even know yet. Just ask someone who relies heavily on computers to conduct business when the business network was down -- and that's dealing with a fairly static set of variables (i.e. computers sitting on a desk plugged into a wall or connected to wireless).
Personally, I think the problem that needs to be solved more than anything is how to we make existing cars safer NOW. How do we reduce distractions and educate all drivers of the horrible dangers of distracted driving? Even if we get automated cars on the roads within the next 10 years, we still have a large number of needless injuries and deaths that could be prevented. Since one of the BIGGEST contributors to accidents is distracted driving, what are we doing NOW to reduce distractions in motor vehicles? I believe these are far easier and cost effective technologies to implement. Things like lane departure, crash detection, cross-traffic warnings are all great technologies that can help reduce driver error.
By all means, we should continue the research for automated cars and continue to work on making all cars safer with these technologies. However, I just don't see driverless cars as the Utopian solution to this problem.
westdccs the other side of this is that that car, and all other cars can learn from the experience and better avoid it later. I'm not saying it is perfect. But I think (if I am thinking correctly this was the woman on bike at night scenario) that a human wouldn't have had better odds at avoiding that in the dark. It is possible. But that late and that middle of no where, no one is really paying attention.
You seem to have forgotten that lawyers exist. You can bet good money that the first time someone is injured in even the most unlikely and freakish scenario, a lawyer will be in contact with the injured party offering to earn a fortune while clogging up the courts, which in the US means that Joe Numnuts and his barely achieved high school diploma will be sitting on a jury trying to decide what millions of lines of code should really say. The only reason we have not seen this already is that Uber settled out of court.
"I'm building to it"
That color video is going to happen some day... some day, maybe before the heat death of the universe, but it's coming!
i hope you know that picture of the chair with a face brought me immense joy
Like for the Celty reference btw. Of all the biker chicks in fiction, i name her Best Grill.
I'm blind in my right eye. Plasticity of the brain and the growth provided by life allows me to have excellent monocular depth perception. I have no problem driving, shooting hoops, riding my motorcycle at break neck speeds around twisties, or catching something someone tosses me (unless they have a bad arm). It should be (and one day will be) perfectly feasible to teach an AI to do what i do for fun with only one camera.
These self driving cars better not randomly veer into my lane when i'm passing... Hard to sue a car...
I worked on a self driving car project last year and the ethical dilemmas everyone is freaking out about are a non issue. The car is just programmed to plan to stop. That's it. No swerving, no ethical choices. Just apply brakes and stop. Let the ABS and ESP handle any traction loss from the brakes.
Gilotopia I also work with self driving cars myself the car just stops.
One Man On Fire yes I here it stops quite a bit, like every 5 miles the car tells the driver to take the wheel
Good to see people who know commenting on this kind of video. I am curious though, I get the impression that you guys work more on the technical side of things and in the example in the video where they are driving side by side I would be surprised if this was an issue.
Having said that I still think the idea is valid - what happens when something unforeseen means that some kind of accident is inevitable? From a technical standpoint "slam on the anchors" and minimise the damage is fine, my question would be, what happens when the lawyers get involved? You can bet that there will be at least one who throws a spanner in the works.
Are there legal representatives on development teams to look out for these issues? I would guess yes, but the world often doesn't work the way we think it does so I'm genuinely curious to find out.
Fernando, I was actually in an accident many years ago where my mental heuristic operated exactly like that. The end result: I did wreak into the parked cars that the a-hole wanted me to, but I took him out with me by literally driving down the shrinking opening I had until I hit both!
> Sometimes stopping isn't the best answer. Sometimes acceleration is the best answer.
The car doesn't have to pick a best answer, it has to pick a valid answer. And breaking is a valid answer in case of some unforseen weird thing happening.
"A problem that will never be solved though..." Is a really bold statement. I can imagine this has been said about all the technological advances ever made.
"If humans were meant to fly, God would have given us wings." Well that one was wrong also. In fact, God's creation of the birds, gave us some clues that heavier-than-air flight was actually possible, and how we might go about doing it.
@@yosefmacgruber1920 'gods creation'
7:50 - OMG that chair looks so happy! What a good chair!
If I had to sit in a car that doesn't even have my input on where it goes I would just take a bus. I would see a lot of people starting to ride motorcycles or bicycles once self driving cars are the norm. People will always want to have input on a vehicle. If I had to sit in a car that I paid for that I can't even control I would hate my life.
Cycles, people would never do that. People want a soft comfortable ride.
@@prospero4183 well people still want individualized transport. 2 wheeled vehicles are stored and parked very efficiently so I bet they will become popular again once roads become safer.
How long did it take to shift over from all horses to all cars with a few horses?
Exactly, automatic cars just have to be better than humans and they rule the roads
@@dylanchamberlain3375 They also have to be able to drive alongside human drivers.
I’ll make my own self driving car so I can turn off A.I mode and do things myself.
But then I’d be a problem.
So what’s the point of a self driving car in the first place?!
Yeah, If these cars are supposed to be so smart, then why can't they Co-exist with human drivers?
@@streamtrollmike5348 it's actually mostly that humans will make dumb decisions and good self driving will just be safer. By driving, you're risking other people's lives
@@streamtrollmike5348 That's a key point -- can self driving cars follow not only the written rules, but the unwritten rules of the road.
For example, will they take 10 minutes to pass a semi, because they're both going almost exactly the speed limit? Or will it speed up to get around, so that traffic can flow in the left lane?
I once drove on a dirt road with the lane departure correction turened on in the brand new work car for like 100m. It randomly decided there was a line and started steering me across the road whilst the traction control was taking drive off my wheels for every corrugation i hit. It was like driving on glass, whilst fighting some pirate for the wheel. Pulled over as soon as I could and turned all that nonsense off. NEVER AGAiN.
I thought you seemed a cool kind of guy anyway, but then you throw in that Little Boots caraoke at the end! You, sir, are a legend.
Who gets blamed if it runs over somebody?
the developer
Who gets blamed if a rock rolls over your car on the highway?
@@waltermeerschaert God