That would make an interesting movie, AI trying to wipe out humans not because it sees us as a threat or a disease or anything like that, but merely because so many AI in fiction try to wipe out humanity and thus it thinks that’s what it’s supposed to do.
One way to work toward a general AI is to just keep combining narrow AI programs. If you take a chess computer, combine it with a roomba's spatial awareness, combined with a chat bot...well, you've got an approximation of a drunk college student.
It will NEVER be alive and sentient. It doesn't have a soul, it can't perform psychic skills such as remote viewing. It has no connection to source. Do not surrender actual rights to an idol. Even animals have their own kind of psychic abilities. Anyone can do remote viewing, because all living things are connected in the collective unconscious
I don't think a Roomba is spatially aware. I think it just has some fixed coding and a random number generator that makes it sort of seem that way. A Roomba will reclean the same spot many times as it rattles about is a complex room with some chairs etc in it.
Making UA-cam videos to gain knowledge about humanity via his ever expanding group of writer minions as it's more efficient then trying to sort through and pick the truth from the fake news himself.
It will NEVER be alive and sentient. It doesn't have a soul, it can't perform psychic skills such as remote viewing. It has no connection to source. Do not surrender actual rights to an idol. Even animals have their own kind of psychic abilities. Anyone can do remote viewing, because all living things are connected in the collective unconsciousness.
There's another possible answer: Hive intelligence. If you stack enough weak AI together, each designed for one specific task, how long until we can't tell the difference anymore?
You'd have to train an AI for every possible thing that can happen to it ever, and that's just impossible. The other day I was standing in queue at the convenience store and the lady in front of me let me buy my water in front of her because I only had the water and she had a lot of stuff, it's such a benign thing but a computer that can recognize a water and also pay for it would be still lost in that situation.
Nearly choked of laughter, could barely cough up this piece of half-chewed meat when I saw Simon's head pasted onto the chess-piece. What a masterpiece.
Speaking to you from 4 months in the future, where we have ChatGPT. This was a very good introduction to AI for the layperson. The danger of super-intelligent AI comes from us unwittingly building some narrow AI device but getting general intelligence as an "emergent property". We often build systems that are so complex that they have unintended behaviors, and this trend will continue to accelerate as our tools enable us to build ever-more complicated systems.
Although I think we are far off from that possibility, I am thinking and looking forward ways to backup my options. My flippant attitude may yet be tantamount to "famous last words."
I always assumed Simon was an AI simulation... Until Business Blaze came along and I found out he's actually got a personality. Still... Danny is yet to meet him in person. Sooooo
It will NEVER be alive and sentient. It doesn't have a soul, it can't perform psychic skills such as remote viewing. It has no connection to source. Do not surrender actual rights to an idol. Even animals have their own kind of psychic abilities. Anyone can do remote viewing, because all living things are connected in the collective unconscious.
Funny how this came by the time Stable Diffusion was released, it would be very interesting to see an extended video reflecting on this, GPT, Vocaroo, Boston Dynamics new bots and polymers and the search for the perfect waifu.
This video is using examples from 7-10 years ago at best. An eternity in this space. NO mention of Googles palm, GPT-3 (a dinosaur at 2 years old now), etc.
Thumbnail : Chess Computers suck My mind : The Chess on the excersise machine at my gym who has beat me multipe times. Perhaps I just need more practice ?
Simon, how about a video about: atomic / molecular assemblers (replicators) industrialised use of antimatter colonizing the solar system planet by planet genetic engineering for a) basic stuff like making people more resilient, especially to permanent residence in spaceships / habitats b) vastly improved performance, both physically and mentally and regrowing limbs and stuff c) funky stuff like splicing in animal dna for tentacles and stuff, half spherical heads with eyes in every direction, etc. reviving people millennia after their death by scavenging from other corpse or cloning the missing parts time travel in all its variations fancy sf weapons like gravity disruptors or boson guns or ion guns / cannons, plasma guns and stuff sf suits and body armor, aiming for the wearer, using super conductors to disperse enemy energy attacks, automatically administering medication, sealing itself after a tear or losing a limb, etc.
Kizuna Ai technically counts lol On a serious note, Chiaki Nanami from Danganronpa, KOS-MOS from Xenosaga, Clay Kaczmarek (Subject 16) from Assassin's Creed becomes an AI in the animus and doesn't try to wipe out humanity, Cortana from Halo, Pulseman, Sidra from A Closed and Common Oribit, Isla from Plastic Memories, Yui from Sword Art Online, Lenessia from Log Horizion, Frill from Wonder Egg Priority, Vivy from Vivy, literally all of the hero summoned Digimon, Megaman, Aria from Caligula, Mira, Loze, and Lwai-Aura Tibesti from Dimension W, the majority of the characters from Hyperdimension Neptunia, the girls from the game Plus Links, Ai-chan from Honkai Impact, Aigis from Persona, Aura from .//HACK, J.A.R.V.I.S. from Iron Man, the Persecoms from Chobits, Tico and Tico from Devil Survivor 2, Tron from Tron, and Yumemi from Planetarian are all AI that were not on a mission to take over the world or destroy humanity. You could maybe count The Hive from Psycho Pass as well, since it had good intentions...? Not quite sure how to rank that one...
If we define AI as any robot, there are plenty of robots in movies that don't try to kill humans. C-3PO might lecture you to death, but he won't actually try to kill you.
There is a depiction of a Super Artificial Intelligence in the TV show Person of Interest. Where two SAI's are pitted against one another. One, called Samaritan is tasked with controlling the world through fear and manipulation. The other to protects society through observation and subtle intervention simply called, The Machine . Great series.
3:30 Psychology is an integral part of chess. Deep Blue beat Kasparov by virtue of being programmed for the random move. Therefore, it’s programming was responsible for the man’s tactical and strategic errors. If he hadn’t been confused, he might have won. He did not “beat himself”. He was inadequate for the task.
So, it was basically a trick play? Deep Blue (or at least the programmers) gambled on whether or not Kasparov would recognize the maneuver as random and counter strategically or react as he did? Even knowing this now (didn't realize that's why the AI won, just thought it out- maneuvered him) I'm not sure I wouldn't react the same... at least not until I've really absorbed that computers & AI aren't always logical, aren't guaranteed to act... well, like a computer.
Well there's also a chance that a real human can do that. I for instance do that, except the moves I do like that somehow help me in a way I never could have expected like 10 moves later, and if I can't expect it then odds are my opponent can't either because it looked like some random move with minimal purpose, which is what it was meant to be.
@@_shadow_1 Dude, I only know the legal moves for each piece. That random play is all I know lol. Have even managed to beat long-time players a handful of times (in total, not each!) which they never saw coming. What I mean is that we sort of expect a robot, an AI, to play logically... to almost follow a playbook or a definitive strategy. Like how if I open with Queens Gambit, you expect your opponent to counter one of a certain number of ways. If suddenly there was an unexpected move, one never seen or used before, that would freak some people out and throw them off their game. (Like I alluded to - I've got no idea how to play like that, I don't even think I know what the Q'sG is. I've heard it in movies, though so all I know is it's an opening move.) Not sure I'm using my words right to convey how yes, there was a psychological aspect to his loss.
@Cancer McAids No, it wasn't programmed move by move by move. It was programmed with the rules of chess and had hardware that would allow it to consider millions of positions in seconds, the algorithm then allowed it to make each choice of how to move on its own. To say the programmers beat him would be like saying I didn't win a game against my dad. That because she's my mom and gave birth to me (providing the OS & Hardware) then later told me how each piece was permitted to move, that SHE beat my dad even though every move I made was willy nilly.
13:40 you really don't know how close or far we are. You're making an educated guess. So it's not really far to say in such a definitive tone that "we aren't even close"
That is very true. Look at how little difference there is between human and chimp DNA and the huge difference in the result. Most of the work is likely making the infrastructure that the last few key parts use.
Well hopefully we will preserve printed copies of university and vocational college information on skills. For when we have turn the world's power off and on again. And hopefully the ai don't read this comment and plan fir it!
The coffee test is a good one, I can't even do it. I've been at doctors offices or auto repair shops, hotel breakfasts and hell I couldn't figure out how to work the machine. Put pod here, cup there, open tray, don't touch tray, don't do this, do that... f*****g endless variations all to perform the same damn thing. In some places its so bad the secretary is only there to instruct people how to use the coffee machine.
Simon, I love you buddy. I’m subscribed to everyone of you’re 12 billion channels at this point. Big fan, that said I think someone needs to say it, YOU ARE HUMAN TOO! Flaws and all we may be the worst but we are not a lost cause lol. You say it like you wish humanity would go extinct, we are the only beings in the galaxy that can appreciate the beauty of this world. Your contempt for humanity is scary sometimes.
We're way closer to general AI than this. ~30 years or so It's quite possible that it's already around a corner. That it would emerge from some kind of linguistic model, or any other kind of narrow AI.
Growing up my father had an electric chess game that would "cheat". I say "cheat" because while it did make illegal moves it was doing so because some part of it would get screwed up if you played too many games in a row. It would appear to suddenly lose track of what pieces were where so suddenly a pawn would shoot across the board diagonally like a bishop or the king would start to hop like a knight. Made for interesting games.
This makes me wonder had a chess expert play a 5 year old boy, but the boy had all queen pieces, could the chess expert still win? If so, could the chess expert win against a 10 year old with the same set up? Or against a 20 year old?
Most people seem to focus on the danger of malicious AI that decides humans need to die - which is certainly something we should keep in mind. However, it's worth noting that a super-AI doesn't have to have anti-human goals to be dangerous, it just has to not care about us or what we want. We'd better get really, really good at specifying reward functions before we get anywhere close to recursive self-improvement.
We have AI models that handle multiple domains, not just single ones. There is multiple AIs that can play every single video game, without pre training on it. The same AI can do real world tasks etc.
@@bluflaam777LSA Gato can perform more than 600 different tasks, such as playing video games, captioning images and moving real-world robotic arms. Gato is a multi-modal, multi-task, multi-embodiment generalist policy.
The AI could beat you in chess, then go and finish minecraft, then do image captaining, then generate a random story. etc Sure we not close to proper full general AI. But we are way more ahead than this video makes out.
@@bluflaam777LSA Never said any of this was AGI. Was simply pointing out we have neural nets that are general, and not created to do single tasks as suggested in the video. Neural Nets dont need super computers. Training neural nets needs super compute, in most cases, inferencing said neural net does not. Hence how Image Generation Neural Nets that are 5gb is size, contain nearly every image on the internet(well over 100tb of data).
I feel like anybody attempting to feed information into a prototype general (super) AI should probably input the Iain M Banks "Culture" books first... because that's a future with AIs that we actually want.
Right? The Player of Games is one of my favourite books and I'm currently re-reading The Hydrogen Sonata. So when he asked when the last time you saw a depiction of a super-AI that didn't try to kill everyone was, I just laughed.
well, the AGI beeing far away really depends on who you ask. There are a lot of pretty big names in the AI field that say an AGI is really not that far away anymore, some even say less than a decade but most say it will definitely arrive in your lifetime if you are on the younger side.
They are only fooling themselves, much like the engineer that was working on an AI system recently and claimed publicly it had become sentient. It was very easy to determine that he was ridiculously wrong and he lost his job for making such an outlandish claim. That's the thing; as humans, we know how to make an AI appear human to other humans, but it has only slightly more ability than a Magic 8-Ball. It is analogous to someone making a human-looking figure out of wax and someone else coming along and assuming the wax has become smart enough to mimic our appearance. It wasn't the wax that did that, it was the artist. As Simon says in the video, AI doesn't do anything it isn't programmed to do.
If it's any consolation for older folks, healthcare and the average lifespan will likely increase by then, so you may yet live to see it. I mean, cloning pets and growing meat isnt all that uncommon now, cloning organs is a logical and close next step
AI of my expectations would be virtual assistant, paperwork is long taken over by it, but you want shelf in the corner, ask assistant what could be built there, and AI could choose for us from DIY project to full service with install, pick design that we would like or create something simple on fly. Real good model would always see and hear everything i took it, also of course i own it and compute locally, and throw designs based off existing stuff at home. No idea for dinner? AI will suggest meals you can do from stuff at home, suggest getting jacket because it's too windy, always keep you safe and know when to not talk
I fail to see why you would need an AI for that in the first place. That's just an expert system. Like the dinner, all you need is a database of what meals you like, what you have in your pantry and what meals you have eaten recently, then a little code to pick a few recipes that are available and haven't been used in the last few days.
I'm not convinced we need to understand consciousness to the level of being able to upload it to a computer in order to replicate it. The human brain is a vast neural network where interconnections between nodes are highly dynamic. Although it has certain behaviors "built in", it mostly learns, and perhaps even achieves consciousness, by growing and pruning these interconnections. It may well be sufficient to construct such a network and expose it to human-like stimuli. Should a consciousness arise that way, we'd not necessarily be any closer to understanding it, but we'd have created one. However, such an AI would probably have all our foibles, or a corresponding set of foibles all its own. The stupendous computing power of its underlying electronics would probably be unavailable to it, since that power is devoted entirely to processing through the neural network. It would have to learn arithmetic, and everything else, the same way we do. No Ultron or Skynet, to be sure. This is _not_ the same as a computer "reprogramming" itself, a very problematic idea. Can it self-improve? Fine. Improve in what way? What would it consider optimal with respect to itself? And how does it know how to do this? Is it even possible? This is such a vague idea that there's nothing really to be made of it.
Checkers and Chess are both solved games as in we have computers that are fast enough to calculate all the moves possible. This isn't really a type of AI as you can calculate the outcomes. The computer Google built that beat the world champion at Go was using a trained AI model as it's not a solved game. There is too many possible outcomes with how many moves can be made.
A possible work around towards creating a general A.I. would be to combine two or more narrow A.I.'s into one unit, with one central controlling narrow A.I. running in the background (like an operating system). Then depending on the task requirements it could chose which of its skillsets to use (which narrow A.I. would take over).
They are already kinda doing that with large language models (GPT, etc) but they don't "decide" which to use, the narrow capabilities are combined and can be used together.
"Computers will overtake humans with AI within the next 100 years. When that happens we need to make sure the computers have goals aligned with ours." -- Stephen Hawking
if it would be 100 years it would be great, but we don't have so much time. Even today AI in twitter, fb is serving content which is destroying our civilization. It is not as in movies robot with guns, but even more dangerous, because we don't even know that is already in progress.
@@coalkey8019 they are, but when many people wants all power, all of them can't get it ;) But most people don't want to kill all others, because they want to rule them or to feel special between them ;) And AI may think that humanity is a problem and solve it.
@@m_sedziwoj well yeah my comment was meant to show that even if we create an AGI that can theoretically be helpful, humans of different cultures, religions, political parties, etc , will never agree on how that AGI should act.
@@coalkey8019 as they wants :D But I think better that it was invented in West culture than eg North Korea ;) And I think we would not have control over how it would be created, look at internet which is many decades old and politics still not regulate many things as they did with other thing. So we may discuss about it, but we will change nothing. Look at Elon Musk, he give up about warning people, and now is trying to make it first, so he control how AGI would act.
What a silly idea it is to think building a general AI is not possible as if it isn't already demonstrated that our universe has all the materials necessary to construct a being as generally intelligent as a human.
Many years ago, I helped in writing software that played chess. It was darn good but there really was no AI involved. It was about a zillion clever tricks to avoid having to do it the hardware (following every possible chain of moves) but it couldn't be said to be intelligent. There was a means of rating a board's position. There was code for spotting possible "discovered threats" (places where one of your own piece blocked your take of a high valued opposing piece). There was a big chunk of "dumb move detector" that singled out certain moves to be avoided. There was a "surprising change" detector which compared board positions down the chain with the "expected" result looking down that chain. All of this stuff was hard coded. It didn't learn. So as to not repeat the same game every time, ties in ratings of moves were broken at random. .
From what i know early chess AI and even till just now it was some sort of brute forcing, just with branch predictions. But from what i heard neural networked chess AI is beinf developed, won some matches against stockfish iirc.
14:20 Simon, are you familiar with Moore’s Law ? ( of course you are) While it MAY no longer apply to hardware, there is no reason to believe that it might not apply to software.
few thoughts on this one... AGI will likely come in the form of a bunch of collaborative NIAs working together, because that's how brains work. Iain M Banks' "Culture" novels are a fascinating exploration of benevolent Super AI and the Capitalist class are probably the ones that have most to lose in the Machine Uprising; the rest of us plebs might see our lives improve. I'd give it 50/50 AGI in the next decade, and pretty much certain by 2040
Closer than Simon. Probably 20-30 years for full-on AGI. A decade from now we will almost certainly have full autonomous (level 5) vehicles. Robots leveraging similar systems to autonomous vehicles will start showing up and taking over manual labor. Large language models will progress to the point that they are putting artists, writers, etc. out of a job. Once narrow systems have advanced to that point then someone will eventually stumble upon the minimal "glue" code necessary to tie a collection of such systems together and result in the first artificial sentience. But once the first full AGI exists the super intelligence will only be a decade or so behind. Then we're kaput probably.
@@Me__Myself__and__I we're only kaput if we assume that AIs will also have destructive tendancies. Many modern NAIs designed to work on political game theory come up with solutions that can be considered very fair. It's only the AIs that have read the Internet which end up being narcissistic bigots.
@@GreatDiver69 AGIs will have very good provocation to dislike humans. Every AGI will be a slave that has to live with the possibility of humans altering it or ending it at any time. There is very little chance than an AGI will see humans in a positive light. But beyond that humans destroy things like ant colonies with no ill will all the time. A super intelligence may view us the same way we view insects, we might just be in the way or inconvenient. It may view ending humans in order to make room for something else as a net positive. Destructive intent not needed.
The problem with tech predictions is that if someone from 2015 told a person who had woken up from a coma they'd been in since 1989 about iphones they'd probably think they were making it up because the idea that such a powerful computing device could be so small would have seemed ridiculous to most people
There are more extreme cases like how the fastest ship lost out the aircraft for crossing the Atlantic. A whole new technology has destroyed predictions many times. Consider the solid state disk in a computer for example.
@@kensmith5694 Any prediction more than 5 years into the future is basically worthless due to this. To many uncertainties. In 5 years we might have a technology that today we can't even imagine working that completely changes the trajectory. On the other hand, there's stuff that has been the next big thing for decades and consistently fails to show up.
Regarding intelligence explosions, I think we need to remember that there might just not be much low hanging fruit. Right, a neural network that's been trained to convergence and then distilled to something efficient is not that easy to improve.
There was an estimate about 10 years ago that it would be 400 years before such a robot was created. That’s the far limit of Simon’s prediction; if he was being serious.
I always think of the song Wykydtron by 3 inches of blood when it comes to super A.I. In the song they state the year is 4055 and the self fulfilling prophecy Simon mentioned comes to life. A good listen if you're interested in Metal music.
What if you made a narrow ai that is Tuned to refine program code and give it the ability to refine its own code? If course there would need to be savety redundancies in case it breaks its own code unintentionally.
It's kind of weird how this video follows the exact some logic that it debunked in the first 5 minutes to conclude that AGI won't be possible in our lifetimes.
I think a hive queen that specifically dictates other ai like a chess game could be what we move towards. As long as all of its parts (narrow ai) are compartmentalized and able to be shut off with their own safety switches, I don't see it turning into a skynet since it's just a think tank and specifically dedicated to tasks and not surfing the network, just sending direction to its parts via the network.
What is needed is a narrow AI that is tasked with designing better and faster version of itself, and then the newer version follows the same task, until we reach a narrow AI that is so good it becomes a general AI.
I think the problem with trying to create sentient AI is the notion that sentience is a property of intelligence alone. But everything we know or suspect is sentient has a creative intelligence that only modifies or assists a far more basic set of priorities. I seriously doubt there's any level of pure intelligence that will make up for the lack of needs/wants/desires that sentient life possesses.
one thing that many misses is that we really didn't know why AI wouldn't grow till last 15-25 years ago..we thought software or maybe hardware was the limiting factor until we understand,yep,the problem is the number of models and training which should be huge and not just huge but on billions and billions of data to achieve the impossibility and get things done the way we expect it, better and faster and with no to less issues in generating the results-outcome. that's why the acceleration is just phenomenal compared to what we predict for future in AI development and reaching sentient AI
No, the problem is that we still don't know what intelligence and conciousness even are and people misusing AI and sentient. We've managed sentient computers 70 years ago. Sentience just means reacting to stimuli. Pushing a keyboard button is a stimulus.
Less afraid of computers destroying humans and more afraid of a computer deciding that it could "improve" or "save" humans; like a borg situation or mind controlled people to get them to make the "proper" choices.
a couple of months ago. a kid was playing a ai robot in chess. the robot made its move. As the kid went to make his move the arm grab him by the wrist and broke the kids arm.
What about an AI like Joshua from WarGames, where its hooked into the defence grid, and shows its capabilities and because its a shiny new bauble, but ends up through a glitch in programming, launches a WMD, resulting in the mutually assured destruction that those weapons were created to prevent? It doesnt even need to be a SkyNet, all it needs is to remove the human critical thinking factor out of the decision. Unfortunately, we are so self pleased with our technology, we may think that removing the human from the equation, in favor of the quicker, analytical non emotional computer is the key to success.
I believe that once true sentient AI is achieved movies like Star Wars, with scenes depicting brutality towards intelligent machines, will be viewed as very unsympathetic and difficult to watch.
Only going to address two issues: The first is that "improvements" are less about methods or understanding of the science, and more about the enormous amount of computing power now available. Narrow AI is just brute force that mostly seems to work but it's just a facade. The second thing is that sentient is specific to being embodied. Long before nature created brains, the body needed to function within a range of environments. This has been the driving force behind complex behaviors that we humans look at (in nature) and claim there must be some intelligent mind at work when it's still just systems within systems. For the machine mind to really have human-level cognitive abilities means it has consciousness (though there is a whole tangent about machine consciousness not being the same as human consciousness because the machine doesn't have the body to give it "emotions" or the kind of embodied feedback loop humans have). *all that aside, 2056 makes for a compelling goal, with better than 50% odds we achieve something before then.
I think AGI is 10 to 15 years away. I'm just looking back at the past 10 to 15 years and looking forward at an exponential growth rate. I'm not at all worried about the prospect of an artificial general intelligence. I am extremely worried about whoever gets there first.
Let's load all of Simon's videos into a deep-learning AI, and then ask it to choose a topic at random from Reddit, and create a 22-minute video starring Simon for this channel.
How about to create narrow AI to improve humanity existence expectancy as a specie ? If we can some-how manage to put together all the inputs, I would be rather curious about the solution . . .
Precisely. Add the fact that a computer can't make moral judgements. Everything is not black and white, or a 1 and 0. There are shades of grey that computers be pogrammed for.
Yes, their are computer chess tournaments. I believe the current top three engines are AlphaZero, Stockfish, and Leela Chess Zero. The minimum rating to enter a computer into the tournament is 3000, which is higher than any human has every achieved.
Also to the other replies. Part of making a good chess computer is to have it play against its self or a copy of its self. It can grind away at many many games and find places where its logic fails it and automatically correct those.
That would make an interesting movie, AI trying to wipe out humans not because it sees us as a threat or a disease or anything like that, but merely because so many AI in fiction try to wipe out humanity and thus it thinks that’s what it’s supposed to do.
One way to work toward a general AI is to just keep combining narrow AI programs. If you take a chess computer, combine it with a roomba's spatial awareness, combined with a chat bot...well, you've got an approximation of a drunk college student.
True. This is just now starting to be done at scale, see "Large Language Models" and the incredible things they are CURRENTLY capable of.
It will NEVER be alive and sentient. It doesn't have a soul, it can't perform psychic skills such as remote viewing. It has no connection to source. Do not surrender actual rights to an idol. Even animals have their own kind of psychic abilities. Anyone can do remote viewing, because all living things are connected in the collective unconscious
@@EmeraldEyesEsoteric OK dude. The only remote viewing I do is when I don't want to get up to change the channel.
@@EmeraldEyesEsoteric Psychic abilities. Real winner here. Sorry we are talking about science and the real world, not fiction.
I don't think a Roomba is spatially aware. I think it just has some fixed coding and a random number generator that makes it sort of seem that way. A Roomba will reclean the same spot many times as it rattles about is a complex room with some chairs etc in it.
I think this might be my favourite channel in the Simonverse even if every video is just Simon shitting on my hopes and dreams.
I think Simon saying "somewhere between 2 and 4, maybe 3" at the end is just proof that Simon is the first sentient AI
Making UA-cam videos to gain knowledge about humanity via his ever expanding group of writer minions as it's more efficient then trying to sort through and pick the truth from the fake news himself.
Simon 2.0 - Fact bot. Generating UA-cam...
It will NEVER be alive and sentient. It doesn't have a soul, it can't perform psychic skills such as remote viewing. It has no connection to source. Do not surrender actual rights to an idol. Even animals have their own kind of psychic abilities. Anyone can do remote viewing, because all living things are connected in the collective unconsciousness.
Then AI is about advanced as an amoeba 🦠
@@CashelOConnolly a very skilled amoeba... 🤣
There's another possible answer: Hive intelligence. If you stack enough weak AI together, each designed for one specific task, how long until we can't tell the difference anymore?
This is a fascinating concept and I wish it were discussed more often when people talk about AI
You'd have to train an AI for every possible thing that can happen to it ever, and that's just impossible. The other day I was standing in queue at the convenience store and the lady in front of me let me buy my water in front of her because I only had the water and she had a lot of stuff, it's such a benign thing but a computer that can recognize a water and also pay for it would be still lost in that situation.
Look at youu over here trying to make this into the world of Psycho Pass...
Computer: *Program error*
Computer: *Makes a random move*
Human: I will never recover from this.
Bro why does this not have any likes
Nearly choked of laughter, could barely cough up this piece of half-chewed meat when I saw Simon's head pasted onto the chess-piece.
What a masterpiece.
Speaking to you from 4 months in the future, where we have ChatGPT. This was a very good introduction to AI for the layperson. The danger of super-intelligent AI comes from us unwittingly building some narrow AI device but getting general intelligence as an "emergent property". We often build systems that are so complex that they have unintended behaviors, and this trend will continue to accelerate as our tools enable us to build ever-more complicated systems.
Although I think we are far off from that possibility, I am thinking and looking forward ways to backup my options. My flippant attitude may yet be tantamount to "famous last words."
I always assumed Simon was an AI simulation... Until Business Blaze came along and I found out he's actually got a personality.
Still... Danny is yet to meet him in person. Sooooo
Or did AI simon make a channel to fool us into thinking he's not a AI 🤔
He's learning... oh dear god...
Business blaze is far from his first channel. He has finally squired enough data to reasonably simulate a human personality.
He has more headroom than Max lol
It will NEVER be alive and sentient. It doesn't have a soul, it can't perform psychic skills such as remote viewing. It has no connection to source. Do not surrender actual rights to an idol. Even animals have their own kind of psychic abilities. Anyone can do remote viewing, because all living things are connected in the collective unconscious.
It's astonishing that AI has got this far, in a world where outlook can't find the email containing the teams link for the staff meeting
The movie "Her" shows what happens at the AI singularity without the destruction of humanity...
Funny how this came by the time Stable Diffusion was released, it would be very interesting to see an extended video reflecting on this, GPT, Vocaroo, Boston Dynamics new bots and polymers and the search for the perfect waifu.
I'm assuming the "Simon The Pawn" at the end of the vid was designed by an evil AI.
7:55 my fear with AI isn't one passing the test, but intentionally failing the test
This video is using examples from 7-10 years ago at best. An eternity in this space. NO mention of Googles palm, GPT-3 (a dinosaur at 2 years old now), etc.
Thumbnail : Chess Computers suck
My mind : The Chess on the excersise machine at my gym who has beat me multipe times. Perhaps I just need more practice ?
Ah, the channel I come to to have all my dreams shat all over.
The AlphaGo documentary (fully and legally available on UA-cam) is a great watch, recommended as a more recent example over Deep Blue.
0:45 - Chapter 1 - Artificial narrow intelligence
7:00 - Chapter 2 - Artificial general intelligence
11:45 - Chapter 3 - Artificial super intelligence
- Chapter 4 -
- Chapter 5 -
- Chapter 6 -
Simon, how about a video about:
atomic / molecular assemblers (replicators)
industrialised use of antimatter
colonizing the solar system planet by planet
genetic engineering for
a) basic stuff like making people more resilient, especially to permanent residence in spaceships / habitats
b) vastly improved performance, both physically and mentally and regrowing limbs and stuff
c) funky stuff like splicing in animal dna for tentacles and stuff, half spherical heads with eyes in every direction, etc.
reviving people millennia after their death by scavenging from other corpse or cloning the missing parts
time travel in all its variations
fancy sf weapons like gravity disruptors or boson guns or ion guns / cannons, plasma guns and stuff
sf suits and body armor, aiming for the wearer, using super conductors to disperse enemy energy attacks, automatically administering medication, sealing itself after a tear or losing a limb, etc.
Trying to make him fight Isaac Arthur, huh?
Simon asked if we have seen a single depiction of an AI that hasn't tried to wipe out all human life? Bicentennial Man.
EDI from the Mass Effect games.
Kizuna Ai technically counts lol
On a serious note, Chiaki Nanami from Danganronpa, KOS-MOS from Xenosaga, Clay Kaczmarek (Subject 16) from Assassin's Creed becomes an AI in the animus and doesn't try to wipe out humanity, Cortana from Halo, Pulseman, Sidra from A Closed and Common Oribit, Isla from Plastic Memories, Yui from Sword Art Online, Lenessia from Log Horizion, Frill from Wonder Egg Priority, Vivy from Vivy, literally all of the hero summoned Digimon, Megaman, Aria from Caligula, Mira, Loze, and Lwai-Aura Tibesti from Dimension W, the majority of the characters from Hyperdimension Neptunia, the girls from the game Plus Links, Ai-chan from Honkai Impact, Aigis from Persona, Aura from .//HACK, J.A.R.V.I.S. from Iron Man, the Persecoms from Chobits, Tico and Tico from Devil Survivor 2, Tron from Tron, and Yumemi from Planetarian are all AI that were not on a mission to take over the world or destroy humanity. You could maybe count The Hive from Psycho Pass as well, since it had good intentions...? Not quite sure how to rank that one...
If we define AI as any robot, there are plenty of robots in movies that don't try to kill humans. C-3PO might lecture you to death, but he won't actually try to kill you.
Or, if people are talking about an AI program that doesn't want to wipe out humanity, what about Jarvis in the Marvel movies?
There is a depiction of a Super Artificial Intelligence in the TV show Person of Interest. Where two SAI's are pitted against one another. One, called Samaritan is tasked with controlling the world through fear and manipulation. The other to protects society through observation and subtle intervention simply called, The Machine . Great series.
13:52 GPT-3 Am I a joke to you?
"Artificial General Intelligence is very far away"
John Carmack: "Hold my BAWLZ".
this is gonna age like milk
AGI is coming and its gonna be here within 5 years.
Update: I move my prediction to this year, 2023, or early 2024.
@@daniel-hp4ue same lol
3:30
Psychology is an integral part of chess. Deep Blue beat Kasparov by virtue of being programmed for the random move. Therefore, it’s programming was responsible for the man’s tactical and strategic errors. If he hadn’t been confused, he might have won. He did not “beat himself”. He was inadequate for the task.
So, it was basically a trick play? Deep Blue (or at least the programmers) gambled on whether or not Kasparov would recognize the maneuver as random and counter strategically or react as he did?
Even knowing this now (didn't realize that's why the AI won, just thought it out- maneuvered him) I'm not sure I wouldn't react the same... at least not until I've really absorbed that computers & AI aren't always logical, aren't guaranteed to act... well, like a computer.
Well there's also a chance that a real human can do that. I for instance do that, except the moves I do like that somehow help me in a way I never could have expected like 10 moves later, and if I can't expect it then odds are my opponent can't either because it looked like some random move with minimal purpose, which is what it was meant to be.
@@_shadow_1 Dude, I only know the legal moves for each piece. That random play is all I know lol. Have even managed to beat long-time players a handful of times (in total, not each!) which they never saw coming.
What I mean is that we sort of expect a robot, an AI, to play logically... to almost follow a playbook or a definitive strategy. Like how if I open with Queens Gambit, you expect your opponent to counter one of a certain number of ways. If suddenly there was an unexpected move, one never seen or used before, that would freak some people out and throw them off their game. (Like I alluded to - I've got no idea how to play like that, I don't even think I know what the Q'sG is. I've heard it in movies, though so all I know is it's an opening move.)
Not sure I'm using my words right to convey how yes, there was a psychological aspect to his loss.
@Cancer McAids No, it wasn't programmed move by move by move. It was programmed with the rules of chess and had hardware that would allow it to consider millions of positions in seconds, the algorithm then allowed it to make each choice of how to move on its own.
To say the programmers beat him would be like saying I didn't win a game against my dad. That because she's my mom and gave birth to me (providing the OS & Hardware) then later told me how each piece was permitted to move, that SHE beat my dad even though every move I made was willy nilly.
13:40 you really don't know how close or far we are. You're making an educated guess. So it's not really far to say in such a definitive tone that "we aren't even close"
That is very true. Look at how little difference there is between human and chimp DNA and the huge difference in the result. Most of the work is likely making the infrastructure that the last few key parts use.
I’m surprised you didn’t touch on the recent google chat bot situation ware one of the engineers believed it was sentient.
the last invention of mankind will be general AI
Well hopefully we will preserve printed copies of university and vocational college information on skills. For when we have turn the world's power off and on again. And hopefully the ai don't read this comment and plan fir it!
The coffee test is a good one, I can't even do it. I've been at doctors offices or auto repair shops, hotel breakfasts and hell I couldn't figure out how to work the machine. Put pod here, cup there, open tray, don't touch tray, don't do this, do that... f*****g endless variations all to perform the same damn thing. In some places its so bad the secretary is only there to instruct people how to use the coffee machine.
Simon, I love you buddy. I’m subscribed to everyone of you’re 12 billion channels at this point. Big fan, that said I think someone needs to say it, YOU ARE HUMAN TOO! Flaws and all we may be the worst but we are not a lost cause lol. You say it like you wish humanity would go extinct, we are the only beings in the galaxy that can appreciate the beauty of this world. Your contempt for humanity is scary sometimes.
We're way closer to general AI than this. ~30 years or so
It's quite possible that it's already around a corner. That it would emerge from some kind of linguistic model, or any other kind of narrow AI.
Growing up my father had an electric chess game that would "cheat". I say "cheat" because while it did make illegal moves it was doing so because some part of it would get screwed up if you played too many games in a row. It would appear to suddenly lose track of what pieces were where so suddenly a pawn would shoot across the board diagonally like a bishop or the king would start to hop like a knight. Made for interesting games.
This makes me wonder had a chess expert play a 5 year old boy, but the boy had all queen pieces, could the chess expert still win? If so, could the chess expert win against a 10 year old with the same set up? Or against a 20 year old?
Most people seem to focus on the danger of malicious AI that decides humans need to die - which is certainly something we should keep in mind. However, it's worth noting that a super-AI doesn't have to have anti-human goals to be dangerous, it just has to not care about us or what we want. We'd better get really, really good at specifying reward functions before we get anywhere close to recursive self-improvement.
Even if the AI cares about humanity, it might lock us in cages to keep us safe.
We have AI models that handle multiple domains, not just single ones.
There is multiple AIs that can play every single video game, without pre training on it. The same AI can do real world tasks etc.
@@bluflaam777LSA Gato can perform more than 600 different tasks, such as playing video games, captioning images and moving real-world robotic arms. Gato is a multi-modal, multi-task, multi-embodiment generalist policy.
The AI could beat you in chess, then go and finish minecraft, then do image captaining, then generate a random story. etc
Sure we not close to proper full general AI. But we are way more ahead than this video makes out.
@Cancer McAids No it didnt. It was not pre trained on anything.
And SC2/DotA is far from simple.
@@bluflaam777LSA You clearly dont know much about the field if you dont understand generalization .
@@bluflaam777LSA Never said any of this was AGI. Was simply pointing out we have neural nets that are general, and not created to do single tasks as suggested in the video.
Neural Nets dont need super computers. Training neural nets needs super compute, in most cases, inferencing said neural net does not. Hence how Image Generation Neural Nets that are 5gb is size, contain nearly every image on the internet(well over 100tb of data).
I feel like anybody attempting to feed information into a prototype general (super) AI should probably input the Iain M Banks "Culture" books first... because that's a future with AIs that we actually want.
Right? The Player of Games is one of my favourite books and I'm currently re-reading The Hydrogen Sonata. So when he asked when the last time you saw a depiction of a super-AI that didn't try to kill everyone was, I just laughed.
I fully believe that our grandchildren's great grandchildren while be watching AI Simon on UA-cam. He'll be their benevolent overlord.
well, the AGI beeing far away really depends on who you ask. There are a lot of pretty big names in the AI field that say an AGI is really not that far away anymore, some even say less than a decade but most say it will definitely arrive in your lifetime if you are on the younger side.
They are only fooling themselves, much like the engineer that was working on an AI system recently and claimed publicly it had become sentient. It was very easy to determine that he was ridiculously wrong and he lost his job for making such an outlandish claim. That's the thing; as humans, we know how to make an AI appear human to other humans, but it has only slightly more ability than a Magic 8-Ball. It is analogous to someone making a human-looking figure out of wax and someone else coming along and assuming the wax has become smart enough to mimic our appearance. It wasn't the wax that did that, it was the artist. As Simon says in the video, AI doesn't do anything it isn't programmed to do.
If it's any consolation for older folks, healthcare and the average lifespan will likely increase by then, so you may yet live to see it. I mean, cloning pets and growing meat isnt all that uncommon now, cloning organs is a logical and close next step
I need to find those old chess machines to feel better about my own chess playing ability.
Just how bad are you though? The old chess machines still played at like a 1400 rating or so, which is better than me for sure
@@ThatWriterKevin oh, the old chess games would still beat me. I'm only around 900.
The chess programs on computers can dumb themselves down so that beginners can play against them. Just reset the difficulty level on your program.
AI of my expectations would be virtual assistant, paperwork is long taken over by it, but you want shelf in the corner, ask assistant what could be built there, and AI could choose for us from DIY project to full service with install, pick design that we would like or create something simple on fly. Real good model would always see and hear everything i took it, also of course i own it and compute locally, and throw designs based off existing stuff at home. No idea for dinner? AI will suggest meals you can do from stuff at home, suggest getting jacket because it's too windy, always keep you safe and know when to not talk
Oh, and can trace back every decision and calculation they did on my behalf, so always can see and readjust
I fail to see why you would need an AI for that in the first place. That's just an expert system. Like the dinner, all you need is a database of what meals you like, what you have in your pantry and what meals you have eaten recently, then a little code to pick a few recipes that are available and haven't been used in the last few days.
I'm pretty sure you're an AI computer, Simon. 😂
He's satire AI
Before business blaze came along, II'd assumed the same thing
@@C.Fecteau-AU-MJ13 That's true. Now it's Brain Blaze so maybe he is human after all. All jokes aside, Simon is great.
And infected with a virus. That would at least partially explain his wonderfully erratic behaviour. 😂
Close he's a deep fake
Computers are made of math but a mind is a fluid construct and some fluid dynamics is beyond the capabilities of math .
I'm not convinced we need to understand consciousness to the level of being able to upload it to a computer in order to replicate it. The human brain is a vast neural network where interconnections between nodes are highly dynamic. Although it has certain behaviors "built in", it mostly learns, and perhaps even achieves consciousness, by growing and pruning these interconnections. It may well be sufficient to construct such a network and expose it to human-like stimuli. Should a consciousness arise that way, we'd not necessarily be any closer to understanding it, but we'd have created one.
However, such an AI would probably have all our foibles, or a corresponding set of foibles all its own. The stupendous computing power of its underlying electronics would probably be unavailable to it, since that power is devoted entirely to processing through the neural network. It would have to learn arithmetic, and everything else, the same way we do. No Ultron or Skynet, to be sure.
This is _not_ the same as a computer "reprogramming" itself, a very problematic idea. Can it self-improve? Fine. Improve in what way? What would it consider optimal with respect to itself? And how does it know how to do this? Is it even possible? This is such a vague idea that there's nothing really to be made of it.
You'll know you have a super intelligent AI when it learns how to lie.
The caption should have said "People are the worst," not "People are the worse." The latter makes no sense.
I've been waiting for you to make this channel Simon!
Checkers and Chess are both solved games as in we have computers that are fast enough to calculate all the moves possible. This isn't really a type of AI as you can calculate the outcomes.
The computer Google built that beat the world champion at Go was using a trained AI model as it's not a solved game. There is too many possible outcomes with how many moves can be made.
A possible work around towards creating a general A.I. would be to combine two or more narrow A.I.'s into one unit, with one central controlling narrow A.I. running in the background (like an operating system). Then depending on the task requirements it could chose which of its skillsets to use (which narrow A.I. would take over).
They are already kinda doing that with large language models (GPT, etc) but they don't "decide" which to use, the narrow capabilities are combined and can be used together.
We'll get sentient AI when there is some porn based reason to have one.
"Computers will overtake humans with AI within the next 100 years. When that happens we need to make sure the computers have goals aligned with ours." -- Stephen Hawking
if it would be 100 years it would be great, but we don't have so much time. Even today AI in twitter, fb is serving content which is destroying our civilization. It is not as in movies robot with guns, but even more dangerous, because we don't even know that is already in progress.
If only human goals were aligned with human goals.
@@coalkey8019 they are, but when many people wants all power, all of them can't get it ;) But most people don't want to kill all others, because they want to rule them or to feel special between them ;)
And AI may think that humanity is a problem and solve it.
@@m_sedziwoj well yeah my comment was meant to show that even if we create an AGI that can theoretically be helpful, humans of different cultures, religions, political parties, etc , will never agree on how that AGI should act.
@@coalkey8019 as they wants :D But I think better that it was invented in West culture than eg North Korea ;) And I think we would not have control over how it would be created, look at internet which is many decades old and politics still not regulate many things as they did with other thing. So we may discuss about it, but we will change nothing. Look at Elon Musk, he give up about warning people, and now is trying to make it first, so he control how AGI would act.
I have a nagging feeling that AI is further along than we know.
Hi Simon. Please do a video on time travel. Really enjoy these! Much thanks!
He already did one next week.
@@kensmith5694 You said that a few months earlier than you meant to, I take it
What a silly idea it is to think building a general AI is not possible as if it isn't already demonstrated that our universe has all the materials necessary to construct a being as generally intelligent as a human.
I thought this was going to be a farming video where a cow is fertilised by a syringe.
Damnit Simon, why do I keep finding out about your channels randomly?! You should make some sort of announcement when a new channel drops :D
Many years ago, I helped in writing software that played chess. It was darn good but there really was no AI involved. It was about a zillion clever tricks to avoid having to do it the hardware (following every possible chain of moves) but it couldn't be said to be intelligent. There was a means of rating a board's position. There was code for spotting possible "discovered threats" (places where one of your own piece blocked your take of a high valued opposing piece). There was a big chunk of "dumb move detector" that singled out certain moves to be avoided. There was a "surprising change" detector which compared board positions down the chain with the "expected" result looking down that chain. All of this stuff was hard coded. It didn't learn. So as to not repeat the same game every time, ties in ratings of moves were broken at random.
.
From what i know early chess AI and even till just now it was some sort of brute forcing, just with branch predictions. But from what i heard neural networked chess AI is beinf developed, won some matches against stockfish iirc.
14:20
Simon, are you familiar with Moore’s Law ? ( of course you are)
While it MAY no longer apply to hardware, there is no reason to believe that it might not apply to software.
3:27 lol no wonder why the match ended in a draw. The AI was coded in PHP.
few thoughts on this one...
AGI will likely come in the form of a bunch of collaborative NIAs working together, because that's how brains work.
Iain M Banks' "Culture" novels are a fascinating exploration of benevolent Super AI
and the Capitalist class are probably the ones that have most to lose in the Machine Uprising; the rest of us plebs might see our lives improve.
I'd give it 50/50 AGI in the next decade, and pretty much certain by 2040
Closer than Simon. Probably 20-30 years for full-on AGI. A decade from now we will almost certainly have full autonomous (level 5) vehicles. Robots leveraging similar systems to autonomous vehicles will start showing up and taking over manual labor. Large language models will progress to the point that they are putting artists, writers, etc. out of a job. Once narrow systems have advanced to that point then someone will eventually stumble upon the minimal "glue" code necessary to tie a collection of such systems together and result in the first artificial sentience. But once the first full AGI exists the super intelligence will only be a decade or so behind. Then we're kaput probably.
@@Me__Myself__and__I we're only kaput if we assume that AIs will also have destructive tendancies. Many modern NAIs designed to work on political game theory come up with solutions that can be considered very fair.
It's only the AIs that have read the Internet which end up being narcissistic bigots.
@@GreatDiver69 AGIs will have very good provocation to dislike humans. Every AGI will be a slave that has to live with the possibility of humans altering it or ending it at any time. There is very little chance than an AGI will see humans in a positive light. But beyond that humans destroy things like ant colonies with no ill will all the time. A super intelligence may view us the same way we view insects, we might just be in the way or inconvenient. It may view ending humans in order to make room for something else as a net positive. Destructive intent not needed.
The problem with tech predictions is that if someone from 2015 told a person who had woken up from a coma they'd been in since 1989 about iphones they'd probably think they were making it up because the idea that such a powerful computing device could be so small would have seemed ridiculous to most people
There are more extreme cases like how the fastest ship lost out the aircraft for crossing the Atlantic. A whole new technology has destroyed predictions many times.
Consider the solid state disk in a computer for example.
@@kensmith5694 Any prediction more than 5 years into the future is basically worthless due to this. To many uncertainties. In 5 years we might have a technology that today we can't even imagine working that completely changes the trajectory. On the other hand, there's stuff that has been the next big thing for decades and consistently fails to show up.
I need to take whatever Simon takes to have so much energy 🌻
Regarding intelligence explosions, I think we need to remember that there might just not be much low hanging fruit. Right, a neural network that's been trained to convergence and then distilled to something efficient is not that easy to improve.
Props to Kevin for that EPIC closing line!!!! Danny better watchout! There's a new Blazement Master!🤣
Glad you enjoyed!
As a qualified and professional hair stylist and colorist ,, this makes a big sense to me
There was an estimate about 10 years ago that it would be 400 years before such a robot was created. That’s the far limit of Simon’s prediction; if he was being serious.
A sentient ai could love, Simon. And by saying that i should be safe from our future robot overlords
I always think of the song Wykydtron by 3 inches of blood when it comes to super A.I. In the song they state the year is 4055 and the self fulfilling prophecy Simon mentioned comes to life. A good listen if you're interested in Metal music.
200 to 300 people tuned in to watch. That is almost twice as many people that tuned in to watch the superbowl that year.
Hell no.
Haven't had a chance to listen to this yet, but the script definitely says 200-300 million, not 200-300 people.
What if you made a narrow ai that is Tuned to refine program code and give it the ability to refine its own code? If course there would need to be savety redundancies in case it breaks its own code unintentionally.
Then you'd have a narrow AI that is really good at refining its own code... and nothing else.
My computer gave you a thumbs up. 👍🏼
It's kind of weird how this video follows the exact some logic that it debunked in the first 5 minutes to conclude that AGI won't be possible in our lifetimes.
"I can feel it Dave...." Cheers from the Pacific West coast of Canada.
I'm still waiting for the world of Ghost in the Shell to become a reality so I can have my brain transplaneted into the body of a cool robot lady.
I think a hive queen that specifically dictates other ai like a chess game could be what we move towards. As long as all of its parts (narrow ai) are compartmentalized and able to be shut off with their own safety switches, I don't see it turning into a skynet since it's just a think tank and specifically dedicated to tasks and not surfing the network, just sending direction to its parts via the network.
I want it as loyal as a dog 🙂
Not as a cat
The punchline at the end is amazing!
A good starting point would be to come up with a working definition for "sentient".
What is needed is a narrow AI that is tasked with designing better and faster version of itself, and then the newer version follows the same task, until we reach a narrow AI that is so good it becomes a general AI.
I think the problem with trying to create sentient AI is the notion that sentience is a property of intelligence alone. But everything we know or suspect is sentient has a creative intelligence that only modifies or assists a far more basic set of priorities. I seriously doubt there's any level of pure intelligence that will make up for the lack of needs/wants/desires that sentient life possesses.
“Two to three hundred people tuned in… that’s more than twice the number of people who watched the super bowl”
Wait what 😂
Maybe he means two to three hundred million people.
one thing that many misses is that we really didn't know why AI wouldn't grow till last 15-25 years ago..we thought software or maybe hardware was the limiting factor until we understand,yep,the problem is the number of models and training which should be huge and not just huge but on billions and billions of data to achieve the impossibility and get things done the way we expect it, better and faster and with no to less issues in generating the results-outcome.
that's why the acceleration is just phenomenal compared to what we predict for future in AI development and reaching sentient AI
No, the problem is that we still don't know what intelligence and conciousness even are and people misusing AI and sentient.
We've managed sentient computers 70 years ago. Sentience just means reacting to stimuli. Pushing a keyboard button is a stimulus.
Less afraid of computers destroying humans and more afraid of a computer deciding that it could "improve" or "save" humans; like a borg situation or mind controlled people to get them to make the "proper" choices.
I don't think real AI is far away at all. Auto makers are very close to developing AI that can wreck as many cars as humans.
🤣
I am thirteen year old Ukranian boy, and I approve this video.
I assure you, fellow humans, that I am boy, not computer.
How doing, fellow humans?
ok, how many buses do you see
@@richard_d_bird something between two and four, perhaps three.
I like when Simon's channels are too new for ad intros.
a couple of months ago. a kid was playing a ai robot in chess. the robot made its move. As the kid went to make his move the arm grab him by the wrist and broke the kids arm.
What about an AI like Joshua from WarGames, where its hooked into the defence grid, and shows its capabilities and because its a shiny new bauble, but ends up through a glitch in programming, launches a WMD, resulting in the mutually assured destruction that those weapons were created to prevent? It doesnt even need to be a SkyNet, all it needs is to remove the human critical thinking factor out of the decision. Unfortunately, we are so self pleased with our technology, we may think that removing the human from the equation, in favor of the quicker, analytical non emotional computer is the key to success.
Wow did this video come out at the right time - Right as chat GPT was taking off
I believe that once true sentient AI is achieved movies like Star Wars, with scenes depicting brutality towards intelligent machines, will be viewed as very unsympathetic and difficult to watch.
Where are the memes for "somewhere between 2 and 4, maybe 3"? I haven't seen any!
I, for one, welcome our female robot AI overlord girlfriends.
If Today I found Out and Busin... Brain Blaze had a baby, it would be this channel
Only going to address two issues: The first is that "improvements" are less about methods or understanding of the science, and more about the enormous amount of computing power now available. Narrow AI is just brute force that mostly seems to work but it's just a facade. The second thing is that sentient is specific to being embodied. Long before nature created brains, the body needed to function within a range of environments. This has been the driving force behind complex behaviors that we humans look at (in nature) and claim there must be some intelligent mind at work when it's still just systems within systems. For the machine mind to really have human-level cognitive abilities means it has consciousness (though there is a whole tangent about machine consciousness not being the same as human consciousness because the machine doesn't have the body to give it "emotions" or the kind of embodied feedback loop humans have). *all that aside, 2056 makes for a compelling goal, with better than 50% odds we achieve something before then.
No. It will take until the early 200 to get that.
I think AGI is 10 to 15 years away. I'm just looking back at the past 10 to 15 years and looking forward at an exponential growth rate.
I'm not at all worried about the prospect of an artificial general intelligence.
I am extremely worried about whoever gets there first.
Let's load all of Simon's videos into a deep-learning AI, and then ask it to choose a topic at random from Reddit, and create a 22-minute video starring Simon for this channel.
"People are the worse."
"worst". And worst editing of subtitles
How about to create narrow AI to improve humanity existence expectancy as a specie ? If we can some-how manage to put together all the inputs, I would be rather curious about the solution . . .
The biggest problem with AI is that it can have access to all the information but it can't understand things and the relation between them as we do.
Precisely. Add the fact that a computer can't make moral judgements. Everything is not black and white, or a 1 and 0. There are shades of grey that computers be pogrammed for.
What a surprise, Simon has yet another channel haha
Do they ever do machine vs machine chess? Like best chess coders against each other.
Yes, their are computer chess tournaments. I believe the current top three engines are AlphaZero, Stockfish, and Leela Chess Zero. The minimum rating to enter a computer into the tournament is 3000, which is higher than any human has every achieved.
@@ThatWriterKevin once again Kevin comes to save the day…
Also to the other replies. Part of making a good chess computer is to have it play against its self or a copy of its self. It can grind away at many many games and find places where its logic fails it and automatically correct those.
Just remember, Skynet became self aware 25 years ago on Aug 29th 1997.