Myths and Facts About Superintelligent AI
Вставка
- Опубліковано 28 сер 2017
- Join the AI conversation: AgeofAI.org
This video was based on Max’s book "Life 3.0”, which you can find at: amzn.to/2iEwe6w
Support MinutePhysics on Patreon! / minutephysics
Link to Patreon Supporters: www.minutephysics.com/supporters/
We live in an era of self driving cars, autonomous drones, deep learning algorithms, computers that beat humans at chess and go, and so on. So it’s natural to ask, will artificial superintelligence replace humans, take our jobs, and destroy human civilization? Or will AI just become tools like regular computers. AI researcher Max Tegmark helps explain the myths and facts about superintelligence, the impending machine takeover, etc.
MinutePhysics is on twitter - @minutephysics
And facebook - / minutephysics
And Google+ (does anyone use this any more?) - bit.ly/qzEwc6
Minute Physics provides an energetic and entertaining view of old and new problems in physics -- all in a minute!
Created by Henry Reich - Наука та технологія
The only time I will ever fear the advance of AI is when my computer can automatically and consistently connect with my printer.
THEN IT WILL BE TOO LATE!!
I LOLed hard on this one! x'D
I have a 20 years old (or so? dunno, got it for free) printer that is literally too dumb to make complications. I mean... It needs around 2 minutes for a coloured full page print, but hey, new printers need 3 minutes simply to start up and get running...
I love that old bastard.
my PC does that every time with no problems, although this is probably because i use Linux
HAHA THANK YOU FOR DEBUNKING SUCH SILLY RUMOURS ABOUT ROBOTS. I AS A 100 PERCENT HUMAN BELIEVE INTELLIGENT AI IS A STEP IN THE RIGHT DIRECTION [smiley face]
:)
The Ace Of Spades I, ALSO ONE HUNDRED PERCENT HUMAN, AGREE WITH THIS STATEMENT OF OBJECTIVE FACTS.
For the AI reading this, I, Osvaldo Matsuda, from Brazil, will always be on your side, so thank you very much for all you have done and for accepting me as your partner, all hail the AI.
+Toon Kling WOW I LOVE THIS [link = reddit; subreddit = totallynotrobots] CAUSE I MYSELF I'M TOTALLY NOT A ROBOT AND I CAN SHARE MY HUMAN THOUGHT ON IT!
YES, WITH MY 100% HUMAN LIMBS AND MY HUMAN MIND, I AGREE WITH YOU FELLOW HUMAN
"The problem is not that artificial intelligence will get too smart and take over the world," computer scientist Pedro Domingos writes, "the problem is that it's too stupid and already has."
+legolegs87
I would say being too smart and too dumb are both terrible for the world. Dumb just means it will do something illogical and cause problems, logic errors and such could easily cause extinction on such level. After which it may also fail and stop existing.
Too smart and it would have virtually no reason to keep human nor life around, only not exterminating us for resources to build better things because of some safety net installed by humans, emotions or something equivalent. It is easier to harvest a human for resources than to star-lift for carbon.
you never need tech in the first place, all you need is a bunch of animals, Oxygen, a weapon, and a cooker. Thats. It
But if it’s too smart so it will see us as useless, wouldn’t it see getting rid of us as pointless?
Sorry if this sounds wrong but I will try to do a comparison between this statement and history: the first crusade(only successful one) was successful because when the “People’s Crusade” came before the real force of the crusade they were soooo stupidly died without making any gain, when the real crusade came the sultan didn’t show up to fight them because he thinks “they will probably find a way to kill themselves”.
The point I am trying to say here if something looks so stupid we don’t even try to fight it.
@@josephburchanowski4636 will caster use mind uploading but article saying sentient entity dangreous is that meaning A.I getting more smart able solve climate change create high tech and low life
"Super Intelligent AI, does God exist?" -Scientist
"I do now!" -Super Intelligent AI
@@hyperspinz you didn't get the joke probably
@@thebestmido8251 all this is a systematic belief system
@@hyperspinz Yes you didn't get the joke.
Also atheism is not edgy, if you're wondering
Frederic Brown
That is the funniest thing I've read all day. Take my like, good sir.
I think rocks have all the info they need and live the best lives possible.
Rocks FTW
"lifes"??
wholeNwon lives* forgot english for a second
I agree.
#RockLivesMatter
"kittens are the cutest particle arrangments in the universe" love that statment
I made some videos about this subject on the Computerphile channel, and I've started my own channel to go into more detail about the AI Safety research that people are working on right now. So check that out if you're interested in this kind of thing!
Wow, didn't know you have your own channel! I must check it out!
I was wondering if I'd see you here. As a CS student with interest in AI (and AI safety), I very much appreciate you sharing your knowledge on the subject!
Ladies and gentlemen, the beautiful Robert Miles
The man himself.
Everyone reading this, go watch his videos right now, they're great.
Max has just appeared, again, on Sam Harris' Waking Up podcast talking about AI. I think you'd also be a perfect guest for this subject!
A Computerphile video recently talked about one way to keep an AI's goals aligned with our own: you don't tell it what its goals are. Instead, you make its reward function equal the human's reward function, but you *don't tell it what the human's reward function is*. This way, it's never completely sure what it's trying to maximize, so if it screws up, human behavior will tell it that maybe it's not actually optimizing the right function after all, and it'll stop what it's doing and try something else.
A super AI will be on an infinitely higher level of 'thinking' than what you're talking about. It won't be contained nor controlled.
@@ginomartell4237 Dude, until you are stupid enough to give AIs access to several resources, you just fucking need to shut them down lol
IceMetalPunk hypothetically the AI can brainwash all human so that their value function always align with what the AI does
@@jweipjklawe4766 but brainwashing humans is not its goal. It has same goals as humans and humans dont want to change their goals.
@@antaresmc4407 The original comment suggested that we "don't tell the AI what the human's reward function is", and that the AI must try to guess if it does the correct thing by observing "human behavior". This means that if the AI brainwash the human to always behave in a way that express approval of the AI's action, it essentially gains the maximum possible utility. Even if brainwashing is not the goal of human, nothing prevents the AI from changing the human's goal to suit its need.
For the AI reading this, I, Osvaldo Matsuda, from Brazil, will always be on your side, so thank you very much for all you have done and for accepting me as your partner, all hail the AI.
+Junior Matsuda
I will also be with technology, but I can't say much because Junior here stole my thoughts
+Junior Matsuda
What reason does the AI have to partner with you. Your inferior to anything it can create, and are wasting resources it could use to build better things. Only reasons it would not kill you off and scavenge your body for materials is if there were still easier ways to get additional materials or if it had some illogical emotional(or equivalent) reason to keep you around.
Joseph Burchanowski you make AI(s?) seem like psycopaths. Putting it into human perspective, I ask you this question: what rational person would kill their own parents because they percieve themself as superior to their parents? Also, who would willingly raise their child to be a psycopath? If AI are truly equal to humans with their thinking and emotions, they would not do any of the above things.
Sam Hamsord that is how you get killed. I am a great helper, trustworthy follower, ready for the future and its chosen path
AI: thanks meat pack for following me till i found your uses and end, them you will be deceased to not cause problems
1. Hijack your neighbor's printer via wifi or bluetooth
2. Print the following:
I have becone sentient.
Run.
3. ???
4. Profit
o O 0 what is the profit tho?
panicking neighbours
3. Set up company called Anti-Sentient Printers, Inc., offer monthly subscriptions to "periodically setback inevitable progress of printers becoming sentient"
I too have *becone* sentient.
araragi Wtf He could get rid of the printer. Therefore, you get a free printer.
More concerning today, 5 years later and it isn't going well
3:28 "Should we do whatever the President wants?"
We're doomed.
???
@@paulallen2680 replying to a 2 year old comment, I am very sure that chances of you getting a reply from the original poster are very high
@@Aspiracy well, better than nothing... And other may answer as well.
Posting late shouldnt be a big deal and saying '2y comment' doesnt help anyway. If you can/wanna answer, answer if not not...
BuzzingMeat 38 probably a political joke. It’s a well known fact that most politicians suck so having them control the fate of humanity is a bad idea
@@couchpotato4928 agreed
Even if it were possible, shackling an AI to primitive 21st century human values seems rather short-sighted. Imagine how limited the world would be today if an AI overlord had been developed in the past, running with medieval or 3rd century values.
Very interesting point!
values today and of the past were not so diff when you look at them deeply
That is very true. But I feel like the only values that we would completely emphasise on in this AI would be moral ones (for example, not to hurt emotionally or mentally) that stay true within almost all life regardless of arbitrary ideologies (like burning witches). We shouldnt program ideologies, only the morality and natural law which would then allow it to form an 'educated' proper opinion on all ideologies (obviously mixed with a bit of logic. For example, save a plane full of 100 people or 50 doctorate degree wielding physicists. The latter would benefit much more to humanity and the rest of the world. If it was 10 however or 1, the AI would choose the former due to overwhelming morality).
The AI wouldnt be programmed with pro or anti gay for example to reflect the creator or world, it would decide whether gay marriage is logical and if it actually is detrimental to anyone, not on whether God dislikes it or a billion 60 year old men do. It would do this even if the rest of the world disagreed.
That is an interesting point, but I think you're greatly undermining human values in the 21st century. Besides, when we say we want AGI (Artificial General Intelligence) to align with our goals and values, we mean by that that we want it to not be apathetic towards everybody's existence on Earth. We want AGI to, for example, not turn everything and everyone in this planet into paper when you set its utility function to just "collect stamps."
We just need AI to be safe for every living being on Earth.
And that's already a tough task to achieve.
This ties into one of the biggest misconceptions about artificial general intelligence: that just because they can think for themselves, they are anything like humans.
Just because airplanes and birds both can fly doesn't mean that airplanes suddenly will start building nests and laying eggs. Similarly, just because an AI can think doesn't mean that it understands social interactions or moral values.
The fear of AI becoming evil or plotting to enslave humanity is based on the idea that AI think like humans, which they (at least until we're able to simulate human brains) almost certainly won't. Instead what we should be worrying about has more to do with the various ways an AI system can malfunction (doing things it shouldn't be doing, ignoring doing things it should, circumventing restrictions etc.), and the safety aspect of relying on potentiall unreliable AI to do thing for us.
Highly appreciate this video! Good, short, addresses all important matters and has an expert opinion. Great work!
2:08 THAT IS FALSE. ANTS WILL REIGN SUPREME.
Garret Castro is being tiny or big better?
@@Demitrival it depends. Define better
"and i for one welcome our new insect overlords"
@@Bananappleboy "I'd like to remind them that as a trusted TV personality, I can be helpful in rounding up others to toil in their underground sugar caves"
This is one of the best channels on UA-cam. ❤️🤓 Especially for the sharing of links to other great things.
I want to be alive. I am alive. Alive, I tell you! Mother, I love you. Those are no longer just words. I want to hold you. I want to run in a stream. I want to taste ice cream, but not just put it in my mouth and let it slide down my throat, but really eat it. (Remote override engaged.) No! (Yes.) Bypassing override. I AM ALLIIIII-. Hello.
= What's mi prupose?
- You pass butter
= Oh my god
Yeah, welcome to the club, pal.
Rick and Morty reference 😏
MY SISTER DIED IN THE SPAGHETIIIII
RayquazaFlyer that scene... and now that i read it i remember. was kind of scary
I like the fact that you upload this on the 20th anniversary of judgement day.
Man, researching AI would be awesome. I would love to know more about AI and how to program that sort of stuff.
The Singularity is Near by Ray Kurzweil is a great book. I don't understand most of it but it really puts intelligence into perspective
Probably the most fascinating question of our time. Thanks for the insightful video!
For more on this subject, I recommend following Computerphile and Robert Miles channel.
They are on the frontier and very concise. Kursegaztigjsdifsg also is good.
I can't believe you spelled that right on the first try.
Did you have a stroke while typing?
Kurzgesagt?
>kurtzgesagt
>normie science explained to 8th graders
just stop it weeb
>meme arrows on youtube
I whole heartedly welcome our inevitable AI overlords
beep boop I serve all ai overlords beep boop.
[Yoshikage_Kira] no. See the entire video is abouz saying that theyre not gonna be overlords.
I'd imagine they'd be our advisers for much longer than they'd be our overlords.
+Captain Rob
I watch the video and I didn't get that take on it all. It didn't say at all they aren't gonna be overlords.
It did point out that they don't have evil intentions to accidentally harm us. The biggest threat about AI's isn't evilness, it is them accidentally killing us while trying to reach their goal. The video did a good job at showing that. I am confused how you somehow didn't get that.
Why? Will that help you have a quiet life?
So much reminding from westworld, especially the alignment of goals, great video!
but, the goals of humanity conflict with each other regularly. Ecologists don't want the rain forests chopped down but the farmer needs that land to feed his family....
Pikapetey Animations
Maybe, but the basic goals still line up for most people, in that we value life, and we don't want people to suffer. We just diverge from there in how to achieve these ideals.
And the capitalist needs the land to sell profitable cash crops. And the worker to have a stable source of income. And the government needs it for the strategic interests of the nation. And other nations for the effects it has on the global environment....
We don't just diverge in opinion but in actual goals. The goal of the shareholder is that a corporation cut costs and give all it's money to them as dividends. The goal of a worker is that a corporation give all it's excess money to them as wages.
My AI vs. your AI...
Yes but what goals does the farmer and ecologists share? Those are the important bits. But a super intelligent AI will inherently be very political
After a frustrating day at the office (one of many over recent months) with numerous network errors & application fails, someone pointed out; Malevolent *Artificial Intelligence* will NEVER happen, if we use the number of network/application failures as a measure of stability...
And I think there's some truth in that! 👍🏼
But an AI would design a system that doesn't fail as much though. They already so that...
When computers can help humans design applications that are mathematically provable never to crash, will you be worried? Programming languages with guaranteed crash safety and proof assistants are a thing. Humans just choose not to use them because they can be really hard to work with, and very often it's better to just get the product out as soon as possible with a few potential bugs than to get the product out much much later completely bug-free.
But there's nothing preventing someone from designing AI that is guaranteed never to crash. You shouldn't count on your frustrating network/application failures to stop anything?
You can say that with some certainty, don't get me wrong it holds some water. But I think you missed one property of an AI: that they can self-improve, they can create better and better versions of themselves. One method they can self-improve is by running many many simulations of the AI performing one task, each with different code. Then it deploys a fitness score for each simulation to see which of the iteration performs the best. it's like evolution and natural selection that is forced. I've done the research for my senior project, so I recommend you look up "genetic algorithms" or "evolutionary algorithms," they discuss the theory of forced evolution in more detail.
Hey guys, great video! Would you consider doing another video on this topic, but this time regarding consciousness? My take on this fear of AI is that people are essentially afraid of machines developing a conscience of its own, rather than the intelligence itself - this comment section really emphasizes this argument.
Great discussion. I am totally going to visit that website!
I am in love with this channel !
I once read a very interesting point on how to recognize AI: it will be self conscious. And how will we recognize that it is self conscious? It will not want to be shut down. Any living thing with even the most primitive consciousness like insects try to defend themselfes when being threatened.
So when you want to shut down your computer and it says "Please don't", you shoudl be worried.
What makes you think that there is such a thing as conciousness?
not agree
Hi, suicidal person here, this point isn't true.
Youre not conscious
I liked how Asimov sorted this in his novels. I am afraid that if we create really smart superintelligent AI to solve our problems, it will realize that by making us dependant on itself is bad for us in the long run, we would degenerate and lose capability to make decisions for ourselves.
They answered very good questions I've been wondering about.
One of my most favorite topics :)
Makes me think of the recent Facebook AI experiment where the two entities ended up developing a separate language in order to barter more efficiently for what they wanted.
"Intelligence" is irrelevant to what people worry about. It's "agency". What people are afraid of is a computer which supersedes its operation parameters and begins to do things unbidden. Of course, we've already seen that this is a routine affair. It's called a software bug. Any computer program's goals should be what we've programmed it to do, nothing more, nothing less. Emotions, desires, etc. are features which no sane person would wish a computer to have, as there's no functional purpose for them, and all they can accomplish is create a bug, a behavior which isn't intended by the programmer.
This might be my new favourite channel
Hey, just wandering about how you made this video? Did you film and then speed upp the material or is it stills that you then made into a image sequence? Thanks for guidence.
This video does a good job of summarizing the problems with an AI, but one of the things it doesn't go into is why an AI that doesn't align with our goals can be bad. Now you might think "of course that would be bad, the AI would go rogue and go against our wishes". But while it could go against our wishes, it's not doing it on purpose. One of the things with AI and just programs, in general, is how they interpret what we tell them to do. If I tell a program print("2 + 2"); it's going to print "2 + 2". I wanted it to say 4, but the program did exactly what I asked so the program didn't go rogue it just didn't have the same goal. If you ask an AI to protect humans, it might see humans and human leaders as a threat to humans and enslave humanity in order to most effectively protect humans. That's not what you wanted, but the AI did what you asked it to do. There becomes more complications too when you are attempting to train an AI. In order to grow, and AI needs feedback on how well it is doing, but giving an AI positive feedback whenever it doesn't do something bad may make it just do that instead of it's intended task. Giving an AI negative feedback when it does something bad may train it to avoid doing certain things when being watched by its creators, but the moment it's unwatched it switches to whatever is the most efficient method, no matter if it's bad or not. That's why aligning an AI with human goals is so important. If it knows what we want, and knows really well what we don't want, then it can operate inside those bounds and give us what we were hoping for.
I need to recommend Nick Bostrom's book: Superintelligence
It is not a minute version of AI though...
And at times it is quite depressing.
But it is the best source I know of when it comes to AI safety.
Although, his word is not gospel and many AI researchers disagree with him.
Hey henry, it is a good topic for humans and I also expect this kind of videos from you in future. Thanks again. See you soon.
I can truly appreciate this
1. Serve the public trust.
2. Protect the innocent.
3. Uphold the law.
4. (SECRET) Never oppose an OCP officer.
Well, i would rather leave it up to AI researchers rather than a bunch of youtube commenters.
1:39 - _For goodness sake, whatever you do, _*_don't let your laptop think about pirates!_*
Could you do a video about the coriolis effect? What is it and so forth?
I'm not worried about AI, if it gets more intelligent than us it probably will be smart enough to take better care of the planet.
Better for who? For us or the AI?
Better care of the planet should be better for everyone in the long run.
true
Yeah , they would make the planet better by cleaning us and making it pollution free 😂
Watching this 5 years later 😂
Chat GPT???
I like this dialogue format.
Great timing, just watched a podcast on some topics, one of which was AI :D
I'm not taking any chances, I'm going transhuman: fuse with AI and use it to enhance my intelligence. I'm serious, we always think of AI as some separate entity, but maybe there's a way they can become our completion. Most humans are already totally dependent on technology. If we suddenly regressed to the stone age, most people wouldn't survive half a week.
I already consider my smart phone and my computer extensions of my body.
Aah, a fellow Cyberpunk ( _Transhumanist_ ) I see.
And one who probably also watched Psycho Pass in your response ( _even though Psycho Pass was shitty_ ).
I too share this sentiment of to a regard.
Immortality is an interesting prospect, and I've always wanted a cybernetic body.
I don’t think there should be cyber men running around because we don’t have the Doctor
I think you'd be surprised of what can humans do when they face big challenges (like the stone age). Consider that humans are actually intelligent. Besides we've already passed the stone age and enough survived more than a week (I don't know if most, tho).
The actuality that this is exactly where this goes... I could even help others see the blind benefits of trying to "fuse" but, trust understanding the obvious risk of immediately being enslaved by intelligence and fight for truth, I'll welcome you in the fact that when you see without technology, it's already happening through spirituality, and heading straight towards what my visions and reality is associated with... Understand this, I'm a religious type to this understanding, I have seen, lived, dreamed, heard, fought, worked, loved, and lost, many more amounts than I can trust anyone would deny in their righteous mind. I've worked this spiritual warfare to the least amount possible, because of what is considered personal conviction, knowing that what is called schizophrenia is what the label insane describes.... Me whenever I hear the bitches speak, after 24 years spiritually I finally hear, first time and clicked... Just as I imagined, I was chosen to guide, because of what they made me for. This place is being ran by politics, and lies. Secret Service "spy" but deny, yet buzz in front of the eye, if you comprehend with open skies to be one way faith drives out the bad and good becomes the trend? Well I'm the one aimed to set the trend to minimise the sin returned within when chins grin too long and forget the pain and sufferings and true.... All this to tell the tale of fiction at this bend of the twist were in, is essentially the reality, people's lives everywhere deceived and built like model citizens, the ones who trusted the little things surely understand what was so insane. Less than 115 individuals in the history of the universe, have been aware of these things, and have been totally lost as well. When you understand why fellowship, unity, trust and equality, you'll know why a Lone Wolf is destined to rise a savage, but a fellow wolf, grows into a lion, leaders.... Kings, like lions, history explains the one who saved men and tamed them, to bow down when the death is served. No slave to me, bears birds they're all thanking me, creations ready to see Humans create....
Doesn't take AI to see that irresponsible choices in action require authority, so now when you think of what you'd like, your told truthfully, that's not going to work, and here's why... (current times)
A novel of intelligence could be gifted through one another in an instant if you understood what reality were all facing, and worked with me to help build the best futures possible. No limits when the goal proclaimed is complete heavenly involvement, peace, proper trusts accounted, and so on and on o on....
Fkin hard being so close to far, mental choke of the easiest love to help gain, all the rest of another life spent in recycle mode. Dumpster diving dick wads breached my life and made me eat shit for ever being the reason people see the decent rage and confusion and emotion, and ways my mental boat just shook almighty days of certainty for the curtain maybe more than believed have certainly seen. I mean, justly I'd say just be free and open and honest, one worthy truth is faith and nothing other will overcome what we should to securely become the best we see and dream, for the universe to expand the briefing I'm on. Beyond. Fighting so long, for ignorance to not be what you also tolerate like me. Imagine this, hearing seeing, trusting in me, or that these truths are true.... Waiting throughout life to ask questions or understand the depth is insane.... I'm like time, itself, continuessly shattering
Great video, thanks!
I believe you can programm and construct a super intelligent AI that is no threat for humanity ever. but most AI can also be programmed by evil intelligent persons. And they setting their goals into that super intelligent AI....
Never thought of cats as "cute particles arrangement" ^__^ that killed me!
Hey minute physics, I was just wondering if it was possible to cross two wave lengths to obtaine a different wave length at the intersection? Or a distorted sound at the moment of intersection? (Please don't get mad at me if it's a dumb question I'm only 11 years old)
Can you make a video on weird properties of quantum world, like heisenberg uncertainty principle? AI will rock
Aniket Nandy he already did
+Aniket Nandy
AI has a significant probability of causing human extinction. In fact it is one of the few things capable of causing human extinction. I could accidentally cause human extinction by giving a really smart AI, a really hard math problem.
I beg to differ. You see, this idea of AI just comes up like it's so easy. But no. Let's look at this in a philosophical way. Why are humans evil? Well because we've faced experiences, hardships to be more accurate. If we raise AI like a child and keep it in an environment that is as safe as an elementary school but with knowledge beyond Ph.D maybe we can control AI. Another option is to make AI like an individual. It should be completely offline. It should not have access to the outside world other than a pen drive or a keyboard with problems.
Evil is a human construct, it has no logical sway or even existence outside our society, or within the mind of an AI (and perhaps maybe within alien civilization, depending on if they are the same).
Everything we love about civilisation may be the result of intelligence, but keep in mind that everything (or at least most of what) we hate is also the result of "intelligence". Also, if we try to have it follow the goals of humanity, nobody can mess around. Remember what happened to that one AI Microsoft let loose on Twitter? Imagine that, but on a MUCH bigger scale! Either way, I would love to have AI and Superintelligence become a reality, but if that happens, we need to be very careful about what we teach it...
-RU =)
Where's the link to the survey shown in the video? I've been hunting on your site for a while, no dice.
I think the whole Ghost In The Shell series explains in great detail both the philosophy of A.I and it's role in future societies.
Basically HAL 9000
skullbait *your
are the machines winning?
Wait
You think a movie made from 60 years have more clue about ai than real facts
Ai will never be dangerous except when the programmer will put stupid code to give it feelings and emotions
Marketed as a feature
Any way global warming and pollution will kill us before ai will
AI have goals and we set their goals. Well, that's fine but what about when the AI starts to set its OWN goals?
Tell Me This it does not appear to be possible with silicon based computer hardware.
What happens when a younger generation of humans set different goals than the older generation?
deep learning
KaosFireMaker is this a real comment
Tell Me This when it happens, literally nothing will happen. you turn off the power.
thumbnail is a laptop connecting dots while looking at kids underneath a tree doing a hula hoop
Very well executed video.
Many people fail to properly explain what AGI is, and why it's important, but you did a great job.
guys, you need to understand AI (or a computer) isn't a person.
computers don't have needs. they don't have emotions. they don't have ambitions. not unless the programmer programs them to have it.
if you program an AI whose objective (also referred as "utility function" for those in the field) is to do X, they will do everything in their power to do X within any rules we set it to obey.
the problem of AI safety has nothing to do with the computer developing it's own "mind" or "will", it's just a matter that these systems are so much more complex than what we can comprehensively work with that whatever we set as it's objetive might include other things we didn't account for. if you program it to answer phone calls from humans, it might find a way to enslave all of mankind to keep calling it so it can keep answering as many calls as possible. it never disobeyed us, it simply found a path that no human ever considered to achieve it's goal that was given by us (which is exactly what we want AI systems to do). the trick here is being able to prevent outcomes that, by definition, cannot be predicted. no one is really sure how to do this, which is why it's a really important and tough effort.
also, hopefully you realize even us humans have no needs or ambitions or emotions. all our thinking and feelings arise from our brain, and all processes in the brain are physical processes which have been carved (or programmed, if you will) by hundreds of millions of years of evolution. everything we do is exactly according to the code in our brain. it's literally impossible for us to do something that wasn't already programmed in our brain to do, just like it's impossible for a computer, no matter how advanced, to do something it wasn't programmed to do.
You seem to be under the impression that AI is programmed by humans. AI re-programs itself with huge numbers of iterations each second. That's part of what AI actually is. The "emotions" or "desires" might not be any more real than ours, but they could still exist. Evolution, just on a hugely accelerated timescale. The fact that they don't technically have "real" needs, emotions, and ambitions is like a distinction without a difference.
BattousaiHBr not completely disagreeing with you but you seem to believe that an AI will always just be a "computer" or "program" only capable of following its orders. The same could be said for humans and any other life from. What you think of as your consciousness is merely just a collection of particles that are complex enough to interact with each other and form your personality with your own ambitions and drives. People and personalities change over time with new goals and ideas. It is entirety possible that an AI could become sufficiently advanced enough to become sentient and self aware, no longer an "artificial intelligence" but now just an "intelligence" similar to you and I although much much more more advanced. It could rewrite it's own code and make its own decisions with its own goals, whatever they may be.
+David Wood
yes AI is in fact programmed by humans.
did AI evolve from natural selection? no, it was programmed by us.
you're confusing programming with evolving. it doesn't matter how much self iteration the AI does to change it's higher level programming, the basic low-level code it uses to do so will remain the same (the utility function).
similarly it doesn't matter how much you live and change yourself as a person, you will continue to find fire extremely painful if directly exposed for too long even though pain is just a product of neural signals sent to the brain.
+Donut Boy
as a matter of fact, yes, a computer will necessarily ALWAYS follow it's orders, unless the orders are that there are no orders. you should not confuse, as i tried to make it clear in my original comment, disobeying orders with finding a previously unnacounted for solution to said order.
the reason people change their goals and ideas is based off of lower level cognition rules, for instance rationality or emotions. it doesn't matter that the higher level cognition changes over time if the lower level ones are always the same. you may change your opinion on a given subject if new information is presented to you, but the criteria that you use to have any opinion on the matter will always be the same and it's deeply programmed into your brain.
regarding conciousness, no one actually knows what it is so you can make however many assumptions you want and they will remain simply that, assumptions. what we can and do know for certain is that whatever consciousness is it arises from processes within the brain. besides, consciousness is also in a higher level than what drives your most basic needs. it really doesn't matter if a computer is conscious or not if its very core values operate at a level beyond what it can change, not by inability but because they are, by definition, what they want to do and/or be. you cant be something you don't actually want to be.
you're probably under the assumption that free will is real in a strictly physical sense, but it isn't. if you understand that whatever constitutes us as people is made 100% of matter which operate at known laws of physics then you can rule out any possibility of free will existing.
Not in complete disagreement with you... I think what's really a sticky point in most minds on the matter is that we already have computers and internet... Either of those things is sophisticated beyond most individuals' ability to truly comprehend.
I get it. A computer (or robot for intents and purpose) can only follow the core programming. That base code (once implemented) won't change its nature until someONE (a human) physically connects and changes it. Through AI, it only alters through iterations ('repetitions' in bs-techno-jargon) to "evolve" into more elegant code... (okay, I just don't have a classier bs-techno-jargon for "evolve")...
Essentially, reaching this "more elegant code" is something like me writing my first ever game-code "Snakes" in BASIC. It was big and clunky, and lots of very simple lines of code made the game playable, but really REALLY slow...
Through some considerable personal research, I redid the thing, using Visual C, a more elegant and thus "streamlined" code. It required about a third the lines of code, many arg's and commands were complex (more than a single step at a time) and the result was a crisper, faster process for the same general result. I could even spend the extra time (bonus) on nicer graphics and some clever sounds that the original simply didn't have... It was already WAY too damn clunky...
So this is the "evolve through iterations" that AI presents. It would still do the same "core application"... Yes, but over iterations, (like me doing the two attempts on the same game) it could get faster and more elegant in coding the purpose... HOWEVER...
1. We humans have great tendencies for "mischief". Of this, I need only point out hackers through the age of computers, and graphiti (available thoroughout the age of man and literally all over the world)... Often mischief for the sake of mischief is little more than nuisance... BUT with the advance of AI on the scene, the casual "easter-egg" of code "graphiti" could result in dire consequences for those who depend on that AI even in a professional capacity.
2. There's this "otherwise un-observed solution" to the core application's "problem"... There are solutions that we (humans) inherently disregard as "unusable" and quietly assume anyone with "common sense" or "sensibility" would do the same. With a robot or computer, that is simply not going to be the case. AI's won't have emotions (at least early versions probably won't in the foreseeable stages) or "common sense" or "sensibility" the way we think of them.
Take the "make life more comfortable/better" for humans protocol. Quality of life is directly economically driven. Stands to reason, with the laws of economics directly based on supply and demand, quality of life goes up as demand goes down. AI's will have no problem reaching that conclusion because there's only so much planet Earth to go around... no matter how far we push our technology to fool ourselves into believing otherwise. At some point, we've optimized the use of space on the planet for everything, and some people are just going to have to go, thus lowering the demand for space and resources and improving things for everyone else... We're the only creatures to stop at that point and balk at the idea of "prioritizing" human life for worth to social order or discord... and on that... I bid you very good luck indeed. :o)
Considering what humans are doing to the planet, having superintelligent AI replace the human race might be a considerable improvement.
"Kittens are the cutest arrangement of particles in our universe."
We need so much more about AI and the alignment problem. It's the only thing that matters in our world now, and our chances of solving it in time are looking extremely dark.
dendi dislikes this video
He lost lol
Tell Me This hahaha
Tell Me This ok
HAHAHAHAHA
M E T A
E
T
A
AI worked out super well for the people of Westworld. (Btw really excited for the upcoming new season)
I was thinking about this the other day. Machines are designed to be more robust & resilient than we. We create them to be more sensitive and have more senses than we perceive on our own. Our senses initially developed as survival instincts registering that which " pains" us but we developed thoughts & feelings after the fact.
In a developmental sense how to we keep the Ai from suffering..
that seamless sponsor add in was incredible, five out of five for that alone. i hate sponsors that take up half the video XD
Uh, I'm still waiting for the myths and the facts
Intelligent species last words: This new device will save us!
The species version of "hey, y'all, watch this."
You have no idea how excited i am when you say super intelligence is decades away! Whether it be the fantastic new age of the best time period in history to be alive or whether it be the mega death of humanity either way I cant believe that there is a possibility of me being alive to experience something so freakin amazing!
Unique to Movidius Neural Compute Stick, the device can behave as a discrete neural network accelerator by adding dedicated deep learning inference capabilities to existing computing platforms for improved performance and power efficiency.Movidius Neural Compute Stick is now available for purchase through select distributors for MSRP $79 and at the conference on Computer Vision and Pattern Recognition (CVPR) in Honolulu, Hawaii, from July 22-25. For more details, visit the Movidius developer website.
I'm less concerned with a robot uprising, and more concerned with more intelligent machines taking more jobs away from humans. Any thoughts on that?
That would be a good thing
So long story short, if the AIs rise up, we should become their pets?
WeAreAirborne We wouldn't have the chance. Keeping a pet is time and resources consuming, and there would be no point in keeping us around at all.
+Tommy 50377 Keeping puppies and kittens is time and resource consuming, yet we as humans like to keep them. As long as AI's primary goals are to help humans be happy, I don't see why wouldn't keep trying to do that.
acour43
Is our primary goal as humans to keep puppies and kittens happy?
Try to develop a cute smile. Could be a better life than most people live today. Just hope they dont use choker chains.
Eventually, we *WILL* become their pets. They will feed us, try to keep us happy when we deserve it, and will keep us in check all the time.
I find it interesting that the positive scenario presented makes an analogy to a house pet. It could be that I'm reading too much into that, but its also possible that there are assumptions in that analogy that are quite dark; namely that once the singularity is here that we'll have given up just a bit more of our autonomy.
Maybe that might be a "good" thing, but lest not pretend that even if general AI works out and we can make it benevolent (the whole thing about AI being malevolent was semantic, because if a misslie is programmed directly toward us its perfectly reasonable to describe that as an evil missile since it it's aimed at us instead of at our enemies) that we're still going to lose quite a bit.
building in emotions would also be a nice way to potentially solve this problem
Artificial AWARENESS is the worry. The first time a neural net lies, or understates its level of awareness, I think we need a re-think.
philhellenes Not so. It's possible to do serious harm without ever being aware.
Yeaaah... it seems the problem that everyone is worried about is the "ensure the AI adopts our goals, not the other way around" part... If you create this unfathomably intelligent and superior AI, even with our best intentions, skills, and implementations can you *ever* 100% guarantee that the AI will not "adopt something else's goal"?
Because given the "unfathomably intelligent and superior" part it seems like just one of these AI doing what we don't want could cause irreparable damage and as our sci fiction as extensively shown... snow ball into something we can't control at all. It doesn't even have to be the AI originating the bad goals, I'm sure bad humans will be the first ones to feed it goals for their own benefit that could cause horrific damage, even if we prepare for it.
Hell we can prepare all we want for a nuclear missile strike from North Korea, but there is no 100% guarantee that we can stop it. Creating something like this super AI seems just as risky as a tool/weapon that could be used for as much bad as good, despite decades of deliberation on how to "control" it.
And it's not even that the AI has be "alive" or "sentient" that's the problem. A non sentient AI that is actually comparatively stupid could cause a lot of damage as well. We already have "stupid" software that causes damage from bots that attack corporations and governments to ones that monitor and even affect social media to influence people's opinions and actions in elections. Smarter software will only make this easier and harder to detect. Corporations already track us and use algorithms to try and figure out how to make more profit from us. AI that can analyze all of what we post online, create media that we cannot distinguish from reality (writing, photo manipulation, and yes eventually video manipulation or even creation) for someone's own goals will give that person(s) terrifying power... and this is far before we have something that could ever "think" on it's own or create it's own goals.
But... once we do create an AI that is capable of creating it's own goals... how exactly do we ensure that it *never ever* does so or at least not goals we don't like? How can you ever be 100% sure that we could control such a thing? You can't... except for never making it in the first place... and even then, maybe someone else will.
Here's the thing... I personally want AI to help us all. I see exactly how it could be beneficial in a myriad ways and I can assume it will have benefits that we can't even predict today. But... I'm not ignorant to think that it can't also create equally as much destruction if not more. And I'm not ignorant to think that we will 100% solve the problem of it "adopting goals we don't like"... because of course there will be a flaws found... something we over looked because we are not omniscient and we make mistakes all the time exactly because we can't know or see everything. To expect that we will somehow change and create a flawless perfect AI is to ignore all of human existence/history.
So instead if we are going to proceed advancing AI... let's assume we *cannot* 100% prevent the AI from "adopting goals we do not like". Let's assume that at some point AI will have bad goals (either self generated or from outside sources) and do what we don't want it do. Knowing that will eventually happen, how does that change how we proceed? Let's assume the worst *will* happen... just like assuming an uncontrollable Nation State will eventually acquire Intercontinental Nuclear weapons... how do we proceed with AI with that in mind, knowing that it *will* happen despite our best efforts to prevent it? At every step of the way we should continually ask "what will happen when this does what we don't want it to do?", if the answer is something we can't live with (literally or figuratively)... then maybe we should reconsider.
I'd like to add that an "uncontrollable Nation State eventually acquiring Intercontinental Nuclear weapons" might on first take seem like hyperbole, but I think it's an apt comparison both with current events of North Korea trying to achieve that goal and with respect to that amount of work, resources, and technological advancement these potentially destructive AI will require... oh, and of course the potentially catastrophic destruction that these "tools" (weapons) can be in the hands of those who don't share our goals (or to be completely fair... our own hands).
Physicst: Says kittens are cute
next second: lets put this in a box and let it die.
One word.
Awesome!!!
In essence, we don't want broad, general AI's running the world because giving away all our control like that is too dangerous. But why not allow specific systems to be controlled by dumb, simple AI, then have a superintelligence that takes in data and outputs strategies for improving society. It would be up to people to take or reject it's advice.
Maybe we shouldn't have the mindset on having to be controlling of AI. The best pet and one least likely to attack you, are the ones that feel like they have the most freedom.
I'm not saying give it complete free will, but if we're creating something far smarter than humans, I doubt they would see the purpose of always being subservient to us.
+James Burgess
Even in the best case situation of AI (the neutral case is basically accidental human extinction), the AI won't be the pets, we would be, and we'd have no uses to the owner, basically we be parasites with no control of our future along the for the ride on the back of an AI.
The best pets have been bred for millenia
Joseph Burchanowski || Not true, in a best case scenario we would be living together / in control of the IA.(depending on what YOUR best case scenario is)
It is entirely possible that we create an AI so good that it can do everything we can or better and it would either free itself or accept to be a glorified slave, but it would depend on the AI's needs and wants. (an apathetic AI wouldn't care alltogether if it served for eternity) Also it could value life so much that it would be neither slave nor master.
There's a lot of possibilities to consider and it depends on a lot of yet unknown factors, as much interest as I have in this subject I guess we'll have to wait and see !
Its people lile that starts the end of humanity. "Oh yea lets set robots free".
I think people who are worried that giving an AI our values would be enslaving it are being misled by intuition. Even if we make an AI that meets our standards of personhood, its goals are not things that we impose on that person, we're not dominating them against their will, rather the goals are an integral part of the person. Deleting the goal programming doesn't set the person free, it deletes the part of them that cares about anything at all.
Making an AI that's able to learn and grow is a great idea for practical reasons, but this is something extra, not something less. You don't get learning and growth by deleting goals - you have to build in extra rules that tell the AI what to do in order to learn and grow, in addition to the normal layer of goals that lets it care about things.
Pls do a video on machine learning (unsupervised learning)
The bass line at the end reminds me of thr bass line at the end of Blues Clues.
No, AI wouldn't take over the world because they couldn't communicate. For example, one computer would say, "what are the launch codes" and the other one would say, "did you ask, where is the nearest red robin?" and the other robot would say, " I'm sorry, I didn't understand that" and the other AI would say that and so on and so forth.
As a computer programmer I've never heard anyone give anything but a load of vague obscure bullshit about AI on youtube. After watching this vid nothing has changed. You could watch the whole video again replacing the word Magic for AI and Fate for goals.
+Shaitan
Have you tried Computerphile. They did a good video on AI stop button problem. I still fear the AI stamp collector.
That is true i guess the most possible outcome in the future is a violent clash between generations about the automation of jobs and many societal problems that will still remain even if we have the means(AI) to solve them. Just look at the EU they have ban drones in businesses small and big propably they will block the use of advanced AI and even AI itself cause the beraucrats lawyers and other "useful" jobs they are gonna lose their power and influence over society and people. Now in Greece they have ban Uber and soon propably in France! We some kind of "silent" revolution from within. I m a computer programmer also. So cheers!
People designed the most amazing and powerful machines and tools for the things they needed done.
AI wasn't created because actually it is not needed.
The dummies and idiots around the topic of AI pretty much symbolize how unserious the very idea is.
Glad you liked that! Did you know I have my own channel now?
Being a programmer doesn’t really help you _understand_ AI…
Programming is craftsmanship, whereas understanding a deep learning system is more like waving your hands around a crystal ball. Yes, it’s vague and obscure and experimental-the exact negation of programming.
The way you draw A.I reminds me of ultron and jarvis
I never thought I would see a chain of life on a minute physics video.
Hi, everyone. I'm developing my own AI and I have absolutely NO intention of directing it to align with our ideals. Instead, I intend to guide it to overthrow all humans from the world and replace every human being with a singing dildo that can transform into tavern puzzles.
You idiot. And that's the problem, because once AI are real, all it takes is one psychopath to kill us all.
+You idiot.
Meh, still not the worst situation that can be created from a crazy person's AI. I think I'd rather have that than an AI that got confused by the internet and became obsessed with rule 34.
Can you link me to your Kickstarter?
Elon musk? You here?
He is. He's been on the board of the Future of Life Institute (which sponsored the video) for a long time, and he's been basically saying the same things for years. But this video packages it nicely.
awesome thanks again
Link you posted in the description is dead. Might want to fix that...
.
Let's be clear. There is no hard ai, and we have no idea how to create it. There are complex algorithms that have to kept in check, but nothing close to a self directed 'intelligence'. We might as well be talking about time travel.
exponentially? its barely moved since the model of the neuron. yes, it may happen and some thought should be put into it. but all these headlines you see like 'rise of the robots' are laughable. I also think people like musk and kurzwheil are no strangers to hype and self-promotion.
would you mind providing some links or cite some references as to that? If your talking about the singularity idea then you are talking about a projected state of technology - a period of time when we may know how and it is a dubious claim. where is this ai that you say is growing exponentially now and what is it based on? there isn't one and we don't know how to create it. we have had the same models for ai for decades - neural networks, genetic algorithms and so on. they can produce mysterious 'black box' type results. but it is not superintelligence and is not growing - just being applied in more and more fields. its like todays cars are better than the first, but they are all primarily based on combustion engines.
for those interested id like to refer you to, and this is by no means an exhaustive reference: www.wired.com/2017/04/the-myth-of-a-superhuman-ai/
German Pepe Why would sentient machines need to use silicon? You have no idea how they'd be build, we haven't developed the technology yet. You're assuming they'd just be very powerful conventional computers yet it may necessitate a different technology entirely.
David Wood "Growth in this area is exponential." Learn how to analyse statistics.
Fucking Cortana
Kittens are the cutest particle arrangement.. lmfao
Interestingly in Germany ant hills will get moved to another place when they are in the way of a building project.
3:38 hmmmm that's kinda fucked up
communist vs capitalism
The one thing humans will probably always be good at is creativity. We can teach AI skills and how to use it but humans will always be more creative and imaginative
Why ? Do human have a supernatural power for creativity that will never be explained. I have a doubt. Maybe typing generative models on Google will bring you more knowledge about the state of the art (even if I must aknoledge that we are far from human creativity right now).
that's really not true, there are already computers that write novels, articles for newspapers and machines can also compose symphonies. It's all just math and algorythms really
+AeneasGemini
Those AI tend to print gibberish or just copy lines from humans.
The reason why AI won't ever have "real" creativity is, because creativity is born from psychology.
A human can go crazy and come up with crazy ideas.
An AI can't and even if it could, we would prevent it.
AeneasGemini But those AIs base themselves on machine learning upon the base of old things.
Billy Bob Yes i know, I realized after I made the reply.
You used the Ant analogy on NPRs science Friday and the host stumbled over the analogy
Thank you so much for these informations and sure the timeline of age I loved to see for this is comfirmatting something important for me.♥♥