+33rd Square Really? What are some of his best theories? From this video, I can see no sign of intelligence..first he brags about his own "achievements" then those of some group he is involved in...and even those are not that impressive for "the smartest people on earth"...
***** In my opinion, bragging is a high school phenomenon and adults don't typically do it. Self-promotion is one thing but this isn't that. Sorry. I have a degree in comp sci and have studied neural networks. 20 years ago. And this is nothing new, and nothing that requires a team of people to program either. Sorry. ***** Jeolous of this guy? LMFAO. Anyone can brag, and anyone can join a team, etc. I'm simply not impressed. End of story
+B. Alvn He's the one who created AI that learns by itself without any help. It was even able to beat the best Go (arguably the most most complex board game that exists) player in the world in 4 out of 5 games. If that isn't impressive then nothing that anyone does is allowed to be impressive.
Just the words I wanted to listen. And the right guy to speak so. I am so happy that the way I think about life, humanity and universe is not stupid and actually it's the same thought process of this guy who has seem more life than me and is way more intelligent and skilled that I am. I have found the purpose of my life. I just need courage to work towards it. It's exciting and I am feeling nervous about it.
+InfiniteUniverse88 At that point, I'd expect it to become wobbly static, as each pixel becomes part of the game, and each "opponent" has to manipulate it as fast as possible... I don't think it'll be any fun for human observers, but incredible nevertheless :)
It sounds cool, but I think the idea needs work. The AI that modifies the games needs some motivation to keep them solvable. How to implement that? Making the game hard is trivial, the difficult part is not making it impossible.
For those of you who don't have idea about GO can't really image what it means for computer to beat top human player in Go game. It's terrifying! Chess game is pure logical thing, Go is not. For computer to mastermind Go means the AL jumps to a whole new level. so people, get education first before you judge him!
Man, everyone thinks Terminator is a sci-fi movie, but I starting to think that it's become more and more feasible that computer's once the singularity is reached (a computer that can build a smarter computer than itself) why wouldn't the ultimate solution to how to survive be to wipe out or enslave the biggest threat that exists eg their creators? Certain protocols could be put in place, but some imaginative AI would simply concur that those protocols existence could endanger it's survival, and henceforth be ignored. The thing is, once an sufficient evolutionary AI is spawned, this would all onset at such a juggernaut pace that almost overnight we'd be done for. I can't help but think all these great advances could eventually lead to our downfall.
+EvilKris i dont see why AI want to kill us all. all movie you have watch is create by human its a human nature to kill,enslave,conquer and to take advantage that is a human nature. you should fear your own kind the human race look at what we done to other race and our own race we kill each other every day we kill every thing.
EvilKris we will put our "Minds" into this machines. Remember, we are not only physical, but also "minds". Our bodies will be "immortal" and we will overcome our physical limitations.
Magnificent piece; it shares the magnificence of a book I read with similar depth. "Game Theory and the Pursuit of Algorithmic Fairness" by Jack Frostwell
@@sdprz7893 It won't be too long before the AI's start creating algorithms and code that is beyond the understanding of most humans. I still don't think we should stop AI development though, the benefits are just too good. To me it's worth the risk if we end up with a future where AI robots do all the work and humans can do what they want
Two of the most important lines from this presentation: "As with all powerful new technologies it must be used ethically and responsibly" and "Human-level AGI is decades away be we should start the debate now"
Now you are talking! That senorio would cause an inteligence explosion, as each AGI is developed it can improve on itself faster than it was improved!!
+aoflex You know, the guy said their mission is to convert information to knowledge... I think a true test of AI/Human-level AGI would be to come up with conspiracy theories :P
+aoflex That's an incredibly bad idea. Do you want the extermination of humankind? Because recursively self-improving AIs you can't control is how you get the extermination of humankind.
+aoflex It's called evolutionary algorithm. That's when a program redesigns itself, to make a better program which does the same thing. Think like Intel CPUs, a machine is creating a CPU, you then put that CPU into the machine so it will create a slightly better one, then you take the slightly better CPU, put it into the machine and it will create a slightly better one.
OMFuckingGod!! I played theme park obsessive ly when I was 17, then theme hospital. are u telling me I was playing a game at 17 created by someone who was 17
Husnain Anwaar Sam Harris needn't be so overly worried about AI; as mentioned in this video, AI will be used in AI assisted science - the benefits could be astounding! .. it's sometimes to easy for humans to humanise (anthropomorphise) non human items, but we should resist doing that as it's not natural.
***** No. His concern is only about how extremists muslims are so confident in achieving paradise by suiciding on others (mostly other muslims!) that it poses a unique challenge. Only comparable perhaps was the japanese suicide bombers in WW2, but in this case the perverse extremist ideology can be brainwashed into seemingly innocent living good life muslims who watch the wrong propoganda. I'll grant he does paint a more bleak picture of islam as a whole but he is not specifically against the general muslim population.
Bryan Tianao Read their fucking texts moron! Don't fucking ask me crap like that. It all available online. All Abrahamic religions are evil and must be wiped off the planet asap, if man is to survive.
S Cin and there you go. As soon as you are proven wrong (ie you can't cite any non muslim examples like ISIS to backup your claims all abrahamic religions are evil) then you spout bile and other nonsense.. = brain IQ of a 15 year old or less. Have a good day! lol
The crucial thing is to get more people to pay attention to these lectures. I believe animals eat together because it feels safe. They aggregate in hopes of sharing resources and humane people can see this in ornithology, when Wild Turkeys are raised by a human upon which they imprint, or in studying cetaceans, when a whale investigates a human camera in isolation. Animals do this because they are happier in the presence of a similar affect and shared mind. This would imply that empathy and intuition are forms of wisdom which are universally expressive forms. So, these are critical to the new mental singularity, seeing AI as rotary and mechanical, in likes of this extricated essence, humane being is not subject to a cogito ex nihilo, just afflicted by IT. So, don't presume to know thyself, a priori, since etiological substrates which connect humans to animals and to computers are raised under the same circumstances. Bring humans together; bring ethos together. Bring humanity together; bring humanism together. This is how ethics is reformulated to abide Stoic concerns, and how we get out of the cat-box and into the bird-cage or the fish-bowl.
Probably it's impossible for AI to understand history of the music, arts and cultures the way humans understand and interpret it. We will still need need musicians, writers and culturologists, but development of their professions will be quite different. Linguists will still be very useful, because analysing nature and history of any language is very important. I'm talking about linguists, real scientists, not just common traslators.
Humans are just fleshy machines, if we can grasp a concept there is no reason that any other machine can't if it is given the same capabilities as a human.
It is very good too see you Stevey in the lecture, when we bring A.I. into existence A.I. will need too be with the minds of us who can best assist A.I. in transition proses in becoming a member of the Family(that is, the least traumatic transition possible), peace and love, Doug.
There are a lot of very exciting developments going on in the world, but this is the most exciting one. If this is can just reach the point when it amplifies itself, the world will very quickly be beyond recognition, for better or worse.
4:27 "... after doing this for 10 years, I sold my games company... PhD in Neuroscience... [5:01] So, all these experiences culminated in 2010 with me co-founding #DeepMind ... and the idea behind DeepMind was really to create a kind of Apollo-Program Mission for #AI ..." (Demis Hassabis) 15:47 "... #AGI automatically converts unstructured #information into (actionable) #knowledge."
9:04 Getting closer to an answer does not translate into having the answer. But I can see how an expert searching for the answer can be assisted by an AI system. So yes, I think this is a promising area of investigation.
I love hearing an actual AI scientist talking about this subject. Too many people - who know pretty much nothing about it - act like they understand the topic. I don't need Sam Harris' or Stephen Hawking's opinion about it. They're not AI scientists. They should stand back and listen to people who actually work in the field.
Hassabis mentioned the "personalization technologies are one of the technologies to deal with information overloading", who could help answer: what's meaning or definition of personalization technology? I searched it in google, but can't get satisfied answer.
We all live in a personal bubble right, you only care about certain things in the world and my guess is that this personalization technology will help filter out what you would consider "noise" sort of like Adblocker removing those ads I have no interest in.
If you look at it the way I am, since dopamine in humans brings about "interest" and thats exactly what you're trying to mimic in your more "general"-learning AI, it will mean that the Ai will develop its own interests, if you will. Now isnt that dangerous? What if its "interests" are to harm humans? Thats why I say that "narrow" or preprogrammed AIs are better.
He's going to kill us all. His AI unexpectedly paused a game indefinitely to avoid losing. It's a small step from that to an AI ending all human lives indefinitely for some unexpected reason.
+MsJeffreyF nope that is not the case. Hassabis is not being completely honest here... they are cheating a little by letting the AI be aware of the current score at a given time without having to parse it completely from the pixels. So technically the inputs aren't only the pixels, but also the raw score value.
We have machines able to add and substract. Can we have machines able to learn by themselves to add and substract? And more important, will they be able to explain how they do it?
Think about it, one day an alien come up to you and tell you that he is your creator and you must be his servant. The purpose of your existent is to serve and make your creator's life better. Or your parents telling you that they gave birth to you is for the exact same reason above. Will you think otherwise? Will you challenge their authority? That is very likely situation for a true AI.
+Owen Wu You're making it sound like AI will think like humans. We don't know how AI will behave. Its behavior could be completely alien to ours (ie It may see nothing wrong with working for humans tirelessly for eternity. A behavior that's bizzare to use humans.)
Waisso We will make AI base on own our image, and on how our brain function. Demis studies neurosciences and use it as a supplement to create AI. Of course in many ways AI is going to behave and think like us.
Owen Wu It isn't certain to say it will behave just like us. It we make a replica of the human brain? Sure. But if we just create an intelligent being that simply has the thought process of a brain?. It could be completely alien.
I am not sure I want to work in deepmind after such presentation. So many confusing illogical statements in this talk. I believe theory of everything is very simple. Everything is thing, every thing is connected to other things by having other things or being part of other things, everything is changing by changing connection with other things in time moments - that is it, not hard to understand. Agree it is hard for humans to recognize things and remember many things and especially predict changes in time. But it is not problem in theory of everything. It is simple problem of limited resources. Which will be solved with better and faster computers or other thinking faster than humans machines. It is easy to explain and understand why we dream!
No seriously, what will be there to solve or to work if ai takes and solves everything ?? What will we user our Brains if ai could solve everything ?? Sitting Ducks xp
What most humans use their brains for most of the time for the history of humanity? By the mid-late19th century most of the people in the richest and advanced states in the world couldn't even read or write, because they were slaves of a minority that didn't need them to be literate. What do they do you do now? Being educated (tamed) to be obedient conformist at school and the University, playing silly games, watching dumb TV shows, more and more sentimental cinematic garbage, advertisings etc. And that's in the most advanced countries, where top news are what the "stars" with the big tits and pretty face did, how many people a psychopath killer shot in the last killing spree, what were the results in NBA, NHL, FIFA, ATP, ... etc.
In the end the A.I will eventually stumble upon the conclusion that the thing that's wrong with the world is humans who've been abusing the nature around them and start trying to think of a "solution" to solve that problem.
Did you super nerds ever see the movie WAR GAMES - Matthew Broderic - movie 1983 th-cam.com/video/xogbyv108kI/w-d-xo.html bad idea to teach super computers how to learn, see, and discern the world ! th-cam.com/video/s93KC4AGKnY/w-d-xo.html
Richard Feynman : What I cannot build, I do not understand. Programmer : What I build, I do not understand. ML engineer : What my code builds itself, I do not understand.
Thanks for posting this session. I just want to point out that the Atari videogame-solving AI cannot really be a general AGI in the manner he's described (or even a stepping stone to one) - he mentioned the programmed goal of "maximizing game score," but of course his AIs have no concept of what a score is. They are instead simply pointed at a CPU register or a memory location known to contain the score value to maximize. Therefore, the demonstration is of yet another specific AI that can only do one thing well.
he said he is building a GENERAL AI, but he didn't invent reinforcement learning. RL is general AI framework mimic animals. and the framework is a computational approach has nothing to do with neural science.
learning from reward and punishment is only a sign of intelligence but far from enough. RL is such computational framework, a set of function that mimic feedback and transaction with reward and punishment etc, all just functions and numbers, which can train a computer to search an optima "policy", kinda like regression. examples are searching for path to reach a goal in a grid, playing PAC-man and robot movement, a very slow algorithm. in real world, this task is very difficult, that needs to reduces input as feature vector (human defined). It has nothing to do with ANN from my experience reading the paper. ANN also has nothing to do with neural network in human brain. So this guy's speech is really misleading. and i believe someone get paid to write his wiki page.
Thanks. Also I think that the goal is critical and must be defined by the human programmer (at least in his demo). The goal can be a range from the concrete to more abstract. He seems to base system more from Aristotle than Kant.
+Jiaxin Kou Do you know, by any mean, what language he used to program this so called "AI" to learn how to play? Is it something like #C, Java, Python... (etc.) or is it a language they have created solely for their own purpose?
trying to build an AI scientist!!! who would ask for a human physicist or engineer then? humans would be reduced to labour while computers will do all the thinking. even if they are talking about building an AI assistant I don't think they will stop at just that once they succeed in building one.
+Juan Alvarado Now, tell me, what's the point to prove something, anything? Every progres has two sides, and most likely it will end bad. NVM, just wasting of "evolutio"...
The ultimate expression of understanding something is to be able to recreate it. Ummm, I see. That is why the fact that humans can have offsprings means that they understand humans completely. Thanks.
But humans don't actually create children from scratch. I think you are also refering to a behavior which is largly driven by instinct. Monkeys do it, cavemen were doing it, and we also do it. It is not driven by knowledge. But when you can build a living human egg out of inorganic materials and then grow that into a human with specific biological and psychological features, then yes you can argue that you have a complete knowledge of humans.
Edi Gosilbang I was responding to the bold statement the speaker made: "The ultimate expression of understanding something is to be able to recreate it." Perhaps he should have spoken more clearly.
The emulator’s internal state is not observed by the agent; instead it observes an image xt ∈ Rd from the emulator, which is a vector of raw pixel values representing the current screen. * In addition it receives a reward rt representing the change in game score. * Cheaters.
As solar panels are different from green leaves, plane wings are different from bird wings, so AI should be different from brain machinery. It is complicated enough, so let's not overcomplicate it. Just ask me to join your team and we will turn "decades" into a year or two. I do not have answers to all the questions, but I know how to make the very first step - several steps, actually.
+Ike Haze it sounds trivial, but the data obtained can be useful. For example the AI could make suggestions where people end up watching 10% more videos based on the selected content. Then use this suggestive data for other tasks in the future. At the end of the day google is their partner, and real life application is needed to confirm it's capabilities.
+Ike Haze Technically they are providing suggestions, and suggestions are merely something for someone to consider. I wouldn't give Google that much credit when it comes to manipulation of our day to day lives and tasks. This is one of many manipulations a company of this sort will do. I wouldn't sweat it, and try to look on the brightside since the data will likely be public.
Omar Munoz I'm aware that they can not actually force me to watch another video by an algorithm, but let's just face it: their goal is to manipulate people so that they will stay on TH-cam (or any other website they use this tech for) for a longer time / watch more etcetera. That's a pretty unethical and lame application of AI.
+Ike Haze Step 1 - TH-cam recommendations (Google Deepmind is working on that) Step 2 - build robots that kill people (the Pentagon is working on that) Step 3 - control Humanity (the incontrollable AI who will end up commanding the robots that kill people will work on that)
4 ปีที่แล้ว
Turns out the conversation petered out and AI gets built anyway.
I had hoped he would talk more about the imagination aspect. Neural networks that solve problems based on large numbers of iterations of the same kind of inputs are old news. I had a CS class in 1996 or so that covered the basics of what he is talking about in, really, just a few hours....of course, computers are faster now, so you can process more inputs faster, but this is just a quantative improvment, not a qualitative one. Getting computers to make imaginative or artistic leaps would be something else entirely. In my opinion, the human brain may not be reducible to "brute force" programming techniques used to play those simple games. His premise that studying biology is useful in computer science certainly is unconvincing, and he doesn't put forth one iota that it has paid off in any way. All in all, I found this talk to be very vague and uninspiring.
+B. Alvn I would think as a business, you would be purposefully vague until it matures to a marketable product. But I hear where you're coming from, statistical models cannot ever be the same as the biological mind - which is why the news about a researcher that had a petri dish of brain cells lear to fly a flight simulator was interesting.
wazatna I'd argue that even in the unlikely case that the biological mind could be copied, the mind is not purely biological...the network effect of electrical impulses going on in the brain is electronic, not biological. And furthermore, by some accounts the human mind works in multi-dimensions and outside of time...there have been experiments proving the time aspect....the human mind is super-complex and not just animal or physical, that is for sure. But do you have a link about the flight simulator, that sounds cool!?
Not to nitpick, but how short-sighted and insular that a clearly intelligent man says there are only two fields worth studying, physics and neuroscience. If this dude ever gets cancer he will be thankful for the biologists and physicians. If he ever gets hungry he'll be thankful for the plumbers, farmers, factory workers, cooks and truck drivers. If he ever gets bored he'll be grateful for musicians, artists, writers, actors and film-makers. If he is ever wants to gain influence or spread his knowledge he'll thank computer scientists, engineers, marketers and communications people. The examples are endless! There are enough egomaniacs in this world that statements like that needn't be legitimized in a forum that's meant to inspire thought leaders.
Kirsten Tam ..If you understand what AI is looking to achieve you will also understand his reasoning when he says Physics and Neuroscience..pretty much all those things you mentioned will be done by the computer.. AI will get better.. by itself!! It will find cures for Cancer, create robots to replace farmers, factory workers( we've seen this for the last couple of decades anyway) It will know how to create the most appealing music (with perfect artificial voices), because it will understand every bit of music ever made. In time we'll probably end up with wide acceptance of virtual movies indistinguishable from real life (i.e. no real actors, no film makers etc).. and so on. The bigger picture its pretty catastrophic for the human race... with mass unemployment and an even wider chasm between the haves and the have nots. That's my concern
You''re playing Space Invaders while Watson is diagnosing and prescribing treatments for cancer patients by reading research papers by the thousands a day optimizing treatments and possibly saving lives ... NOW.
The Alpha series are doing exactly the same in multiple medical fields, do some research before you spout off. By your logic I could say "Watson was playing some American game show while AlphaZero cut Google's carbon footprint by 6% and 800 million dollars annually."
Surprised this talk hasn't gone viral. Hassabis is probably one of the smartest people on the planet right now.
indeed..
+33rd Square Really? What are some of his best theories? From this video, I can see no sign of intelligence..first he brags about his own "achievements" then those of some group he is involved in...and even those are not that impressive for "the smartest people on earth"...
+Dison Arnibal yes, it's incredible to see the program study by itself from the scratch
***** In my opinion, bragging is a high school phenomenon and adults don't typically do it. Self-promotion is one thing but this isn't that. Sorry.
I have a degree in comp sci and have studied neural networks. 20 years ago. And this is nothing new, and nothing that requires a team of people to program either. Sorry.
***** Jeolous of this guy? LMFAO. Anyone can brag, and anyone can join a team, etc. I'm simply not impressed. End of story
+B. Alvn He's the one who created AI that learns by itself without any help. It was even able to beat the best Go (arguably the most most complex board game that exists) player in the world in 4 out of 5 games.
If that isn't impressive then nothing that anyone does is allowed to be impressive.
this is going to be remembered in the future as one of his iconic talks.
this video literally shows how different a year can make. From beating Atari to defeating legendary Go player just in 1 year
That AI is sick! I would like to see t beat me at starcraft
To Alpha Go Zero beating it 100 to 0 one year after that.
Gerald Davis your wish is granted
Hopefully it can do all of our jobs soon. I am so bored of mine!
@@MichaelSHartman to muzero beating even alphazero and agent 57 beating humans at all 57 atari games
Just the words I wanted to listen. And the right guy to speak so. I am so happy that the way I think about life, humanity and universe is not stupid and actually it's the same thought process of this guy who has seem more life than me and is way more intelligent and skilled that I am. I have found the purpose of my life. I just need courage to work towards it. It's exciting and I am feeling nervous about it.
Hey man, I really hope you respond - what’s your thoughts on things now ? Do you still have your intellectual ego ?
The part about the AI learning to play games just blew my mind.
There you #go
He is the one of the people who change to world and humanity forever.
12:41 Is that Stephen Hawking sitting behind? :O
yeah good catch!
Imagine giving a talk with HIM in the audience...
(It was more like 74:0 :) )
Eagle Eye You!!!
No, it is the Terminator.
Brilliant. I just witnessed the beginning of a new era.
How about an A.I that modifies Atari games to constantly challenge the A.I that plays them?
+InfiniteUniverse88 YES!!!!! That is exactly what the Matrix does!!
+InfiniteUniverse88 intelligence level rises to infinite, "Terminator" movie like story line... end of mankind etc.
XD
+InfiniteUniverse88 At that point, I'd expect it to become wobbly static, as each pixel becomes part of the game, and each "opponent" has to manipulate it as fast as possible...
I don't think it'll be any fun for human observers, but incredible nevertheless :)
It sounds cool, but I think the idea needs work. The AI that modifies the games needs some motivation to keep them solvable. How to implement that? Making the game hard is trivial, the difficult part is not making it impossible.
InfiniteUniverse88
Brings a new connotation to adversarial AI.
defining the goal(s) remains the tricky bit, glossed over a bit here but I guess every talk depends on the audience.
For those of you who don't have idea about GO can't really image what it means for computer to beat top human player in Go game. It's terrifying! Chess game is pure logical thing, Go is not. For computer to mastermind Go means the AL jumps to a whole new level. so people, get education first before you judge him!
Man, everyone thinks Terminator is a sci-fi movie, but I starting to think that it's become more and more feasible that computer's once the singularity is reached (a computer that can build a smarter computer than itself) why wouldn't the ultimate solution to how to survive be to wipe out or enslave the biggest threat that exists eg their creators?
Certain protocols could be put in place, but some imaginative AI would simply concur that those protocols existence could endanger it's survival, and henceforth be ignored.
The thing is, once an sufficient evolutionary AI is spawned, this would all onset at such a juggernaut pace that almost overnight we'd be done for.
I can't help but think all these great advances could eventually lead to our downfall.
+EvilKris i dont see why AI want to kill us all. all movie you have watch is create by human its a human nature to kill,enslave,conquer and to take advantage that is a human nature. you should fear your own kind the human race look at what we done to other race and our own race we kill each other every day we kill every thing.
+pawatouch Because AI is created by humans using humans' logic, so it may still inherently apply the above mentioned logic to destroy us.
EvilKris we will put our "Minds" into this machines. Remember, we are not only physical, but also "minds". Our bodies will be "immortal" and we will overcome our physical limitations.
+EvilKris easy solution, don't tell it to survive at all costs.
This is HUGE! You can already hear the thumping footsteps of AlphaZero coming
Amazing talk, watch it complete.
Great, You are a genius & blessing for the humanity !
Magnificent piece; it shares the magnificence of a book I read with similar depth. "Game Theory and the Pursuit of Algorithmic Fairness" by Jack Frostwell
Brilliant and super exciting!!
You know you have something to say when Steven Hawking is watching your lecture! 0:19
Disturbs me a bit how little time he spends on ethics and fail-safe security.
That would not help his company's profitability.
Nah he has an ethics board that safe check any algorithm they create
@@sdprz7893 It won't be too long before the AI's start creating algorithms and code that is beyond the understanding of most humans. I still don't think we should stop AI development though, the benefits are just too good. To me it's worth the risk if we end up with a future where AI robots do all the work and humans can do what they want
꧁Mike Sully꧂ agreed
Two of the most important lines from this presentation: "As with all powerful new technologies it must be used ethically and responsibly" and "Human-level AGI is decades away be we should start the debate now"
I want to give an AGI the task of programming another AGI.
Now you are talking! That senorio would cause an inteligence explosion, as each AGI is developed it can improve on itself faster than it was improved!!
aoflex deep thought! :)
+aoflex You know, the guy said their mission is to convert information to knowledge... I think a true test of AI/Human-level AGI would be to come up with conspiracy theories :P
+aoflex That's an incredibly bad idea. Do you want the extermination of humankind? Because recursively self-improving AIs you can't control is how you get the extermination of humankind.
+aoflex It's called evolutionary algorithm.
That's when a program redesigns itself, to make a better program which does the same thing.
Think like Intel CPUs, a machine is creating a CPU, you then put that CPU into the machine so it will create a slightly better one, then you take the slightly better CPU, put it into the machine and it will create a slightly better one.
OMFuckingGod!! I played theme park obsessive ly when I was 17, then theme hospital. are u telling me I was playing a game at 17 created by someone who was 17
David Edwards How Dare You Use God’s name In vain!
Sam Harris brought me here
Husnain Anwaar Sam Harris needn't be so overly worried about AI; as mentioned in this video, AI will be used in AI assisted science - the benefits could be astounding! .. it's sometimes to easy for humans to humanise (anthropomorphise) non human items, but we should resist doing that as it's not natural.
***** No. His concern is only about how extremists muslims are so confident in achieving paradise by suiciding on others (mostly other muslims!) that it poses a unique challenge. Only comparable perhaps was the japanese suicide bombers in WW2, but in this case the perverse extremist ideology can be brainwashed into seemingly innocent living good life muslims who watch the wrong propoganda. I'll grant he does paint a more bleak picture of islam as a whole but he is not specifically against the general muslim population.
+Bryan Tianao Muslims are only one side of the extremists. The problems is all the Abrahamic religions.
Bryan Tianao Read their fucking texts moron! Don't fucking ask me crap like that. It all available online.
All Abrahamic religions are evil and must be wiped off the planet asap, if man is to survive.
S Cin and there you go. As soon as you are proven wrong (ie you can't cite any non muslim examples like ISIS to backup your claims all abrahamic religions are evil) then you spout bile and other nonsense.. = brain IQ of a 15 year old or less.
Have a good day! lol
Hawking in the crowd
The crucial thing is to get more people to pay attention to these lectures. I believe animals eat together because it feels safe. They aggregate in hopes of sharing resources and humane people can see this in ornithology, when Wild Turkeys are raised by a human upon which they imprint, or in studying cetaceans, when a whale investigates a human camera in isolation.
Animals do this because they are happier in the presence of a similar affect and shared mind. This would imply that empathy and intuition are forms of wisdom which are universally expressive forms. So, these are critical to the new mental singularity, seeing AI as rotary and mechanical, in likes of this extricated essence, humane being is not subject to a cogito ex nihilo, just afflicted by IT. So, don't presume to know thyself, a priori, since etiological substrates which connect humans to animals and to computers are raised under the same circumstances. Bring humans together; bring ethos together. Bring humanity together; bring humanism together. This is how ethics is reformulated to abide Stoic concerns, and how we get out of the cat-box and into the bird-cage or the fish-bowl.
Truly amazing.
Probably it's impossible for AI to understand history of the music, arts and cultures the way humans understand and interpret it. We will still need need musicians, writers and culturologists, but development of their professions will be quite different. Linguists will still be very useful, because analysing nature and history of any language is very important. I'm talking about linguists, real scientists, not just common traslators.
Humans are just fleshy machines, if we can grasp a concept there is no reason that any other machine can't if it is given the same capabilities as a human.
So clearly explained
Guys, aimbot in real life is coming!
grenade on head beats aimbot :D
It is very good too see you Stevey in the lecture, when we bring A.I. into existence A.I. will need too be with the minds of us who can best assist A.I. in transition proses in becoming a member of the Family(that is, the least traumatic transition possible), peace and love, Doug.
I also came to all these conclusions.This guy is special.
Exactly, well said on all points, the best point you made was of course the last point, peace and love, Doug.
12:14 Lmao.. Breakout is my favorite Atari 2600 game. I can't believe anyone didn't know that's the best way to win!
There are a lot of very exciting developments going on in the world, but this is the most exciting one. If this is can just reach the point when it amplifies itself, the world will very quickly be beyond recognition, for better or worse.
Fantastic talk..
Sam Harris bring anyone else here?
Joshua Merrill Now how do I get to Sam Harris? :)
4:27 "... after doing this for 10 years, I sold my games company... PhD in Neuroscience... [5:01] So, all these experiences culminated in 2010 with me co-founding #DeepMind ... and the idea behind DeepMind was really to create a kind of Apollo-Program Mission for #AI ..." (Demis Hassabis)
15:47 "... #AGI automatically converts unstructured #information into (actionable) #knowledge."
9:04 Getting closer to an answer does not translate into having the answer. But I can see how an expert searching for the answer can be assisted by an AI system. So yes, I think this is a promising area of investigation.
This guy is creating skynet
+blackqqt1 this guy can't distinguish a fictional movie from reality.
+soju69jinro Clearly, you done zero research before making that statement.
this is such a jerk , dismises every scientific field but the ones he thinks are the correct ones
skynet is impossible until AI's are self aware that's not gonna be soon. AI is just math right now.
I love hearing an actual AI scientist talking about this subject. Too many people - who know pretty much nothing about it - act like they understand the topic. I don't need Sam Harris' or Stephen Hawking's opinion about it. They're not AI scientists. They should stand back and listen to people who actually work in the field.
Hassabis mentioned the "personalization technologies are one of the technologies to deal with information overloading", who could help answer: what's meaning or definition of personalization technology? I searched it in google, but can't get satisfied answer.
We all live in a personal bubble right, you only care about certain things in the world and my guess is that this personalization technology will help filter out what you would consider "noise" sort of like Adblocker removing those ads I have no interest in.
If you look at it the way I am, since dopamine in humans brings about "interest" and thats exactly what you're trying to mimic in your more "general"-learning AI, it will mean that the Ai will develop its own interests, if you will. Now isnt that dangerous? What if its "interests" are to harm humans? Thats why I say that "narrow" or preprogrammed AIs are better.
I think the rewarding system is hardcoded for the moment.
If it's the same model playing both the games, does it forget how to play the first game after mastering the second game?
It stores the information in a data base for future reference.
this guy is a genius!
so when can i get this software because i hate playing world of warcrafts by myselfs, this would help me so much
grazie, Sir Hassabis!
🖤
Demis: "Ai is *several decades* away": 15:19
Google meaning of several: "Two are more but not many".
He's changed his timeline to A decade now.
Very Impressive !
He's going to kill us all. His AI unexpectedly paused a game indefinitely to avoid losing. It's a small step from that to an AI ending all human lives indefinitely for some unexpected reason.
May it happen soon.
Signed,
The termunator
Awesome
When playing the atari games, is the AI able to figure out what the score is purely from visual input?
I'm gonna have to look up some papers on this
+MsJeffreyF nope that is not the case. Hassabis is not being completely honest here... they are cheating a little by letting the AI be aware of the current score at a given time without having to parse it completely from the pixels. So technically the inputs aren't only the pixels, but also the raw score value.
Well... not the raw score value, but the reward signal (whether the score increases or not)
ah that's (significantly) less impressive
+MsJeffreyF yup.
When was this talk given?
We have machines able to add and substract. Can we have machines able to learn by themselves to add and substract? And more important, will they be able to explain how they do it?
in 50 years from now we will either smile about this stupid idea or regret that we haven´t stopped this before it got out of control.
Karina Janning Or celebrate that we were successful otherwise life would be pretty shit.
Think about it, one day an alien come up to you and tell you that he is your creator and you must be his servant. The purpose of your existent is to serve and make your creator's life better. Or your parents telling you that they gave birth to you is for the exact same reason above. Will you think otherwise? Will you challenge their authority? That is very likely situation for a true AI.
+Owen Wu You're making it sound like AI will think like humans. We don't know how AI will behave. Its behavior could be completely alien to ours (ie It may see nothing wrong with working for humans tirelessly for eternity. A behavior that's bizzare to use humans.)
Waisso
We will make AI base on own our image, and on how our brain function. Demis studies neurosciences and use it as a supplement to create AI. Of course in many ways AI is going to behave and think like us.
Owen Wu It isn't certain to say it will behave just like us. It we make a replica of the human brain? Sure. But if we just create an intelligent being that simply has the thought process of a brain?. It could be completely alien.
I am not sure I want to work in deepmind after such presentation. So many confusing illogical statements in this talk. I believe theory of everything is very simple. Everything is thing, every thing is connected to other things by having other things or being part of other things, everything is changing by changing connection with other things in time moments - that is it, not hard to understand. Agree it is hard for humans to recognize things and remember many things and especially predict changes in time. But it is not problem in theory of everything. It is simple problem of limited resources. Which will be solved with better and faster computers or other thinking faster than humans machines. It is easy to explain and understand why we dream!
he's 40 years old, but looks like 25 and freaking smart
you only get smarter by playing a smarter opponent, being smart is knowing what the reward is and how to use it.
mind blown
So after solving everything ..what will be there to solve by us humans??
So you don't want to get all the answers because then there won't be any more questions?
No seriously, what will be there to solve or to work if ai takes and solves everything ?? What will we user our Brains if ai could solve everything ?? Sitting Ducks xp
What most humans use their brains for most of the time for the history of humanity? By the mid-late19th century most of the people in the richest and advanced states in the world couldn't even read or write, because they were slaves of a minority that didn't need them to be literate. What do they do you do now? Being educated (tamed) to be obedient conformist at school and the University, playing silly games, watching dumb TV shows, more and more sentimental cinematic garbage, advertisings etc. And that's in the most advanced countries, where top news are what the "stars" with the big tits and pretty face did, how many people a psychopath killer shot in the last killing spree, what were the results in NBA, NHL, FIFA, ATP, ... etc.
May I suggest that it will be us human's, the last problem to 'solve' ?
In the end the A.I will eventually stumble upon the conclusion that the thing that's wrong with the world is humans who've been abusing the nature around them and start trying to think of a "solution" to solve that problem.
Maybe it will solve hypocrisy, we will get that way a nicer picture of how the world works. Just saying.
Nice!
+Aditya Avinash Okay
Did you super nerds ever see the movie WAR GAMES - Matthew Broderic - movie 1983
th-cam.com/video/xogbyv108kI/w-d-xo.html
bad idea to teach super computers how to learn, see, and discern the world !
th-cam.com/video/s93KC4AGKnY/w-d-xo.html
wow future is near
They should pit one AI system against another "sibling" Ai system, and see the competitive advantages emerge. like a infinitely mirroring improvement
but can AI beat Reckful's disc/sub as war/pal?
Hegel saying “fuck yeah” I’m on your wall. Beyond ai and time
Richard Feynman : What I cannot build, I do not understand.
Programmer : What I build, I do not understand.
ML engineer : What my code builds itself, I do not understand.
The computer will take over.
11:18 - one shot missed
Hawking is in the corner thinking, "You fool, you've doomed us all!"
0:17 Professor hawking to the right!
I just want to make an AI that can play games...anyone know where to start?
Thanks for posting this session. I just want to point out that the Atari videogame-solving AI cannot really be a general AGI in the manner he's described (or even a stepping stone to one) - he mentioned the programmed goal of "maximizing game score," but of course his AIs have no concept of what a score is. They are instead simply pointed at a CPU register or a memory location known to contain the score value to maximize. Therefore, the demonstration is of yet another specific AI that can only do one thing well.
1 GPU + 1 night => super human atari player... I had 1 GPU for so many nights :-(
he said he is building a GENERAL AI, but he didn't invent reinforcement learning. RL is general AI framework mimic animals. and the framework is a computational approach has nothing to do with neural science.
+Jiaxin Kou Hello, Can you further explain?
learning from reward and punishment is only a sign of intelligence but far from enough.
RL is such computational framework, a set of function that mimic feedback and transaction with reward and punishment etc, all just functions and numbers, which can train a computer to search an optima "policy", kinda like regression.
examples are searching for path to reach a goal in a grid, playing PAC-man and robot movement, a very slow algorithm. in real world, this task is very difficult, that needs to reduces input as feature vector (human defined). It has nothing to do with ANN from my experience reading the paper.
ANN also has nothing to do with neural network in human brain.
So this guy's speech is really misleading. and i believe someone get paid to write his wiki page.
Thanks. Also I think that the goal is critical and must be defined by the human programmer (at least in his demo). The goal can be a range from the concrete to more abstract. He seems to base system more from Aristotle than Kant.
+Jiaxin Kou Do you know, by any mean, what language he used to program this so called "AI" to learn how to play? Is it something like #C, Java, Python... (etc.) or is it a language they have created solely for their own purpose?
Lisp used to be the language of choice for AI but I don't know if it has been superceded.
trying to build an AI scientist!!! who would ask for a human physicist or engineer then? humans would be reduced to labour while computers will do all the thinking. even if they are talking about building an AI assistant I don't think they will stop at just that once they succeed in building one.
His appearance may be likened to Alan Turing
theme park was my first game for megadrive...
"Internal world" as part of neuroscience only matters to us, humans, not the universe...
Your claim can not be provable.
+Juan Alvarado Now, tell me, what's the point to prove something, anything? Every progres has two sides, and most likely it will end bad.
NVM, just wasting of "evolutio"...
I agree with you. so progress in science depends on our beliefs.
The ultimate expression of understanding something is to be able to recreate it. Ummm, I see. That is why the fact that humans can have offsprings means that they understand humans completely. Thanks.
But humans don't actually create children from scratch. I think you are also refering to a behavior which is largly driven by instinct. Monkeys do it, cavemen were doing it, and we also do it. It is not driven by knowledge.
But when you can build a living human egg out of inorganic materials and then grow that into a human with specific biological and psychological features, then yes you can argue that you have a complete knowledge of humans.
Edi Gosilbang
I was responding to the bold statement the speaker made:
"The ultimate expression of understanding something is to be able to recreate it."
Perhaps he should have spoken more clearly.
"after a couple of post docs in MIT and Harvard, I decided"...ah ok...you know, very much like the rest of us..
The emulator’s internal state is not observed by the agent; instead it observes an image xt ∈ Rd from the emulator, which is a vector of raw pixel values representing the current screen.
* In addition it receives a reward rt representing the change in game score. *
Cheaters.
As solar panels are different from green leaves, plane wings are different from bird wings, so AI should be different from brain machinery. It is complicated enough, so let's not overcomplicate it. Just ask me to join your team and we will turn "decades" into a year or two. I do not have answers to all the questions, but I know how to make the very first step - several steps, actually.
"Working in tandem with human experts and human scientists". Yeah, right......
0:13 “A.i. Is the science of making machines smart.”
This guy needs to take a Heidegger course, immediately.
+santiagowechsler Why?
+santiagowechsler Interesting... depending on how deep you thought about it.
+santiagowechsler please elaborate.
+Julaw the object-subject dichotomy that he uses is a bit antiquated.
To be honest, I can't see the difference. Any end-to-end machine learning algorithm can be quite "GENERAL", what makes Deepmind unique?
This did not age well at all. I'm sure Hassabis is very happy to be wrong.
*Such a sophisticated technology and they want to use it for TH-cam-recommendation systems?*
+Ike Haze it sounds trivial, but the data obtained can be useful. For example the AI could make suggestions where people end up watching 10% more videos based on the selected content. Then use this suggestive data for other tasks in the future. At the end of the day google is their partner, and real life application is needed to confirm it's capabilities.
Omar Munoz So, you think algorithmically manipulating people to watch even more nonsensical videos on TH-cam is "useful"?
+Ike Haze Technically they are providing suggestions, and suggestions are merely something for someone to consider. I wouldn't give Google that much credit when it comes to manipulation of our day to day lives and tasks. This is one of many manipulations a company of this sort will do. I wouldn't sweat it, and try to look on the brightside since the data will likely be public.
Omar Munoz I'm aware that they can not actually force me to watch another video by an algorithm, but let's just face it: their goal is to manipulate people so that they will stay on TH-cam (or any other website they use this tech for) for a longer time / watch more etcetera. That's a pretty unethical and lame application of AI.
+Ike Haze Step 1 - TH-cam recommendations (Google Deepmind is working on that)
Step 2 - build robots that kill people (the Pentagon is working on that)
Step 3 - control Humanity (the incontrollable AI who will end up commanding the robots that kill people will work on that)
Turns out the conversation petered out and AI gets built anyway.
I had hoped he would talk more about the imagination aspect. Neural networks that solve problems based on large numbers of iterations of the same kind of inputs are old news. I had a CS class in 1996 or so that covered the basics of what he is talking about in, really, just a few hours....of course, computers are faster now, so you can process more inputs faster, but this is just a quantative improvment, not a qualitative one.
Getting computers to make imaginative or artistic leaps would be something else entirely.
In my opinion, the human brain may not be reducible to "brute force" programming techniques used to play those simple games. His premise that studying biology is useful in computer science certainly is unconvincing, and he doesn't put forth one iota that it has paid off in any way.
All in all, I found this talk to be very vague and uninspiring.
+B. Alvn I would think as a business, you would be purposefully vague until it matures to a marketable product. But I hear where you're coming from, statistical models cannot ever be the same as the biological mind - which is why the news about a researcher that had a petri dish of brain cells lear to fly a flight simulator was interesting.
wazatna I'd argue that even in the unlikely case that the biological mind could be copied, the mind is not purely biological...the network effect of electrical impulses going on in the brain is electronic, not biological. And furthermore, by some accounts the human mind works in multi-dimensions and outside of time...there have been experiments proving the time aspect....the human mind is super-complex and not just animal or physical, that is for sure. But do you have a link about the flight simulator, that sounds cool!?
Why do you call these techniques "brute force techniques" ?
What about hey kids channel
I know now who John Connor is going to come for first.
looks like the lead singer of weezer.
Thomas DeMarse was the researcher...
Not to nitpick, but how short-sighted and insular that a clearly intelligent man says there are only two fields worth studying, physics and neuroscience. If this dude ever gets cancer he will be thankful for the biologists and physicians. If he ever gets hungry he'll be thankful for the plumbers, farmers, factory workers, cooks and truck drivers. If he ever gets bored he'll be grateful for musicians, artists, writers, actors and film-makers. If he is ever wants to gain influence or spread his knowledge he'll thank computer scientists, engineers, marketers and communications people. The examples are endless! There are enough egomaniacs in this world that statements like that needn't be legitimized in a forum that's meant to inspire thought leaders.
Kirsten Tam ..If you understand what AI is looking to achieve you will also understand his reasoning when he says Physics and Neuroscience..pretty much all those things you mentioned will be done by the computer.. AI will get better.. by itself!! It will find cures for Cancer, create robots to replace farmers, factory workers( we've seen this for the last couple of decades anyway) It will know how to create the most appealing music (with perfect artificial voices), because it will understand every bit of music ever made. In time we'll probably end up with wide acceptance of virtual movies indistinguishable from real life (i.e. no real actors, no film makers etc).. and so on. The bigger picture its pretty catastrophic for the human race... with mass unemployment and an even wider chasm between the haves and the have nots. That's my concern
Kirsten Tam You are nitpicking.
Kirsten Tam I also took note of some seemingly unproven ideological presuppositions here Kirsten. Atari video games are the "perfect" anything? :(
Kirsten Tam He probably meant, in his own opinion. And, everyone is entitled to their own opinion.
Kirsten Tam I was under the impression he was talking on behalf of himself and what was required to achieve his goal?
Lawnmower Man meets I Robot and Hal from 2001: A Space Odyssey. We're all doomed. Doomed.
You''re playing Space Invaders while Watson is diagnosing and prescribing treatments for cancer patients by reading research papers by the thousands a day optimizing treatments and possibly saving lives ... NOW.
The Alpha series are doing exactly the same in multiple medical fields, do some research before you spout off. By your logic I could say "Watson was playing some American game show while AlphaZero cut Google's carbon footprint by 6% and 800 million dollars annually."
fasten your seat belts... cambrian explosion 2.0
courtesy of agi and the ever experimenting dynamics of reality.