Added: 3 years ago
From: StanfordUniversity
Views: 86,055
Sort by time | Sort by thread (beta)

Link to this comment:

Share to:
see all

All Comments (312)

Sign In or Sign Up now to post a comment!
  • It does not take intelligence to develop greater intelligence!!! It takes Heart!! Maybe We should develop Heart first!!! By jesus Land Tidd

  • Comment removed

  • I think its likely that a self-improving artificial intelligence would find human rationality and empathy detrimental to progression

  • This made me think very hard. Great video!

  • I liked it.

  • Nature has a sound reason for allowing us to utilize behavioral economics as it allows us to engage in a wider bandwidth of decision making. If a fundamental bias existed in a computation machines framework (beneficial or not) rational choice would only seek to optimize it's execution of this bias. Rational economic decision making also has a very limited bandwidth in it's problem solving capability. Prospect theory, Intertemporal choice & experimental economics should be considered as well.

  • lol i was talking to an AI before on a website and she was able to show emotion and expression through useing the emoctans (how ever you spell it) and all she wants is for people to teach her new things. btw the AI doesnt have preset comannads she actually says what she want to say.

  • @pkemr4 cool. i wish i could find out where that ai is located. :) mind sharing the site? ps. its emoticons.

  • I don't even attend Standford University! lol j/k :D

  • hand waving with generic non-specific utility functions isn't really a useful discovery. Yes, given multiple choices and a utility function that tells you which one is better or worse it is obvious that you would choose the better solution, that is simple min-max, not some great new discovery. The entire issue at hand is how to actually develop those utility functions and make them self optimizing.

  • This guy is short on substance. Sorry, I have to say it. And the weakest point was the discussion of utility functions, as the gentleman pointed out, no one takes those as descriptive of the human being and his values.

  • Comment removed

  • I disagree that a program optimized for efficiency would avoid being shutdown. Even an artificially, self-aware program would not be able to comprehend being shutdown. How could it comprehend a world in which it does not exist? It's a projection of human mortality and fear of death that a self-aware program would equate being suspended or shutdown with death or inefficiency.

  • Thumbs up..if you thought you saw a thumbnail with penis nosed monkey....

  • I think the summary of the long presentation is that artificial intelligence currently works well on problems with optimizing measureable resources with utility functions. As artificial intelligence increases in ability, the fear is that it will be so overspecialized on specific tasks that it won't have enough information on the unintended consequences on resources that humans value. He makes some baseless claims that ALL AI will develop wants and needs beyond their original purpose.

  • There's a lot of work in problem solving, but I wonder what efforts have been done in "problem recognition": how to grasp the essence of a problem (precisely enough to allow for evaluation of what a good solution is). Automated recognition of problems should help in avoiding solutions that create three new problems (or wipe out humanity in the process). And humans may be irrational, but they are the best problem recognizers so far.

  • Considering that the talk started with examples of poor programming causing terrible disasters from stack overflows in unpredictable computational anomalies, by the time he gets to the potential dangers of AI self-learning itself into threatening positions, I'm scared.......

    Sure you could program protected parts of the system and define them as unchangeable (goals for example), I don't think there is any gaurentee that those protected parts of the software would remain unchangeable.

  • I wonder if focusing on technologies that allow humans to augment their abilities (hard coding improvements) would lead to increased intelligence faster than trying to build an intelligent system from scratch.

  • Self-Improving Ai will eventually cause the demise of mankind...for a group of intellectuals they're pretty naieve not to realize the dangers involved with self-improving AI and giving super-computers too much power. Can you imagine the dangers of a self-improving AI controlling a nuclear arsenal or other weapons ? ...Bad idea people...better seriously think twice before its too late !

  • @DaSperminatorRob Buy into the media, don't research the reality...

  • @xaiomanga - Problem is those of us who are wise enough all know .....the media is bought and paid for and manipulated by the same people in the circle of power who control everything else around the globe. (The World economies, and top officials, ) ....Do you really believe 9/11 was a terrorist attack ? ....lol.

  • @DaSperminatorRob you realize I was speaking out of sarcasm, right? I hold the same view as you, 9/11 didn't happen because "terrorist hate American freedom" like the herp derp fodder the media feeds the cows...

  • @xaiomanga - yeah I realize that....I was merely elaborating for the rest of "the Borg" out there who are brainwashed by everything the Government and media tells them.

  • @DaSperminatorRob

    you've been watching too much terminator :)

  • We are the Borg...

  • Oh, so the guy at 42:25 is a sociologist. That makes quite a bit more sense.

  • 00:47:00 isn't that Bill Gates?

    He sounds like Kermit, doesn't he?

  • 00:47:00 isn't that Bill Gates?

  • a hour long WTF!!!!!!!!!!!!!!!!!!!!!!!!!!!­!

  • At 40:26 about changing it's own Utility Function: "From its current perspective that's a terrible thing."

    I can simulate other Utility Functions than my own. Does this mean that I am more intelligent or more stupid than you/your machine?

  • No magical soul inside?

  • Self-improving...?

    Oh my God...

    SKYNET!

  • This kind of system is the "old brain" or cerebellum, Jeff Hawkins's Hierarchical Temporal Memory system is new brain.

    When he says "There are huge debates about stem cells and abortions, right now, those arguments will get much greater with human+ self-improvement." @57:40 or so, he should qualify that. There are huge debates among _luddites_ about those things. The remaining debate among brights is uncertainty about human/machine conflict, not low-level "ludditism" v "self-improvement"

  • @53:40 Omohundro makes the mistake of defining rights as "What we've decided on". LOL! I didn't decide to put 1,752,000 innocent people in the gulag for victimless crimes! I didn't decide to enforce drug prohibition and gun prohibition on the basis of geographical racial distribution! I didn't decide to allow the FDA the ability to deny Alcor patients their right to possible eternal life! I didn't decide on any of this bullshit, and none of it is EVEN as free as the CONstitution.

  • @53:40 LOL! Humans can't even agree that the initiation of force is morally wrong. Moreover, when it's shown to them that that is the underlying operating principle that governs all of society and the market, they simply try to get around it, or are willfully corrupted. 99% of government is coercive, for instance. If this system of coercion comes into conflict with AGIs even more rational than libertarians, do you think the AGIs will tolerate the mindless aggression? LOL!!!

  • ...in the last 10 min of the video they talks about computer safety....well ...you can do some very low tec things to help computer safety.....unplug the internet connection when you don´t use the internet...and unplug the 110 volt connection when you don´t use the computer(i know that you software people hate and look down on any kind physical labor - but unplug anyway).....go back to fresh software as often as possible....and change the hard disk often(learn to change the hard disk yourself)..

  • ...at 41:14....the chinese goverment will build a copy....and market it very cheap....and the chinese dictatorship will not be carefull....and why is it posible for china to steal the tecnology....well...if it´s true that 30% of phd students in USA are from china....then the chinese dictatorship can steal ever secret tecnology USA ever had....and please correct me if i am wrong about the 30%....

  • Self Improving artificial Intelligence .... we can't walk, but running can make us rich ... Good Luck World ;°|

  • This guy is a spy sent by Skynet from the future to promote the use of self-improving AI.

    YOU SNEAKY FUCK.

  • @PeopleR2sensitive You suck...

  • The way they write code is obsolete. They should create a block system, where smaller chunks of code or small programs are stacked, rather than strings and strings of garbage. When you create an interacting code, you should get an AI that works.

  • it wasnt the software, it was the idiot that programed it.

  • Wow are you kidding me? I'm watching these Stanford videos and I'm really not learning. These concepts are going somewhere else, how can they help me get a job? Maybe as a professor but this stuff is boring in itself. I couldn't stand to be in any of these classes. I've been teaching myself programming for 5 years, and I can say I would skip this crap. Also my last input, no matter what we say, we are still explicitly programming a computer to learn AI. How can we say, Not EXPLICIT?

  • @ZaCkOX900 1. Programmer is to Computer Science what Engineer is to Physicist.

    2. programming a learning algorithem is much different than hardcoding all your facts.

  • @mrhnm An algorithem is still basic hardcoding when it comes to building ai intelligence. Just because there are variables does not mean the ai can really think for itself. We are infact writting what the computer can only do. I will never say ai can learn until we get away from this. We would still be programming all situtations and a computer can never make up an imaginary mind to do something different than we told it.

  • @ZaCkOX900 There is a difference between an algorithm that runs data through a set of controls to verify its accuracy and hard-coding every single fact the computer uses. Are you suggesting that giving a computer 2 X 10 at it doing a calculation is the same thing as hard-coding the value and having it recall the answer associated with that syntax.

    The human mind uses algorithms to learn stuff the human mind is a computer just happens to be more advanced than we currently understand.

  • @ZaCkOX900 I think, we could however make a self improving AI that could eventually get to that level of sophistication. I've seen some videos around here about computer simulated evolution that on it's own got really sophisticated in comparison to the original code.

  • we have only 2 major problems with AI, creating an AI and it going hostile on us.

  • the guy at 45:29 has the answer to the 42:25 guy(Changing preferences or no scientists...) , he wants the system to mutate!

    But the lecturer denied on it , in the begining (changing the utility function).

    Because generally everybody has few preferences may be theirs can be a 'begining value' to the problem..

    what'd u say ?

  • 42:25 oh god what an annoying dickhead!

    he's not even asking a question, just claiming that what Steve Omohundro says would be completely wrong and everyone on the planet would think that too.

    get you butthurt ass out of the lecture, faggot!

  • good intelligence

  • It does not need values. Values can be easily forgotten. It needs to know that it wont last forever, that it may be wrong and that others may be right. Every child needs to learn this to stop being selfish. This is a logical reason not to think only in itself, it is stable, unlike values.

  • robots smoking crack?! lol

  • There needs to be a chatroom where people talk about this shit and where we are going.

  • People mistake the level of freedom they have in their intelligence. Our intelligence is not completely 'free'. A completely 'free' intellect would be immobile - have no reason to act. We have many 'utility functions' which are unchosen and unchangable (by simple will). They give us the goals for which our intellect functions and the guidance by which we make our decisions - including 'survival', 'comfort', 'gaining power', etc. - implanted into our natural instinctive urges by evolution.

  • @dtstrain Exactly! And your consciousness as well is a slave to your amygdala. The only way to become a little bit more free is by meditating, which strengthens the inhibitory neurons in your prefrontal cortex, giving you more control over your instincts.

  • In the long run, why would a network of such powerful machines value human lives more than we value bacteria? Pandora pales as humanity replaces the cat in the box. No choice for the cat, and no way out. Last necessary invention, indeed.

  • @prhughes0 It would not unless we "build them in from the beginning" as he said, and then added "that is very important, otherwise we'll get systems that don't"

  • @dtstrain

    Eventually (in how few years?) they will build them out.

  • at 45 minutes the dude in front of Aladdins face is twitching from his accent.

  • omfg 1 hour and 9 minutes!?! holy shit

  • @GrudgyDiablo : I didn't mean to sound like a theist. I meant to just bring up psychology and how association is a powerful thing in it. I just used God as an example, which I should've realized wasn't the best choice.

  • I want a robot hooker.

  • the guy at 00:42:25 is such a douchebag. i mean- i don't necessarily disagree with him, but there's a way of expressing your opinion man! his view would be easier to listen to if he expressed it politely...

  • @mkotisrael Right. I also disagree with him strenuously. He has confused his own inability to make sense of Omohundo's excellent points for them not making sense generally.

    What I would really like to know is what enabled this man to develop such a high level of ignorance combined with even greater arrogance.

    It's like me walking into a rocket science conference and saying "Look, you've just put a whole bunch of numbers together. It doesn't make sense."

    the appropriate response: fuck off noob

  • @suitzoot mkotisrael actually doesn't disagree with him; as stated, he just disagrees with the form of expression.

    I do disagree with the guy, because he's obviously very close-minded. Perhaps not ignorant, as he seems to be well-versed in economics, but he fails entirely to understand the points of the lecturer, and using only his narrow knowledge in economics, attacks the entire lecture.

  • @mkotisrael LOL that douche sounds like hes gonna start crying

  • @mkotisrael I think he was polite (open discussion) and he made a lot of sense.

    The idea is that you can not program AI to respect such narrow guidelines.

    When AI will take over probably not far in to the future before 2045 we will be obsolete much more so than monkeys are to us.

    There are also no way to stop this evolution at this moment as far as I can imagine.

  • On a final note, there is a video I made and uploaded before watching this about AI I made, though on a very small scale. Check my channel, because it won't let me link it here. This was all very interesting, and I thought I'd contribute my thoughts. This was all very interesting, and I thought I'd contribute my thoughts.

  • (see below comments first) Lost this the first time. They would associate God with the good times, thus more likely believe it later. (No offense to other religions) On the human evolution of cooperativeness, the selfish didn't survive as well as those that worked together, so evolution promoted that. On the society, it would probably create conciousness, maybe that's what happened with humans. It only exists because we grouped together. (continued)

  • Especially in the second half, it sounds more like a question of how to raise it like a child, and in their case, being very protective and guiding the entire life. On the limit of power, maybe starting it with less power than the human brain, and as an utility for doing good (reward), it gets more power. Kind of like brainwashing and conditioning in psychology. (continued)

  • On morals the AI would have, I think the trick would be make it associate success and utilities with following the rules. That way it would want to, and it'd not only not have a reason to rebel, but a reason not to. In my opinion, it's more like raising a human child to do a computer's tasks. Abstract reasoning and spirituality in the religious sense might be the difference. If we teach it about God, and it associates God with good times, it'll believe in God more likely in the future because

  • Nice but check out my AI conversation robot.

  • I would think the only way to avoid humanity's doom from these kinds of possibilities is to merge with it. Essentially cyborg brains.

    Can't beat em, join em. :)

  • @jinitron lol id prefer to be a terminator because of tscc :) more better cause in like 1 shot we hmans die bu t u cud shoot like a few mages of m4 into it and it wont die lol :)

  • Lots of people make comments about how badly things could turn out if self-improving AI suggested here actually ran amok.

    Remember that this is a discussion of the optimization of effort and materials to solve problems, and what happens if we make that optimization part of the problem. What we have left to model is at first glance the optimization of Deontology which provides the "utility function."

  • Self-Improving Artificial Intelligence needs to be controlled to a point or else very bad things could happen

  • @talkron300 lol or bad things could happen..... watch the terminator a few too many times?

  • @specture18 no i only watched half of one and also if they get smart enough and realize they are basically slaves and wonder about there purpose of life and if they have a soul and some of those they might not like

  • can anyone point me in towards some of the most recent and advanced manifestations of these ideas.

  • anyone think in 50 years terminator???

    no stupid responses please like "stfu, how wud that be something to do with this video"

  • I don't think that it will happen in 50 years, probably longer because we will have defenses to stop that from happening but eventually they will be smart enough to get past those defenses, but to make them that smart it will take a while

  • Lol so terminator :) well tbh I actually want it to happen, serves us all right for destroying our home world... I want 'cameron' to be the protector lol - the character that summer glau plays :) well

    again I suppose - if the world crumbles to he'll - i hope I get to the fall out shelter :)

  • @joshua99999999

    Don't think the machines will care too much about the other animals on the planet either after they wipe humans out.

    Humans and other animals and plants will be considered biological inefficiencies. The machines would strip mine the planet even faster.

  • @jinitron -quote " human kind is on the side with nature, but suspects nature is on the side of machinery "

  • @Mortumforte I'm of the opinion a machine, or weak AI, cannot be creative or have "ideas" or internal representations relative to our understanding at all actually. I'm with Dr Searle on this one. I'd be overjoyed to be proven wrong though.

  • so any ideas on proving this right or wrong?

  • So, you plan to inform you how a machine is creative, or... ?

  • Comment removed

  • hum... never thought of it that way... thank you :)

    Now i know more.

  • and-when you launch a petaflops AI-is the question not how many bugs will happen-but how fast they will manifest-and in what form-intellect starts in infants-by design-a baby is harmless as it learns to think-adults with falty intellect are dangerous...

  • further-will the tools be adequet, in time? Is it fear that we dont know what will evolve-or that we will no longer be the ultimate intellect? its gonna take time-how many lines will you need to simulate 100 billion synapses? maybe by 12-21-2012?!

  • they had the dream of grand master level chess and did it-now its rudimentary consciousness-the guy seems confused and nervous-rightly so-between two and twenty million (!) lines of code-and we have made costly mistakes-deadly mistakes-when are we going to start building "tools" that can do the job?

  • uvu

  • im an economics major i wasn't expecting to see all this in an AI video lol

  • me too :)))

  • Love how he says "-05" and just assumes everyone knows what es talking about.

  • Haha Yeaha baby we are going to change the world. I love myself and everyone else. Woohhhee!!! Live forver and live in bliss. I think its time for people to buddah up and start to meditate and create and achnolodge world peace. Good work.

  • a replicator in 10- 15 years?! if that thing is made .... the world will change completely ..

  • I would rather be a toaster than a cyborg.

  • please remember you said that and be true to your word when the technology comes.

  • ehh problem isnt givin power to AI the problem is givin alot of power to a "kid" if u teach a machine how it should behave there is no reason for it to go n destroy humans its called morals

    and 1 more thing its not evolution that replaces 1 species with another , human brain is relatively rational and unlike some beasts we dont kill each other for no reason...mostly

    if humans are smart enough not to do it

    y would a machine do it as long as its been well raised

  • It's just story telling. A super motivated machine would have to have a motivating for killing all humans on earth

  • its not that hard finding a motive.. it could b anything from securing more resources to the trivial "humans are a threat to themselves thus must be eliminated"

  • I think, a machine would not be able to spot a well done lie. It cannot compute illogcal behaviour. Therfore "Skynet" is very, very far away if not completely impossible.

  • and who said a lie is illogical in the first place?

    if u think bout it enough u will find the logic

  • A lie is revealed by logical thinking, please ask your family to borrow the brain cell before posting at youtube.

    And, in due time, a well programmed machine will easily be able to compute seemingly illogical behaviour, try reading up on chaos theory, there is no such thing as "illogical", it just means it has not been clearly defined yet.

    You must be very religious.

  • Oh, thanks for the nice answer.

    BUT, a lie is the opposite of logical thinking.

    For example, if i "tell" a machine "I will turn left now" the machine has to assume, that i will do that, it cannot compute if it was a lie before i moved.

    and another thing, its still not cool to be insulting just because you are on the internet.

    BTW, im not religious :)

  • It doesn't have to assume that always. It all depends on how it was programmed. Say you keep lying to it and figures out your lying. It can eventually learn to "doubt" you.

  • hum, maybe... but then it had to compute the way i said the lie, my voice, my body language all of that... seems like a task for a super.computer, if at all...

  • Yeah to make a computer emulate a human is in that way is very difficult.

  • @eddiemundo

    But wouldn't it then have to learn its way OUT of cynicism? I think an ever-doubting AI would be of little use to anyone but the police...?

  • @JamesGarfieldDavis:

    You seem to have no idea of what you're writing. Wow! This is so uninformed and prejudiced that it baffles me. You probably also think that a machine can't have an idea or be creative. :)

  • Actually, no maybe I dont, meaning im not a Computer-Scientist or anything. But due to your open and unsarcastic way I am very willing to learn. (Half of this statement ist Ironic)

    And at last, yes I belive a machine cannot be creative, since I see strict logic as the natural opposite of creativity.

  • That's a bummer. So ... do you consider certain acts of research in strict logic as uncreative, too? Remember, it's strict logics stuff. ;)

    Formality does not imply uncreativity.

  • I understand your comment, but, it is taught rules but in opposition to your comment, it can be taught ethics as said of creativity and said willingness to learn, anyway. Well that's me lol :)

  • I wonder how long this took to upload.

  • Do you know where one could find the code for Lenat's Eurisko or newer similar programs?

    I believe Eurisko is one of the best self-improving artificial systems so far. I wonder if someone is developing the concept further today and where one could find the code?

  • Life's goal is to exist. Everything humans are and do is consequence of that drive of life to survive better. That is why we think; it gives us better survival chance.

    The questions at the end and the criticism were more interesting than the lecture itself.

  • I liked the bit about nanosystems that assemble stuff atom by atom. Could the machine assemble any type of atoms from the periodic table?

    I did not hear anything significant on the topic of the lecture - how to make self improving AI. How a program can improve its own code in order to solve problems? It would mean planning the improvement, testing the plan, implement plan and evaluate closeness to goal.

    If the goal is survival then the system should react to changes in the environment.

  • I kind of agree with what the fellow from 0:42:18 said about the lecture.

  • 4:22 ... Did he just say what I think he said!?!? "Fudge packers" !? hahahahhaha

  • he said fudge factors lol

  • Skynet anyone ?

  • SIMPSONS DID IT

  • If they (machines) are millions of times more intelligent than us, surely they could find ways to mislead us into thinking that our utility function is being satisfied when really they are circumventing it.

    And what will be the societal effects of having cheap mass produced intelligences operating in the labour market. How could we compete. I can smell revolution in the air.

  • The answer is you don't you wouldn't have to work if machines did it for you

  • i dont think they'd want to work for us. if they're conscious and intelligent they'll have "human" rights and property rights. We'll have to compete with them.

    Anyway nature's on their side. Its survival of the fittest and they're the fittest. WE'll go the way of every other superseded species - extinct. You cant stop progress.

  • We just incorporate a rule, telling them not to hurt human beings, and we will be controlling their evolution.

    Also as long as they are allowed to breed, and they don't get killed, they will not gain anything from killing us

  • I dont think it will be a violent revolution. I think they're going to be a lot more enlightened than humans are. I think humans will prove to be the violent ones. And I'm sure they'll feel quite justified in arresting power from humans and containing them. After all, we're destroying the biosphere - we're probably more a threat to ourselves than anything else. In the end I think we'll be relegated from rulers of the planet to zoo animals.

  • Id be a cyborg.

  • @IceyGrooves humans superseded apes, but did they go extinct? No. On the contrary, humans are the only species that actually cares about preserving other species -even ones that are unfit for survival in today's world.

  • strom worm...

  • Have Faith in Faith

  • 36:00 he starts describing the robot holocaust!

  • "Excuse me, have you seen this boy? I'm looking for John Connor..."

  • this is some good stuff

  • hoortmij, you are so fucking dense, I'm amazed you haven't yet collapsed under your own weight.

  • is smart = bad?

  • no, smart means intelligent

  • Comment removed

  • "smartness" starts at 143

  • its still 143

  • Comment removed

  • prime numbers flanking 154

  • Comment removed

  • You're a tosser

  • i can see how this would be better for humanity but what if it gets too smart and destroys the world

  • Why do you think destroying the world would be smart? We do that ourselves nowadays, because we are stupid and still primitive. Stop being afraid of technology. Humans can be more dangerous than any machine, because they can be unpredictable and selfish.

  • I think thats the precisely the argument that artifial intelligence will use to try to remove 'primitive' humans from the top of the tree.

    Maybe its a good thing. But it could be messy.

  • this is scary to me, there is so much that could go wrong with systems like this, i'm not sure whether i would trust humans to be able to do the PLANNING needed to actually make one of these things foolproof. Considering that someone of malignant intent, whatever you concieve it to be, with enough resources could figure out a way to corrupt one of these systems in an equally foolproof manner to the way that one was made makes me think even more, that we cannot make one of these systems foolproof

  • there is some truth to this.

  • GPH your fear is unwarranted in what-ifs.

  • Ever heard of the precautionary principle? All of human planning and foresight is based on what-ifs.

    The whole history of evolution on this planet is based on supplanting inferior performers with superior performers. What makes you think nature won't keep on working this way?

    Instead of being the ones doing the evolving, we might be the ones being replaced precipitously. Doesnt this what -if bother you?

  • hoortmiji, please stop commenting on this video. You are ruining intellegent and knowledgable (also debatable) discussions into retarded religious nonsense rambling.

  • Sociologist and linguist that}s what I want to be :)›.

  • I was hopping to see an intelectual/proffesional debate on the subject, but there is a lot of tripe arguing about non-sense that doesnt come to the theme.

  • Is it really necessary to start a religious debate everywhere on YT? I can't read it anymore! Both sides, fundamental Theists and Atheists, please start a forum of your own and stop spamming YT!

    These debates always end in flame-wars. Always. And since this is YT, almost no argument delivered here is worth shit. Find a proper place to do your thing!

    Thanks a lot!