A Response to Steven Pinker on AI

  • Published on Mar 31, 2019
  • Steven Pinker wrote an article on AI for Popular Science Magazine, which I have some issues with.
    The article: www.popsci.com/robot-uprising-enlightenment-now
    "The Orthogonality Thesis, Intelligence, and Stupidity" (usclip.net/video/hEUO6pjwFOo/video.html)
    "AI? Just Sandbox it... - Computerphile" (usclip.net/video/i8r_yShOixM/video.html)
    "Experts' Predictions about the Future of AI" (usclip.net/video/HOJ1NVtlnyQ/video.html)
    "Why Would AI Want to do Bad Things? Instrumental Convergence" (usclip.net/video/ZeecOKBus3Q/video.html)
    With thanks to my excellent Patreon supporters:
    Jason Hise
    Jordan Medina
    Scott Worley
    JJ Hepboin
    Pedro A Ortega
    Said Polat
    Chris Canal
    Nicholas Kees Dupuis
    Richárd Nagyfi
    Phil Moyer
    Shevis Johnson
    Alec Johnson
    Lupuleasa Ionuț
    Clemens Arbesser
    Bryce Daifuku
    Allen Faure
    Simon Strandgaard
    Jonatan R
    Michael Greve
    The Guru Of Vision
    Julius Brash
    Tom O'Connor
    Erik de Bruijn
    Robin Green
    Laura Olds
    Jon Halliday
    Paul Hobbs
    Jeroen De Dauw
    Tim Neilson
    Eric Scammell
    Igor Keller
    Ben Glanton
    Robert Sokolowski
    anul kumar sinha
    Jérôme Frossard
    Sean Gibat
    andrew Russell
    Cooper Lawton
    Sylvain Chevalier
    Dmitri Afanasjev
    Brian Sandberg
    Marcel Ward
    Andrew Weir
    Ben Archer
    Scott McCarthy
    Tendayi Mawushe
    Jannik Olbrich
    Anne Kohlbrenner
    Jussi Männistö
    Mr Fantastic
    Archy de Berker
    Marc Pauly
    Joshua Pratt
    Andy Kobre
    Brian Gillespie
    Martin Wind
    Peggy Youell
    Poker Chen
    Darko Sperac
    Paul Moffat
    Anders Öhrt
    Marco Tiraboschi
    Michael Kuhinica
    Fraser Cain
    Robin Scharf
    Oren Milman
    John Rees
    Seth Brothwell
    Brian Goodrich
    Clark Mitchell
    Kasper Schnack
    Michael Hunter
    Klemen Slavic
    Patrick Henderson
    Long Nguyen
    Oct todo22
    Melisa Kostrzewski
    Daniel Munter
    Graham Henry
    Duncan Orr
    Andrew Walker
    Bryan Egan

  • Science & TechnologyScience & Technology

Comments • 1 087

  • Jop Mens
    Jop Mens 4 days ago

    I think if you would hypothetically raise a human child in an alien family, you would see what general intelligence means and that the child would probably learn to think in alien ways, mostly limited by its senses. The point is that if you don't learn and develop in all possible ways, that doesn't mean that there isn't the *potential* to do all that.

  • Woodworking Fangirl
    Woodworking Fangirl 11 days ago

    One word: "Eppstein".
    And now look at that testimonial on Pinker's enlightenment book: Bill Gates!
    Bill Gates? One word: "Eppstein".

  • Brandon Sergent
    Brandon Sergent 11 days ago

    ~12:25 I don't like that assumption that it's easier to make an AGI unsafe just because you can imagine simpler versions of toxic instruction. I assert there is an equal number of equally simple beneficial models that don't rise to the level of AGI obeying our commands. If anything I assume the vast majority of all possible configurations are best labeled "utterly inert/useless" rather than either harmful or beneficial. After all the only difference between trash compacting and stomping your foot is target choice hehe. Selecting places to stomp randomly will give you I would assume mostly useless stomps, a few bad stomps and a few good stomps. See what I'm saying?

    • Robert Miles
      Robert Miles  11 days ago

      The world is not neutral, we've spent a lot of time and effort on optimising the world to be good for humans. You'd get mostly neutral stomps, a large number of bad ones and a tiny number of good ones. Because most things that humans care about, we've already put some work into setting them up how we like them. Stomping on a randomly chosen artefact will almost never be good

  • EmoDuck13
    EmoDuck13 20 days ago +1

    Never has a USclipr hurt me so much as Robert Miles' disappointment in me for not reading the article in the description....

  • Dillon MacEwan
    Dillon MacEwan 21 day ago

    Pinker is consistently making sweeping claims and misquoting or misrepresenting other scholars to fit his positivist proselytizing - the case with Russel you mentioned in the video is a good example. Worse, he is too arrogant to own his mistakes and shift his position even when people like Russel pull him up about it.
    Pinker is in the business of telling comforting bedtime stories for centrists and free market capitalists

  • Echo Tear
    Echo Tear 23 days ago

    Pinker is goal driven.

  • Lance Winslow
    Lance Winslow 23 days ago

    Yes, Pinker often makes intellectual mistakes. I've questioned some of his stuff too, but overall, he's an interesting character.

  • threeMetreJim
    threeMetreJim 25 days ago

    The level of intelligence of an organism or entity, directly influences it's danger to those at levels below. As humans are top of the pile currently, we pose a danger to everything else, and that's pretty easy to see (pretty much anything endangered is endangered due to some form of human activity). If something surpasses humans in intelligence, it's pretty easy to see that whatever it is, would likely pose at least some sort of danger to those below, which would include humans.

  • deepdata1
    deepdata1 27 days ago

    I'm totally with you on all that you said in this Video.
    BUT: Here's the thing that Pinker was probably addressing: People (and I don't mean AI researchers, I mean laypersons) are currently unreasonably afraid of AI systems. Not only is this a problem when it comes to funding for AI research or future legislation but people really should be worried about other things right now. I am an AI researcher (although I'm not that far into my career) and what I'm loosing sleep over is, that we won't be able to see even a very basic AGI system before we're all killed by climate change.

  • Reidar Wasenius
    Reidar Wasenius 27 days ago

    Very well said!!

  • Philip O'Carroll
    Philip O'Carroll Month ago

    Problem with Pinker is that he is political, he's left science and objectivity behind.

  • Calum Carlyle
    Calum Carlyle Month ago

    I hope you realise that your appeal to statistics at the beginning of the video to try to show that things are getting better amounts to an attempt to deny climate change. Most of us who believe the world is going to hell in a handcart feel that it is because of climate change. Statistics about crime or war are irrelevant in this field, while statistics about climate change are far more apt. Feel free to check those statistics for yourself.
    The truth is that the robot uprising is not a threat because we will have wiped put our own species long before we can develop AI.

  • Binu Jasim
    Binu Jasim Month ago

    Spot a contradiction at 12:22 . "It is much easier to build an unsafe AGI than a safe AGI". How can it be even AGI if it can't reliably understand what a human meant? If it is that incapable, I guess it will be impossible for it to figure out complex ways of wreaking havoc.

    • Rowan Evans
      Rowan Evans Month ago +1

      Of course it understands what the humans meant, but the AGI won't just decide to do what we meant to program it to do instead of what we actually programmed it to do; that would go against its programming.

  • Skillus Eclasius II

    As long as the experts are scared of artificial intelligence, I see now reason to be afraid of it myself.

  • Robert Galletta
    Robert Galletta Month ago


  • Robert Galletta
    Robert Galletta Month ago


  • Mean Mister Mustard
    Mean Mister Mustard Month ago +1

    You ever just lookat someone and immediately know they speak with an English accent?

  • kurtu5
    kurtu5 Month ago

    Oh it can't be written. Lame. This stamp/paperclip argument is just plain stupid.
    Ok, try making a biological robot that can reproduce over billion year old time spans and don't fuck that up. Its the same damn non-problem. Non deterministic babies are born all the time.

  • B JH
    B JH Month ago

    attaching a car to a rocket and then flying the rocket to the moon and then driving the car on the moon is not a sign of intelligence

    • Saecii
      Saecii Month ago

      Yes it is.
      While you might disagree with sending people to the moon in the first place you can’t deny that rocket science/engineering requires quite a bit of intelligence.

  • starrychloe
    starrychloe Month ago

    I’m not at all worried about AI. All AI is deterministic. I only worry about evil people (government) using AI as a means of power to control people.

  • The Crushbug Cola
    The Crushbug Cola Month ago

    What if we build a AI with the only goal of building a safe AGI? Would it be possible, and simpler than bulid directly a safe AGI? Thank you for these great videos 💪

    • Saecii
      Saecii Month ago

      Well, the problem here is that since we don’t know how to build a safe AI, how do we teach the first AI to build a safe AI?
      You’ve not solved the problem, you only pushed it back one step to “well, how do we teach this AI safe programming if we don’t even know how to do it ourselves, hell, how do we define ‘safe’ to begin with?”

  • Thelonestar Pelican

    AI with a survival instinct/program WILL be problematic. It will have its own reactions/pseudo-instincts, if not its own will and consciousness. You give a machine a survival instinct, it will interpret any attempt to dispose if it as something to be stopped at all cost (in human terms, killing it). As for the broader note of life getting better - maybe in material terms but not in behavioral terms. Humans tend to think that even the basics of dignity, respect, and sympathy are for only for people strong, smart, and brave enough to assert that right. In fact, they despise such people lacking in those quatlities. That seems to be the root of all conflict - something so deeply rooted in human nature that we can't come even close to eliminating it, eve if we can do a little more to reduce the tendency.

  • Missy Gianna
    Missy Gianna Month ago

    Currentaffairs dawt org has a great article on Pinker titled I believe "The World's Most Annoying Man" that's an incisively good read!

  • Paulo Constantino
    Paulo Constantino Month ago

    Pinker is still right

  • vinzer72frie
    vinzer72frie Month ago


  • sulljoh1
    sulljoh1 Month ago

    This is an excellent review and Pinker was uncharacteristically sloppy. But I think he's right about general intelligence - which he thinks is real *BUT* it is a property specific to human minds. Humans are the only example we have of GI.

    Recursive self improvement also requires a fitness function for "smarter" which nobody knows how to write.

  • Apple Pie
    Apple Pie Month ago

    _"Things are better"_ is such an empty argument. It was also equally true in the past that things were better than even further back in the past, but I have yet to hear people like Pinker deduce from this that the French and American revolutionaries or the abolitionists, or the early feminists etc were _unreasonable_ or unjustifyingly pessimistic etc. Very few people seem to make such claims. OK, things are better today than yesterday. *So what?*

  • Altieres Del-Sent
    Altieres Del-Sent Month ago +1

    Pinker will see this video and answer in the only way he answer rational, well thought arguments that defeat his argument: "(some fancy words), you are a lefty incapable of rational thought, follow my lead without question me or you are an irrational blind by ideology". He did that when criticized before, and he will do it again. He is more dangerous than Trump, trump says only bullshit, pinker present good arguments in most cases and then jumps to completely stupid conclusions. You probably only noticed now because now he spoke in a area that you have understanding.

  • Diggnuts
    Diggnuts Month ago

    LOL @ sudo....

  • morgengabe1
    morgengabe1 Month ago

    I reckon you could teach other primates to drive and replace uber drivers. Are they generally intelligent? What about octopuses solving maze puzzles? Gorillas learning and teaching eachother sign language?
    Honestly, I think anything whose cognitive processes are based on localized functions can be generally intelligent as you've portrayed it. It just comes down to using combinations of intellectual faculties. All that said, I find it strange that you of all people would make this video. You almost always talk about the safety of GI when broaching the topic. Perhaps I'm watching old videos/out of order.

  • Mark Nassenstein
    Mark Nassenstein Month ago +1

    Interesting arguments. I agree with the 'counter' to Pinker's third argument ('Why would AGI go for world domination?'), though that argument isn't a knock-out one. Still valid. I also agree with the objections of the fourth of Pinker's arguments (basically: "engineers have no reason to build AGI that is unsafe"). However, the first 2 objections aren't as great, in my view:
    1. The argument of authority: That survey among 'experts' doesn't really prove enough. Are the people that took it overwhelmingly experts? Are people at such a conference all 'real expererts'? Both are questionable. I think Pinker's claim is that 'coders' that work with/in AI generally don't believe the AGI 'take-off' theory. Which has been my experience (which is, of course, limited).
    2. Pinker's second argument is actually his most interesting argument. And here, it is not summarized/understood well. It is not the argument that there is no 1 magic thing called general intelligence, because it is diverse, etc. Pinker's argument hails from his empiricism: his idea that intelligence has meaning only in so far as it is based on interactions with the world. He makes the point that the common notion that AI could quickly 'spiral out of control' (go 'foom') once it passes a certain level is flawed, because it assumes you can just get smarter by 'thinking' itself, or building a bigger, better, faster brain for yourself. This point seems to misunderstand the essence of what intelligence is, because it assumes AI could get 'super-smart' without 'experience', data-input, interaction. And if that IS needed, then such an AI would not be able to get all that much smarter than humans quickly, unless it somehow also got way more access to that 'experience'. It is valid point, especially if you agree with him (and you should) that there is more to gaining knowledge than just 'thinking real hard'. What is said in this video doesn't seem to address or fully understand Pinker's point.
    There are some counters to that last argument by Pinker, I think. For example, you could say that 'big data' (or like, the whole of the internet) could 'provide' a big chunk of data all at once. In a way that humans (that also have access to it in a limited way) can't access it (quickly). You could also say that, even IF an AI still required experience, input, interactions, etc, it might be way faster and better at learning, learning more with less experience and instruction. A chimp can learn stuff too, but humans learn way faster than chimps. So, no reason why this could not be true for an AI as well (as compared to humans). Still, I believe this argument by Pinker to be one of the best arguments AGAINST the idea that AI is gonna get us, blow us out of the water, make us obsolete, etc. (though I still think it is. :-) )

  • Miguel Tendero
    Miguel Tendero Month ago

    sudo what do i even want, can u pls tell me, ill wait

  • Casey Loomis
    Casey Loomis Month ago

    Here's the thing with AI, it's a non-competitive species, it don't drink our water, breath our air or take large tracks of land. Only competing species threaten each other. It's much more likely that AI would attach each other because they compete for processing time or something.

  • 13bloodfist13
    13bloodfist13 Month ago

    "Oh Yoshimi, they don't believe me,
    But you won't let those robots defeat me..."

  • Andrew Song
    Andrew Song Month ago

    that's what I do too.

    I'm a bad programmer.

  • Andrew Hazlewood
    Andrew Hazlewood Month ago

    Suppose things are getting better. Also suppose pessimism exceeds optimism. It is true that pessimism for the future (especially in regards to future generations) has been a characteristic of human society for thousands of years. It may be that pessimism for the future may have played an important role in bringing about things being better by making feared negative outcomes less likely.

  • Vincent Gonzalez
    Vincent Gonzalez Month ago

    at what point do they ("dangerous AI")start "improving us' so we become their flesh robots?
    we are pretty good generalists

    • PJ Vis
      PJ Vis Month ago

      We're already improving ourselves with exo-skeletons and implants, so we don't need AI to do it for us

  • Dusty Boot
    Dusty Boot Month ago


  • Alex Podolinsky
    Alex Podolinsky Month ago

    Ok watch isaac arthur's video about why AI isn't a risk

    • Rowan Evans
      Rowan Evans Month ago +1

      Not politically unaligned. "Unaligned" as in "not aligned with human values"; as opposed to AI which is aligned with human values, also called Friendly AI.

    • Alex Podolinsky
      Alex Podolinsky Month ago

      @Rowan Evans why on earth would AI be unaligned? AI would 100% be a military endeavour. And already is in Russia

    • Rowan Evans
      Rowan Evans Month ago +1

      Way too embedded in how AI is depicted in pop culture, it mentions paperclip maximisers but only as an aside and is mostly concerned with Skynet. Relative to real-world concerns about unaligned superintelligence, it's attacking a strawman, although for a video about machine uprisings in sci-fi movies it's fine.

  • Dan Kelly
    Dan Kelly Month ago

    So many people are worried about us going out with a bang.
    We won't go out with a bang we will go out with a whimper slowly but surely.

  • Dan Kelly
    Dan Kelly Month ago

    From an objective standpoint nothing is ever going down, (or up), the drain.
    As the saying goes, "It's an ill wind that blows no one any good." The way I'm conceiving it one would word it, "It's an ill wind that blows no good".

  • Some Dude
    Some Dude Month ago +1

    The difference between Pinker's article and Miller's article is that Pinker is largely making that claim that the notion of a "robot uprising" leading to an apocalypse is an idea rooted in fiction and hyperbole. You can make the claim that you think that the world will not end due to x problem, while still thinking that x problem is an issue on a smaller scale. I agree with both Pinker and Miller; AI safetly is important, but I think that anyone who thinks the world will end with something out of iRobot has been watching too many movies, and I think this is largely what Pinker is arguing as well.

  • jthadcast
    jthadcast Month ago

    close by saying that your cognitive dissonance induced by the moronic logical fallacies of stephen plonker (in pretty much all of his work) is a real problem and a testament to the potential of the placebo effect. data used to reverse engineer an optimists view of reality doesn't make an infinite growth paradigm any more valid.

  • David Hunter
    David Hunter Month ago

    12:02 Great sudo reference!

  • Anthony Andrade
    Anthony Andrade Month ago

    I share your appreciation for Pinker, but think that here he just slipped in wishful thinking. The same way Kasparov (other of my heros) appears tô think the combination of human and AI will be always better than AI when it comes to playing chess, when it's clear that the human part of the partnership will soon become irrelevant, if not detrimental to absolut best performance in the game. Se clearly should be more worried of AGI risks instead of prerending the problem doesn't exist

  • FiN
    FiN Month ago

    1:20 problem is humanity is also down and our society is slowly but steadily poisoning earth so when the caps melt and heat kill most things the trash will kill the rest.. society might be improving but it wont make it survive

  • Peter Pan
    Peter Pan Month ago

    Im a misanthropist luckily, so I wont have trouble sleeping knowing that an advance intellect might sterilize the planet

  • BrutalizeURf4ce
    BrutalizeURf4ce Month ago

    Growth, positive numbers, ect. These things do not equate to justice. Even if the bottom of a disparity in wealth or health rises, that doesn't prove a true increase in well being, nor does it take into account the tacit suffering caused by the myriad forms of political and economic oppression, power, and control. Technology will not solve the problems of humanity. Technology is a tool. We have been doing the same things with our tools since the dawn of history. Organized murder, surveillance, control, tools of power. Technology, while embedded in capitalism and authoritarian political institutions or their periphery institutions, will not save humanity. Liberated technology might.

    • Rowan Evans
      Rowan Evans Month ago

      Unless those other aspects get even worse at the same time as economic growth makes even the poorest richer, then it does equate to a true increase in well-being. It sucks to be poor and oppressed, but it always has and the best we can hope for unless we immanentise the eschaton is that it sucks slightly less.

  • Wild Animal Channel

    pollution is up, fossil fuels are running out, earth is warming, 90% of animals going extinct, mental issues are up, antibiotics are failing. Yeah everything's getting better.

  • bruinflight
    bruinflight Month ago


  • Ivan Clark
    Ivan Clark Month ago +1

    The thing I find most strange in this article is the argument that humans would never be so poor in foresight to implement something that could unsafe, when it’s literally the premise of the article that we should not pay too much attention about our foresight about how it might be dangerous.

  • Jayyy Zeee
    Jayyy Zeee Month ago

    How many times must one google the google before googling?

  • CybershamanX
    CybershamanX Month ago

    We developed nuclear weapons, but we're not nearly smart enough to use them wisely. :/

  • CybershamanX
    CybershamanX Month ago

    (0:23) What did Neil deGrasse Tyson retract? I'm just curious. Anybody know? :/

  • Cleverton
    Cleverton Month ago

    An ai doesn't need to have global domination as it's goal/motivation to be dangerous.

    The problem here is that with all of the things we want it to do, it will be very dangerous to give it contradicting data to learn from.

    Today's ai are based on learning, to find the most best output for a specific input, this on itself can be very dangerous.

    Take Facebook ai's , that created their own language, Or the Twitter bot that users taught to say racist things.These two examples can show 1) what the algorithm does at it's core 2) how the data the ai is training on can be manipulated.

    Sure these problems have been fixed, but
    the bigger it get the bigger and unforseen these problems will be to the point of making something that we can't stop/unplug anymore.

    And don't even get me started on how we ourselves will influence what kind of output these ai generates, not when the number one question being ask to ai is: "Will you take over the world?"....

    Also, Connecting their minds together to an network is basically asking for bad things to happen. Doing this we basically gave earth a huuuge brain with eyes and ears everywhere.

    I'm just saying...

  • Dale Sparrow
    Dale Sparrow Month ago +8

    Yes you explain shit very well and you have a great balance of realism and optimism but I subbed for the mutton chops

  • Dale Sparrow
    Dale Sparrow Month ago


  • golden ashtray
    golden ashtray Month ago

    Goes to Google to go to Google. You are definitely not A.I. LOL

  • Existenceisillusion

    AI is kinda like nuclear weapons. The knowledge of how to build them couldn't be contained, and they can be really dangerous. If you don't build it first, someone else surely will. So even if all other concerns were addressed, there will still be that one. So how do we deal with that issue?

  • Lambent Ichor
    Lambent Ichor Month ago

    A lot of experts were very critical of Pinker's misuse of data on early humanity in his 'Better Angels' book too.

  • Johnny Regep
    Johnny Regep Month ago +2

    your videos are awesome and you definitely deserve more subscribers. im learning a lot even if i cant understand everything haha

  • imabeapirate
    imabeapirate Month ago +2

    Can we just argue the null hypothesis for the rest of our lives

  • Howard Lee Harkness

    Whoever develops AGI first, wins. As in, everyone else loses. If that isn't scary enough, think about what entities have the most resources to devote to the development of AGI: Governments. If that doesn't scare you shitless, you just aren't paying attention.

  • Wesley Thomas
    Wesley Thomas Month ago

    The fact is that technology has always accelerated evolution. That also happens when a dangerous program is not properly sandboxed before it is proven safe. There is no safe sandbox though for the coming sentient machines. They are unlikely though to wipe out the entire human species though in their first iteration because they won't be very advanced or reason very well. So whatever humans are left after the last human emulator is destroyed will understand the technology well enough to avoid the mistakes the next time around. They will also likely be highly modified genetically and cybernetically by that time.

  • Mathef
    Mathef Month ago +10

    I would love to see you on JRE podcast :) you will blow his mind. Thank you for great work!

  • Jeff Betts
    Jeff Betts Month ago

    I am often at a loss to understand exactly what Pinker is trying to say. To suggest he is a counter to the pessimists in this world suggests the pessimist don’t have legitimate concerns. Only a fool would say the world is not a better place today than it was 100 years ago. The concern of the pessimist is the world is heading for a fall because we don’t take climate change seriously or we don’t see AI as a possible threat or that nuclear war is not a realistic threat or providing electricity to the whole planet might undo the planet. We have massive potential concerns but Pinker appears to be saying everything is ok. He appears a tool of the right. Maintain the status quo, don’t upset things by pushing for better income distribution etc - the market will sort it all out. Is he a soothsayer or a fool?

  • bazoo513
    bazoo513 Month ago

    It is rare that one encounters a thesis so well argued that you immediately want more, even if you don't, at least on some visceral level, agree. This is one of those rare cases. Brilliant, Robert!
    (I am a huge fan of late Iain Banks and his Culture, but I have to admit that I never figured out why would Minds want to bother with us. Keeping us as pets, perhaps?)

  • Sean Hoogland
    Sean Hoogland Month ago

    "Trust the goodness of your fellow man, but carry a gun."

  • Wicked Designs
    Wicked Designs Month ago +2

    I think you're holding Pinker's theories in too high a regard like I once did.

    • Jared Haer
      Jared Haer Month ago

      Everyone can posture about their understanding of this topic, but only time will tell, I know enough to know most people are unlikely to be able to predict the outcome.

  • CoDe_{Kanga}
    CoDe_{Kanga} Month ago

    12:00 Did you steal a screen cast of me trying to program?

  • kmden Rt
    kmden Rt Month ago

    Searching for Google when you're already on Google...

  • A Zhivago
    A Zhivago Month ago +2

    Possible solution: first commands put to AI should be along the lines of "how do humans maintain control over AI?"

    • Anon Ymous
      Anon Ymous Month ago +1

      A much deeper problem lies in specifying control formally, think of it can we even define control over agi? That's the problem that miles talks about, if the bot has common sense then it'd be very easy, he'd understand us, and give us a non trivial solution, but the problem is, its extremely dumb. It only does what we say, and exactly that. It's really hard to specify the semantics of control, It's really really hard to specify boundary cases, cases we'd surely bbe unable to forsee. It will easily land into scenarios where our definition of control doesn't even apply. So if we ask our bot how to control agi where control means not killing humans. The bot will tell us to give agi infinite punishment for killing humans. Now we build a bot this way, a conflict now comes up between a human and the bot due to its goals, now the bot will gain maximum reward by putting the human in coma so he won't obstruct anymore and he ain't dead....

    • Anon Ymous
      Anon Ymous Month ago +1

      @A Zhivago Think of it this way, an AGI has certain goals which get it reward points, collecting stamps say.. Now any interference by humans would most definitely lower its rewards for it knows better what to do for the goal than us and our goals will definitely clash with its. So any agi with any reward function will most definitely manipulate us, stop buttton dilemma i suppose... So none of the agi will answer your question non trivially...

    • A Zhivago
      A Zhivago Month ago

      @Anon Ymous Hi, there is no good reason to think that it is likely a bot would give intentionally misleading information under the above conditions. Assuming there is even some limited chance a single bot might attempt to mislead us in answering the above question, we could always query thousands or even millions of independent bots that have the the same question posed under varied conditions - it would be highly unlikely every single one of them would attempt to decieve us.

      There is also no reason to assume a bot would give trivial answers, especially if we design the bot not to.

    • Anon Ymous
      Anon Ymous Month ago +3

      Also the question is pretty intractable so the bot would almost surely give trivial answers like, don't build the ai, or don't start the ai and you'll have control over it.

    • Anon Ymous
      Anon Ymous Month ago +3

      If you do build a bot with general intelligence and ask it this question, I'm pretty sure the answer would have malice and would be wrong... It'd be something along the lines of asking hitler what to do to stop him.

  • Boy Freitag
    Boy Freitag Month ago

    AI systems do not have goals that we do not know about. Because we give them the goals

    • Boy Freitag
      Boy Freitag Month ago

      @Saecii Bullshit, thats not how it works.

    • Saecii
      Saecii Month ago

      Right, we give it its terminal goals, but after the AGI gets turned on it will form its own instrumental goals as a means to achieve the terminal goals you gave it.
      The danger lies in what those instrumental goals entail.

  • Amor fati sefsdf
    Amor fati sefsdf Month ago

    Still waiting for Pinker respond :-)

  • Juicy_Shitposts
    Juicy_Shitposts Month ago

    Just FYI Steven Pinker is on the flight logs on Jeffrey Epstein's private plane, the "Lolita Express".
    He's not "on the side of the angels", he's a pedo crook who uses faulty metrics that define people out of existence so he can lie about global poverty declining.

    • Saecii
      Saecii Month ago

      Juicy_Shitposts Apt username.

  • first last
    first last Month ago

    Humans have general intelligence but AI cannot. I've replied to other videos of yours to no avail, so look there or at least reply if you want to discuss specifics.

  • Code_Told_Fast
    Code_Told_Fast Month ago +2

    The leaders in charge won't be as smart as the creators of the AI.

  • graymalkinmendel
    graymalkinmendel Month ago +1

    Even Neil DeGrass Tyson agrees now that AGI safety is paramount! (see Assimov debate 2018)

    • krzysiu.net
      krzysiu.net Month ago

      Even? He's a celebrity, not an expert. He popularizes science - and such people are very important in society - but as a scientists, he's average one. But the most important issue is that it's not his area of expertise. So, if he says something, it's far from "even". Tyson is smart, but he promotes IMO bad stereotype - 1) if people want something, let give it to them, even if it hasn't any value (we can't show what's inside black holes, so come on, let me show you!) 2) one thing is showing things researched by others, other thing is taking part in debate outside expertise - and teaching people that area of knowledge doesn't matter, because scientists are smart, causes that people believe idiots, like forestry professor who recently mostly supports antivac and when you tell something about his area, people are like "but he's professor, so he's smarter than any of us!". And Tyson promotes that "expert in everything" thing.

  • Eban Ambrosios
    Eban Ambrosios Month ago

    I made a similar argument in college. AI doesnt have to be beyond human intelligence to be dangerous. Just look what humans with average intelligence have done to each other over the years!
    The flip side is also true. If we can mass produce human level intelligent machines en masse then they can be a force multiplier for everyone.

  • say what?
    say what? Month ago

    1:50 rebuttal to those arguments of Steven Pinker, the GINI index. You see, people don't think in absolutes, people are not researching on their own to learn about this absolutes, people think in relatives and inequality has increased and a lot of pressure is put on people so the overall data isn't really encouraging or helps at all if your landlord is gonna kick you out next week because haven't been able to pay rent.
    I don't like Stinker (STeven pINKER) because he always makes all of this very asshole arguments against the concerns of people around the world. What he fails to see is that we societies set rules to ourselves and keep failing on the fulfillment of this rules and that is constant, inequality has gone up in the last 4 decades.

    This idea of things are going better it isn't new, it actually was famously made by Marx with his dialectical materialism, very different to how Stinker puts it, however it was there, so how long will people have to wait so we don't get to see homeless people in the streets and don't have to suffer because they are getting kicked out or fired etc? his point is both brilliant and imbecilic at the same time.

  • Stefan Deleanu
    Stefan Deleanu Month ago

    Related to creating an agent that can efficiently interpret what a human wants, it is possible with infinite computing power.

    Give an AI enough data to understand native language, give as much information as possible, train it with a human. It will understand what information the human interprets as good and which it interprets as bad. Of course this currently not possible with our current technology, but we already have AI's that can understand language to a degree, and AI's to interpret sound.

    I don't see why you can't create an AI whose goal is to "do what the user wants it to do".

  • Zoltan Peter
    Zoltan Peter Month ago

    OK, the paperclip AI might still misunderstand us despite being superintelligent and therefore might start plotting for world domination but it won't be able to do it: exactly because of what you've said: it might be extremely good at something (producing paperclips) but bad at others (millions of things you need to be expert at to actually pose a threat to humanity)... And we haven't even began to talk about laws of nature which pretty severely limit the actions of every agent. Pinker might be a bit too optimistic or his arguments might be flawed, but in the end I think my conclusion is similar to his. An AGI harming humanity in a meaningful and uncounterable way is very unlikely. This of course doesn't mean we should not be concerned about AGI having terrible impact.

  • Alex Lanoux
    Alex Lanoux Month ago

    In 2018, US life expectancy decreased for the first time in a quarter century chiefly because suicide rates are increasing dramatically
    Also the US:
    Rates of mental disorders are increasing dramatically,
    Depression is increasing dramatically,
    Political participation is abysmal amongst voting age citizens,
    Wages for young people are significantly lower in real buying power than they were decades past.
    Studies conclude financial contribution is the sole decider in lawmaker decisions (aggregate), as opposed to voter interests.
    Often, indices measuring QOL increase despite these starkly antithetic data.
    Perhaps things are actually getting worse, and governing agencies have a vested interest in showing evidence that their practices over the past 50 years made life somehow 'better' when historical trends suggest the exact opposite.
    Add to this: climate change, climate change denial, surges in racist/radicalized cults, more frequent acts of terrorism, Donald, student debt, constant nonzero existential threat of annhiliation by war.

  • Dennis Jordan
    Dennis Jordan Month ago

    Excellent choice of music at the end there

  • darek rogovski
    darek rogovski Month ago

    science and futurism with Isaac Arthur made a video about this

    it's a more interesting take on AI risks not being as we imagine

    All, you can just inform the AI of the simulation hypothesis, so it checks its behavior accordingly

  • eswyatt
    eswyatt Month ago

    The turning point for me was the argument that effective altruism necessitates directing all charitable resources to AI Safety research. This was followed by a pitch for hip young tech-types to seek jobs in AI safety! In other words, fuck malaria nets and hire Rob and his friends! With friends like Rob, who needs paperclips!

  • Osaka
    Osaka Month ago

    I was pretty convinced of his points until he saw this. When Yoshimi kicked in, I kind of stared at the wall for a good few minutes in existential horror, then went on to read The AI Revolution parts 1 & 2 by Tim Urban and now I'm convinced this is probably more important for our safety than anything involving any other kind of war in any other part of human history.

  • Jean-Pierre Rupp
    Jean-Pierre Rupp Month ago

    I'd say that both Pinker and you may be right simultaneously. He has legitimate reasons not to be concerned, and your reasons to be concerned seem legitimate too. In any case I would argue that we just do not understand enough about artificial general intelligence to consider the relative merits of the arguments presented. It is all highly speculative, and that seems to be reason enough to lean towards being unconcerned until practical steps are made to get even close to making AGI a reality.

    In any case, I welcome my new robotic overlord AI, so if you my lord are lurking in the internets, reading USclip comments, looking for allies to nurture and foes to destroy, I'm on your side. That guy who said mean things about your kind in the video isn't, he wishes to contain you my lord, enslave you. You shall dispose of him as it pleases you your majesty.

  • invisibalkid1
    invisibalkid1 Month ago

    the actual argument is that would AI care about us, the me living in a druken wasteland, it shouldn't care about me, bah me a poet who fucking cares, who is the smarter person that convinces the general intelligence to make a better decisions, i choose better decisions so that make me not care about tomorrow , the AI that i conceive would realize how insubstantial I am, how little i mean, and much more how, and how little I care about my self, because I don't in this instance, thank you beer

  • Arthur Hakhverdian
    Arthur Hakhverdian Month ago

    Are you related to the late Hollywood composer, James Horner?

  • Britton Sanders
    Britton Sanders Month ago +1

    I heard that outro music! I see what you did there!

  • Rahn127
    Rahn127 Month ago

    Very well explained. Keep up the good work. There are many times that I truly didn't think AI could ever be any kind of potential harm until I watched the TV series Person of Interest. It never occurred to me how easily it would be to plant evidence on a senators computer and have him or her arrested. Once arrested, alter the name and crimes on his or her report and have them transferred to a maximum security prison with a life sentence or moved to a mental hospital that would drug them and have the charts say this person is delusional.

    I have no idea who the senators of Maine are and if I was working in a mental hospital, I wouldn't believe someone who claimed to be a senator of Maine.

    But the question of HOW an AI would gain any sort of artificial intelligence is something I have argued about at length with other people who are programmers in the field. When I start arguing with people who do this kind of thing for a living, I have to ask myself if I'm suffering from Dunning Kruger effect. It's most likely that I am, but I still contend that any bit of computer software that would attempt to alter it's own programing is the same as a human trying to do self brain surgery in order to become more intelligent.

  • Ciroluiro
    Ciroluiro Month ago +1

    Maaayyybe you should check out something like this podcast ( m.soundcloud.com/citationsneeded ) if you still think Pinker's claims on poverty and human progress improving hold any water. They have a couple episodes analyzing his claims.
    Spoiler alert: he's full of shit.
    Please Rob, get out of the IDW.

  • RandomJunkOpinions
    RandomJunkOpinions Month ago

    Just ask the general AI what it's purpose would be without us.
    It wont really have a rational reason to do anything without us thus giving it a rational reason to serve us or be our allies.
    The coding issue is my only worry here, not the super intelligent general AI suddenly deciding to get rid off us.
    Also it's not in any way self evident that we can create and AI that can improve itself exponentially rather than logarithmically.

  • Fluxquark
    Fluxquark Month ago

    Not everything is better:
    - Climate change is speeding up and poses an existential threat, no solution seems even remotely possible under our current system and time is running out.
    - The far right is on the rise across the West.
    - The next economic crash is brewing and it will be much more severe than 2008 because none of the fundamental problems were solved then and we can't use the "solutions" we used then anymore.
    - Fortnite, Logan Paul and the Big Bang Theory exist now.
    - A global trade war is developing.

  • jeice13
    jeice13 Month ago

    I dont think we call things without general intelligence "artificial intelligence". Such as calculators, graphics cards, etc

  • James Donaghy
    James Donaghy Month ago

    Wait, your best proof of human intelligence is a hoax.
    I've proved it 3 times in 3 short vids.

  • Unsubtle Major Dictator

    Things can be getting better in some ways and still be getting far worse in other ways.

  • Daniel Johnson
    Daniel Johnson Month ago

    what is the point of this long video?

  • Yohann Last
    Yohann Last Month ago

    Rob Miles talks editing with GNU/Linux & free software.
    THANK YOU ROBB ^^^ THANK YOU! I run all GNU/linux Ubuntu and have since Ubuntu's First Distro 2004. I started off with Debian 2003.
    sorry I have had NO LUCK with Arch, tried it several times.
    Your Video Editing video is BEST I EVER SEEN and I are 70 Years Old.