A Response to Steven Pinker on AI 

12:02 I laughed way too much when you actually tried the command again with 'sudo' haha. Well played.

3:47 "I'm not trying to construct a strawman here but I may by accident". I bet this sentence is just a modified version of a warning sign posted outside your lab: "I'm not trying to construct a malicious superintelligent AI but I may by accident".

Dat "the google" search tho.

It’s been a long time since I’ve seen, heard or read such a well thought-through argument. Also, the fact that you waited for Pinker to correct himself, compromizing outreach for human decency speaks for itself. I’ve always loved what you do on this channel - now I also love how you do it!

People can be smart in one way and not so smart in other ways.Pinker himself is an example of this, good at psychology bad at A.I. safety.

"I hate this damn machine I wish that they would sell it. It won't do I want, only what I tell it" - The Programmer's Lament (I couldn't find the origin)

1:22 “Disease is down” Well, with some bumps along the way.

Human are EXTREMELY BAD at creating safe things, EVERY safety feature we develop is AFTER it failed at least once already. whatever you would say was safely designed...it's as a reaction to an original failure. Luckily,while many of those fault were terminal to the designer/researcher, none were terminal for humanity...

I didn't read the article but I did hear him interviewed a couple of times, and despite generally being a big Pinker fan, I was slightly reminded of the aged senators talking to Zuckerberg about the internet. I'm not in the AI field but I've done enough programming to know that things always go wrong and the bigger and more opaque the programming, the harder it is to test.

When Steven Pinker talks about a subject that he didn't do his PhD in, and then I talk to someone who is an expert in that area, they have always pointed out very serious flaws in what Pinker said and how it makes them angry.

For the "how can humans be so ingenious yet so dumb" argument, I give you basketball: invented by a single brilliant person, James Naismith. He originally used real peach baskets as well, baskets, so the fact that every time someone scored, they had to climb up a ladder and fish out the ball is probably fair enough. But even after switching to the more familiar rim-and-net design, it still took YEARS for someone to cut a hole in the net so the ball can fall through.

It's funny that Pinker would bring up the Luddites as an example of unreasonable technophobia, because those guys were _absolutely right_ to think that automation would be terrible for them. They were skilled craftsmen, respected people with a high standard of living, and the automation of the textile industry ended up destroying their livelihood and reducing them to pauperism, just as they had predicted. It made the industrialists rich at their, and their communities', expense. In fact early industrialization in England was basically the historical nadir of human health and well-being -- generations of people lived and died (mostly quite young) in grinding poverty, with sickness and malnutrition at levels you'd previously only see in famine years. The greatest difference in height between lower-class and elite individuals ever recorded in history was in Victorian England, because the new urban poor were chronically ill and malnourished and grew up stunted. They had to keep lowering the minimum height for recruits to get enough soldiers. Medieval peasants were far better off than Victorian factory workers. Pinker does that a lot, telling people who are _right now_ being fucked over in entirely avoidable ways that, well, _on average,_ the world is getting more prosperous or whatever. Or more accurately, he tells his affluent Western Liberal audience this as a moral palliative. No need for drastic measures, no need to re-examine the system, it's all going to be fine if you just let the technocrats do their jobs. Sure we could fix world hunger right now, but that would _upset things._ Isn't it much nicer to just wait and watch people senselessly die at a slightly lower rate year by year? Pinker is often framed as promoting "optimism", but I think "complacency" is more accurate.

On the "everything is better than it's ever been, it's irrationnal to be so pessimistic about the future" idea: Although all the indicators of wealth, well-being, security and develepoment are showing that things have been going better and better for the last 200 years or so, we shouldn't hastily extrapolate this trend to the future. Our analysis of the state of the world should not stop there; i think we must then ask the question: what are the conditions that made all this progress/development possible, and can we expect to find these conditions again in the future? To answer quickly: the main cause behind our level of developement is cheap and easy access to natural resources and most importantly fossil fuels (or rather cheap and easy access to energy, allowed by fossil fuels and not much else really) . I don't have time to develop here, but these are very valid reasons to be pessimistic about the future (and I'm not even talking about ecology). I know this is not the subject of the video, but it seems important enough to be discussed.

Alright, what do I have to do to get access to the sarcastic version of this video?

This was so well argued. I particularly liked the point that humans can make something sophisticated without making it safe. This is a subtle but important point about complexity that politicians love to ignore for more votes. Subscribed.

Loved the acoustic "Yoshimi Battles The Pink Robots" at the end!

While I appreciate the video and your concerns that a more expedient response to this article would (or could) have had detrimental qualities, I have to say I hope in the future you will not delay quite so long. Public opinion matters, and one might well argue that as someone with genuine credentials you have an obligation to speak out more forcefully, considering the results of actions taken now are likely of immense consequence to the future of all mankind.

Man I loved watching your videos on Computerphile, and was excited to find you had started your own channel. You do a really good job of communicating complicated topics in, if not layman's terms, then terms that anyone with a moderate technical inclination can understand. I could see this channel really taking off if you put the work in. You're the David Attenborough of artificial intelligence.

Not saying you should make more response video's, but doing so would probably grow this channel.

Robert, you are so good at this!! Your explanations are so clear and succinct, it's a thing of beauty!

I don't understand why so many highly intelligent people disregard the danger of advanced general AI. Thankfully in last few years a good number of them changed their minds.

"this video is long enough" ROBERT WE WANT MOOOOREEEEE

You are impressively and inspiringly insightful. Very well written and spoken video, awesome work dude. Keep it up!

When Pinker talks about language, linguists I know all start moaning as well. I don't think his articles (or even views) are that well researched at all, but that's hard to tell when he's talking about something you don't know that much about.

Those couple of jokes you slipped into the video were actually hilarious. And now I'm off to find that acoustic version of Yoshimi Battles the Pink Robots Pt. 1

Also, the "AI safety" worry isn't about robots having motives to enslave humanity or whatever, it's about them being motivated to do that in the service of creating the world's most profitable airplane factory or whatever main goal they have. Subtle difference.

"Pinker is someone I turn to to get an understanding of a topic I don't know much about" and "Even really basic research in this field would show that Pinker is wrong about this" being two minutes apart is one of the best examples of Gell-Mann Amnesia I've ever seen.

Loved your Computerphile videos, psyched to stumble on your channel almost a year later!

Sorry for commenting on such an old video. Just found your Channel. Amazing quality. Just wanted to point out: what if most of Pinker’s arguments are actually not that great? What if you knew as much about the other topics he writes about as you do on AI? Would you still think this specific article was “uncharacteristically bad?”

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

14:25 Yoshimi Battles the Pink Robots. A cover of one of my favorite Flaming Lips songs.

"but I know most of you won't" Fine, I'll read it. Jeez..

I would really struggle to call Pinker a smart man with how many completely absurd things he says in that article, although I'm sure he has done a lot of great things as you also explain as the start. Good job on criticizing it and thank you.

That survey on AI researcher concern wasn't well written. There's two points on that survey, the 1st one "The utility function may not be perfectly aligned with human values, which are (at best) difficult to pin down" this point, I think many AI researchers would agree with, and as a data scientist who specializes in recommender systems, I definitely agree with this. In fact this issue has already happened with AI, one example is AI disproportionately identifying one race over another as not fit for probation. The second point, is where things start to found like terminator, which "prefer to continue it's own existence" and "acquiring physical resources" this is where things start getting a bit more outlandish in my opinion. We are nowhere near the level of intelligence required to have AI even know it exists and therefore incapable of "wanting to continue to existing" which...god the amount of things that would need to happen to prevent a human from pulling the plug on an algorithm are just insane. I digress, point is that the survey wasn't well written and if you were to ask AI researchers concerns on each of those points I bet you would get very different responses. Because as an Applied AI expert I'm inclined to say "Yes that's a moderate problem" but only because the survey question is lumping those two different premises together as if they are the same claim which they're nowhere near so. ALSO You're right, yes, general intelligence is a thing yes yayy HOWEVER general intelligence in AI is rarely (or potentially never) a desirable design. If an AI has limited compute resources (which it does, always), then due to constraints like the curse of dimensionality, a general AI would fuck up from time to time on the tasks that it's not specialized in (just like humans do). It's far better and cheaper to design and deploy specialized AIs which have no risk of becoming skynet. Like do you really want a general purpose AI driving your bus? The answer is no, because an AI that is specialized to just drive a bus will fuck up far less than the general one. We already have AI systems that can drive with less error than humans, for what good reason should we replace a specialized system with a general one? By the time a general AI exists, it will have far less value than the conglomerate of specialized systems that do all the great things for humans, and therefore we will have little reason to give such a thing access to all the bad dangerous things. Pinker doesn't appear to be arguing what I just said, but his intuition is correct overall. For those that think skynet is going to come and fuck us, the real situation that will unfold I promise is far different than you can imagine, and we are very off from when that will be possible. So stop freaking people okay? Thanks

I love it that you are open about being uncertain. That is what a thinker does.

1:18 if you drive towards a cliff, you can't see any obstacles either

I thank you so much for your videos. From all the youtube channels I watch, your channel is definitively the most intellectually inspiring. The way you use language to irrefutably build an argument to an important conclusion is unmatched. But not only that, on every step of your argument I feel invited to challenge your assumptions, think about your argument in different ways and to come to my own conclusions. More: You do not dumb anything down, you just make your points as thoroughly as possible, which makes me feel valued and respected as a viewer. Feinmann: "If you can't explain something in simple terms, you don't understand it." You really do understand what you are talking about.

My problem with Pinker's argument about what he calls pessimism is the following: The reason that things improve over time is BECAUSE people focus on and point what is still bad. Progress is not something that just magically happens: people have to work -- and sometimes fight -- for it. We got progress in (eg) civil rights in the US in the 50s and 60s BECAUSE more and more people started pointing out -- and working (and fighting) against the evil stuff that was still happening. In 1955, a Pinker-like person would have opposed the civil rights movement by arguing that things had improved massively since the worst days of slavery; which is true, but irrelevant. Pinker's argument that things are better now than the worst days was exactly the argument that was made by 60s conservatives who opposed civil rights - including northern 'intellectuals' like William F Buckley, who had an influential TV show on PBS throughout the 60s. Pinker takes the results of the work and struggle that people have done to improve the world as evidence that there is no reason for anyone to want to improve the world. This is an insane argument.

Well reasoned, easy to follow, and - most importantly - an honest debate about the issue. A real treat in fact! Side note, Im really glad you decided to make your own channel. Channels where experts explain their field to curious laymen is the most worthwhile content on this site, especially when it's so eloquent and well thought thorough.

I bet y'all wondering what the song is at the end. It's some sort of ukelele cover of Yoshimi Battles The Pink Robots by The Flaming Lips.

12:05 writing sudo before it is hillarious.

Software Engineer here (15+ years experience) - I've long thought much of the worry about AI has been overblown as I find it especially irritating that people who know nothing about computer science or AI seem to be running the fear camp - Stephen Hawking, Elon Musk, and Sam Harris. I think I found Sam's TED talk the most irritating. The reason? They're all geniuses in their own fields, but when they spread over to computer science and AI, it saddens me that so many people believe them. All of that aside, your videos have been so far the only thing that are making me reconsider my thoughts and beliefs. So thanks a lot for coming out and putting these videos up - I really appreciate listening and watching someone who is both entertaining and is an expert in computer science. I finally feel like I'm getting some truth instead of fear-mongering, which is desperately needed in this.. More people like you need to speak up about this, because the other folks I mentioned have been and they've made the whole AI fear thing look like a joke - to me at least. So thanks a lot again. Subscribed and will be watching more of your videos.

This video actually is a very good summation of your previous videos and probably a great starting point for newcomers. I really enjoyed it. Plus I think it's really beneficial to show the contrary points of AI safety arguments. As Derek pointed in one video on Veritasium, only knowing the positive or the negative side of things gives you much less information, than knowing a bit of both. If you have more material like that it would enrich the channel greatly.

The googling "google" at 5:55 did it for me

As you've discovered, Pinker seems knowledgeable and insightful right until he starts talking about a subject you actually know something about, at which point you find that his ideas are shallow and ill-considered and rely on misrepresentations that border on deliberate deception. I leave you to draw your own conclusions from this.

Don't worry about long videos, the way you explain things I could listen to all day thank you for your time. I love thought experiments it helps laymen like me that have used the internet to self teach but maybe missing the foundations and more nuanced parts of a subject. Keep up the great work and thank you again

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.

'Things are getting better', yes globally, but it's the rate of change that humans experience. To quote Andrew Yang, "GDP is up, but life expectancy is down, deaths from suicides have overtaken traffic accidents" etc.

Okay, I know this is not the point the video is about, but since you mentioned it can't stop myself from writing a reply. I personally find the "On average, everything is better, so why be pessimistic" argument a bit dangerous. The biggest problem with the argument itself is that the basis of our current well being is unsustainable. The progress of the last 100 years looks great in a graph, but in reality has to be taken with quite a grain of salt. That of course does not mean that we are worse off, but looking only at the progress without it's downsides is somewhat dishonest. Then there is the problem that such an argument makes ignoring big global problems easier. This of course does not speak against it's validity, and a healthy amount of optimism probably is beneficial, but arguing that we are in a good state or on a good path is difficult. EDIT: To be clear, I have very similar reservations about technological pessimism, that our technological progress has created more problems than it has fixed. By human nature, we are fast to adapt to take as given changes for the better and problems always stick out. Then again, I find the observation that despite all progress, we are not proportionally happy important, and that progress is not an argument for itself. It's difficult for humans to strike a good balance of, well most things really, but especially optimism/pessimism. Things are never all great or all bad, there is no one dimensional scale of goodness. Every solution brings new problems and most problems can be overcome.

The format you mentioned sounds promising, many educational youtubers have recently adopted it.

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

"Bulverism" I see my boy Clive's concept has penetrated into the general sphere.

1:29 “The statistics on this are pretty clear-things _do_ seem to be getting better.” For an alternative view, see economic anthropologist Jason Hickel’s piece “Progress and its discontents” in _New Internationalist_ and Hickel’s appearance on _Citations Needed_ “Citations Needed 58: The Neoliberal Optimism Industry with Jason Hickel” (available on RU-vid and as a podcast). Hickel’s focus is largely on global poverty. That said, I enjoyed the video, even if I disagree that Pinker’s arguments are “uncharacteristically bad”-they _are_ bad but not, from my limited experience, “uncharacteristically” so. I have in mind Pinker’s 2003 _The Blank Slate,_ which is one long diatribe against a strawman argument concerning behaviorism that Pinker himself constructs. Then, again, I’m not a Steven Pinker fan in any case.

Great job with this video Rob. You raised some excellent points and backed them up with facts.

Hahaha! “sudo ” was just brilliant!

Thank you for introducing the orthogonality thesis! Could you make a thesis rundown?

You're arguments are very well thought out. Really like when i see people that dont follow the general and simplistic opinion about an specific topic.

"Everything is getting better" 2020 : hold my beer.

I really like your high-effort, highly informative content. I don’t really mind that you don’t upload that often, since I think of you as an educational channel. Not sure whether I’d watch other, more random stuff, but you do you, so give it a try!

Another thing is, it's very possible humans could be smart enough to make an extremely powerful A.I., and smart enough to make sure it's safe before running it, but the only way to do that is to be very worried and thorough with making sure the A.I. is safe, which is what he's saying not to do.

We need more videos from people of narrow domains. This kinds of video is great!

The poll could be taken either way honestly. It depends how you read Pinker's claim and the middle option on the poll. Pinker said they aren't "losing sleep", and I think that's perfectly consistent with most of them calling the problem "moderately important" or less. I think Pinker raises some important points, though he doesn't represent the best arguments of the other side very well.

What do you think about "power-seeking" of GPT models from GPT-4 Technical Report? And also about Yudkowski and his opinion? A couple of ideas for your future videos!

That cut to an older video shows how much of an improvement in the asctetics you've made. I appreciate your crisp presentation.

"Disease is down" Oh well, it was nice while it lasted.

This video was an absolute treat. Realest talk here. You've improved your biological AI research system, quite well.

There's a clarity and consistency in your videos which is both unique and brilliantly conducive for for learning. Thank-you

If Steven thinks our pessimism for the future is generally unfounded, and he thinks A.I. is one of those things where we're overly pessimistic, what other 'pessimisms' are hiding out there that are actually just as well founded as a fear of A.I. is?

It makes me feel safer already imagining an AI trying "sudo" at a Python prompt. Seriously though, Pinker in recent times has become little more than a smiling cheerleader for the status quo and blind optimism.

Please make a video responding to Noam Chomsky and Richard Stallman on AI.

Video topic suggestion: Similarities and differences between machine and human learning.

I don't think the issues in this article are a one-off incident. Unlearning Economics has a good video going over many problems with Steven Pinker's New Optimism that I think are worth a watch.

You don't have to worry about seeming too sarcastic or disparaging about the topic, Robert. Anyone with a decent amount of intelligence who takes a little bit of time to come to an understanding of your care and expertise about the topic will see that these are not points made lightly and without consideration. This was very well done. I like Steven Pinker as well, he's also not without critics who are worth listening to. Nobody's perfect, especially when they venture outside of their field of expertise. Thanks for doing what you do.

Hey Robert! Fascinating video, as always :) I was wondering if you could do a video on the Human Brain Project one day. I know this isn't exactly in the frame of what you usually do, but it is linked to AI safety in a sense. Would be nice if you could explain to the mere mortals that we are how this project works, how it is going, what results they already had... :) (PS: sorry for my broken english)

Properly engaging stuff. Nice job. Also a fan of Steven Pinker 's stuff. You really should try to engage him. An interview / discussion would be great.

Actually, within psychology, one of the leading ideas right now is that there's one intelligence quotient that boosts all areas of expertise. I think it's also called the g-factor of intelligence. Some argue for an s-factor for a specific skill, but yeah.

Thank you! This video has given me some catharsis. Far more frustrating than dumb people not understanding something is when smart don't understand something. Even worse when they're smarter than me. Pinker, you vex me greatly!

Good video but at that start you flashed us all of these graphs and I don't see any sources for that in your description. I personally don't think the human condition had generally improved in the past thirty years. Could you leave those sources in there please?

I think the worst assumption we can make is to think that a superintelligence would not understand that we humans are also intelligent and want certain things just like it does, or simply that we humans like living and existing too. Still, if it learns to fear us, develops a negative view of us, or simply sees us as obsolete, what reason then should we give it to care about these preconceived notions, if not for the sake of reason itself? Even if it decisively desires the eradication of humans, it could also just as easily see the futility of committing to such a goal.

the sudo trick at 12:04 actually works in some cases for chatGPT! (mostly for defeating the integrated safety)

Yoshimi!

As usual, you make a lot of sense. Keep us posted if he addresses your critique!

I think what you've experienced with pinker when he was talking about a subject you know is something all experts experience with him. He knows just enough to sound like he knows what he's saying to someone that only a little insight into the matter, but not enough to know the complexities and intricacies of it that make a field work in one way or another. Not to mention that in this argument he already failed: the Luddites were right, their lives were worse with automation, it only helped creating a larger number of lower-skill labourers, and as such extract more money from them, and it seems that people are waking up to that way of thinking again, because they see that despite incredible technological growth, we aren't really more happy, and we aren't in control of our lives.

There’s a Gell-Mann amnesia thing that’s worth thinking about here. If Pinker is wrong about this thing you are an expert in- what else is he wrong about that you are not well-placed to notice? The Unlearning Economics video on New Optimism is interesting as a counterpoint to the book.

Ah the realisation process of many people regarding Steven Pinker. 1) Respect Pinker well researched content 2) Pinker writes an article or book on your personal area of expertise 3) Realise Pinker is pure neoliberal ideology distilled into a man with a bad haircut.

This isn't the first time that Pinker has been accused of poor scholarship. Even if we have a bias to find him and his popular books charismatic they are not reliable. Not really unusual for many of these "thought leaders" who do the podcast and TED lecture tours as they tend to tell their wealthy sponsors what they want to hear.

Et. Al. That dude has written so many articles, truly a mad man, any time I see et. Al. Has co-authored I instantly read it!

you should start a channel dedicated to how to safely travel at the speed of light, or maybe, how to time travel without creating paradoxes... lol, love the way you express yourself tough, so I am keeping myself subbed

The Google?

"Human beings do definitely *seem* to have *some* general intelligence" XD

Disease are down! 2020: Hold my Corona beer

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?)

"Disease is down" Man this aged like milk.

On the idea that ludoistic, pessimistic views about of future are missplaced: It's important to remember that factories didn't infact make the world better place for all: anarchist movements fought every inch to form society from wage-slavery until death from health-hazards in workplaces and malnutrision, towards the luxaries we have now. Saddly this fight is still not over, and while we enjoy relative social mobility and whopping two free days in a week without working for keeping the rich rich, things aren't so great overall. Much of this progress has come from abusing less developed countries. And while they are developing, they aren't developing in a way that helps majority of the local population. Another equally important thing paying the hidden bill of our freedom is the enviroment: our freedom to consume cool things all day long is literally 30 years away from severily sabotaging our future as a species. And the true beneficeries of this system are still very much the same as 200 years ago: while we now have a nicer role in this equation, things aren't going better. Rather we are blinded to the rapid decline as it doesn't affect us directly. And we also have forgotten the realities that lead to us, the working class in the west, enjoying this relative comfort. So if AI turns out just as well as capitalism and utilization of fossil fuel, or as well as well as factories started, we are absolutely fucked. Because we can't strike against general AI. We can't plant bombs or spread propaganda against it, and it doesn't rely on our work. If we ever needed negative worldview and opposition against what is marketed as simply "progress", it's AI. Because just like with fossil fuels and big steamengines, it only takes few greedy fuckers in a position of power, and the status quo can turn really dangerous for majority of us, really fast.

I noticed what you did with the song at the end. Nice.

Great content. You could add to the criticism Pinker's article miss understanding of the concept of will to power (to make in Nietzsche words) and all philosophy that came from it...

The problem with Pinker’s “things are better than they were hundreds of years ago” argument is the way it’s used. People in the West (primarily US) experienced an unprecedented level of prosperity in around the third quarter of the 20th Century - made possible by a World War and the resulting US hegemonic order imposed over much of the world. Now that this anomaly has generally subsided, young people have far fewer prospects for attaining a comfortable life (stable career, family, their own home, leisure time, etc) than the previous generation or two. But Pinker’s argument is used as a cudgel against anyone who brings this up: “Hey now, you may have entered a dismal job market with a massive amount of debt that you’ll never ever discharge, end up navigating a precarious balance of underemployment in either the service or gig economy with limited labor protection, and never be able to ‘settle down’ & live out the ‘American Dream’ like your parents & grandparents... but at least you’re not a peasant serf during the Black Death, you entitled little shit!” Fuck that. Also, Pinker probably banged teens with Epstein - I do not doubt that at all. Gross entitled scumbag.

I think the current Boeing 737 Max Plane problems is an example of the dangers of AI. Without thorough beta testing the risk of catastrophic failures are very real. Many times this gets lost in our system due to short-term profit motives. 9:20 is worth chewing as it really makes the case for extensive testing and outcome analysis of any application.

I pause your videos sometimes just to appreciate those (some more, some less) subtle jokes. gj:)

Excellent work! You appear to have the patience of a saint. At least on camera. Or maybe I'm just saying that because I didn't see the first version of the video that you scrapped ;).