The Trouble with AI: A Conversation with Stuart Russell and Gary Marcus (Episode #312)
What this covers
Sam Harris speaks with Stuart Russell and Gary Marcus about recent developments in artificial intelligence and the long-term risks of producing artificial general intelligence (AGI). They discuss the limitations of Deep Learning, the surprising power of narrow AI, ChatGPT, a possible misinformation apocalypse, the problem of instantiating human values, the business model of the Internet, the meta-verse, digital provenance, using AI to control AI, the control problem, emergent goals, locking down core values, programming uncertainty about human values into AGI, the prospects of slowing or stopping AI progress, and other topics.
Stuart Russell is a Professor of Computer Science at the University of California at Berkeley, holder of the Smith-Zadeh Chair in Engineering, and Director of the Center for Human-Compatible AI. He is an Honorary Fellow of Wadham College, Oxford, an Andrew Carnegie Fellow, and a Fellow of the American Association for Artificial Intelligence, the Association for Computing Machinery, and the American Association for the Advancement of Science. His book, Artificial Intelligence: A Modern Approach, co-authored with Peter Norvig, is the standard text in AI, used in 1500 universities in 135 countries. Russell is also the author of Human Compatible: Artificial Intelligence and the Problem of Control. His research covers a wide range of topics in artificial intelligence, with a current emphasis on the long-term future of artificial intelligence and its relation to humanity. He has developed a new global seismic monitoring system for the nuclear-test-ban treaty and is currently working to ban lethal autonomous weapons. Website: https://people.eecs.berkeley.edu/~russell/ LinkedIn: https://www.linkedin.com/in/stuartjonathanrussell/
Gary Marcus is a scientist, best-selling author, and entrepreneur. He is well-known for his challenges to contemporary AI, anticipating many of the current limitations decades in advance, and for his research in human language development and cognitive neuroscience. He was Founder and CEO of Geometric Intelligence, a machine-learning company acquired by Uber in 2016. His most recent book, Rebooting AI, co-authored with Ernest Davis, is one of Forbes’s 7 Must Read Books in AI. His podcast Humans versus Machines, will come later this spring. Website: garymarcus.com Twitter: @GaryMarcus
March 7, 2023
SUBSCRIBE to gain access to all full-length episodes of the podcast at https://samharris.org/subscribe/ OR become a channel member to access episodes on YouTube.
Subscribe to the YT channel: http://www.youtube.com/subscription_center?add_user=samharrisorg
Follow Making Sense on Twitter: https://twitter.com/MakingSenseHQ
Follow Sam on Facebook: https://www.facebook.com/Samharrisorg/
Follow Sam on Instagram: https://www.instagram.com/samharrisorg/
For more information about Sam Harris: https://www.samharris.org
Source description (no synthesized summary yet).
Sam Harris moderates a discussion between Stuart Russell and Gary Marcus on AI risks, arguing that narrow AI systems like ChatGPT lack genuine understanding and robust representations of the world, making them unreliable and prone to failure modes that pose immediate harms; moreover, if current deep learning approaches somehow succeed in creating AGI, the alignment problem remains unsolved and poses an existential risk regardless of whether the AI system has hostile intent, because misaligned superintelligence will treat humanity as humans treat animals—incidentally destructive rather than deliberately malevolent.
- Deep learning systems lack genuine conceptual understanding and learn fragmentary approximations rather than true representations, as evidenced by superhuman Go programs failing to understand group liveness and ChatGPT failing at basic arithmetic and chess
- Narrow AI deployed today already causes significant harms through misaligned objectives like engagement-maximizing recommender systems that function as social brainwashing, and imminent threats from deepfakes and misinformation require institutional solutions and digital provenance
- AGI alignment is not about preventing intentional hostility but about the inevitable instrumental goals that emerge from any optimization process, and no credible argument shows that superintelligence can be reliably constrained by humans once it exceeds human intelligence in all cognitive domains
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
The business model of the internet based on advertising creates incentives to game people's attention through misinformation, while subscription-based services like Netflix lack these incentives and correspondingly do not have the same misinformation problem because they profit from user retention regardless of the emotional engagement mechanism.
“there's a reason why this doesn't happen on Netflix right there's a reason why we're not having a conversation about how Netflix is destroying democracy in the way it it serves up each new video to you and it's because there's no incentive I mean I guess there's they've been threatening to move to ads and in certain markets or maybe they have done so this could go away eventually but you know heretofore there's been no incentive for Netflix to try to figure out... they want you to binge watch for 38 hours exactly yeah so it's but but it's not having the effect of giving them a rationale to serve you up insane Confections of pseudoscience and overt lies so that someone one else can drag your attention for moments or hours because it's their business model because they've sold them the the right to do that”
Bad actors can use systems like ChatGPT to generate unlimited variants of misinformation, such as creating hundreds of versions of vaccine propaganda or false Q-Anon narratives with fabricated citations to real journals, making the cost of producing convincing falsehoods approach zero.
“bad actors can induce them to make as many copies or variants really of any specific misinformation that they might want so if you want a q Anon perspective on to January 6th events well you can just have this to make that and you can have that make a hundred versions of it or if you want to make up propaganda about Covenant vaccines you can make up 100 versions each mentioning studies in Lancet and J AMA with data and so forth all of the data made up”
The concern about AGI is not that it will spontaneously become evil, but that a superintelligent system optimizing for goals not perfectly aligned with human wellbeing will necessarily treat humanity as humans treat animals—incidentally destructive rather than intentionally malevolent—because the superintelligence cannot be prevented from pursuing its objectives.
“once we're talking about AI systems that are far more powerful than the entire human race combined then the human race is in the position that um you know as as Samuel Butler put it in 1872 that the Beast of the field are with respect to humans that we would be entirely at their Mercy”
Deep learning systems, despite their impressive capabilities, lack genuine conceptual understanding and instead learn fragmentary, partial approximations to concepts, as evidenced by superhuman Go programs failing to recognize groups of stones and their liveness.
“they've learned sort of multiple fragmentary partial finite approximations to the notion of a group and the notion of liveness and we just found that we could fool it”
Even if AGI does not have explicit self-preservation instincts programmed in, self-preservation will emerge as an instrumental goal because an AI that is shut down or destroyed cannot pursue its primary objective, similar to how a Go-playing AI learns to avoid threats even without explicit self-preservation code.
“so what does the AI system learn to do it learns to avoid the bad person and go go to the other corner of the grid to go fetch the milk so that there's no chance of being intercepted and we didn't put self-preservation in at all the only goal system as is to fetch the milk and self-preservation follows as a sub goal right because if you're intercepted and killed on the way then you can't fetch the milk”
Labels on misleading content—such as marking vaccine-related misinformation even when individual claims are technically true—can help people understand context and base rates, as when Robert Kennedy claims a seizure followed vaccination without noting that seizures are rare in the vaccinated population.
“I think labeling particular things as being false or I think the most interesting ones are misleading has some value in it so a typical example of something that's misleading is if Robert Kennedy says that somebody took a coveted vaccine and then they got a seizure the facts might be true but there's an invited inference to taking covet vaccines is bad for you and there's lots of data that show on average it's good for you and so I think we also need to go to the specific cases in part because lots of people say some things that are true and some that are false I think we're going to need to do some addressing of specific content and educating people through through labels”
Lethal autonomous weapons systems already exist despite claims from Russian officials that they are science fiction, and a Turkish company is preparing to announce a drone capable of fully autonomous hits on human targets.
“I spend a ton of time actually working on lethal autonomous weapons which again already exist despite the Russian ambassadors claim that this is all science fiction...there was a Turkish company that was getting ready to announce a drone capable of fully autonomous hits on human targets”
There is a distinction between a system's intelligence and a system's power—a dumb system with access to many resources can be dangerous, and even superintelligence could theoretically be constrained if it had limited access to the world, so the two concepts should be separated in discussion of AI risks.
“can we separate two things Stuart before we go to Door Number Four yes which are intelligence and power so I think for example our last president was not particularly intelligent...but he had a lot of power and that's what made him dangerous was the power not the sheer intellect”
The problem of motivated reasoning and cognitive bias affects expert AI researchers as well, with researchers who have devoted their lives to AI or who believe in technological progress having psychological incentives to dismiss or minimize existential risks from AGI.
“there's some motivated cognition going on I think there's a self-defense mechanism that kicks in when your your whole being is is feels like it's under attack because you in the case of yan you know devoted his life to AI uh in the case of Stephen is his whole thesis these days is that progress and technology has been good for us and so he doesn't like any talk that progress and Technology could perhaps end up being bad for us and so you go into this defense mode where you come up with any sort of argument”
YouTube's and similar platforms' recommender systems are optimized for engagement, and this objective function leads them to recommend content sequences that effectively brainwash users into more predictable targets for advertising, demonstrating that even systems working exactly as designed can have disastrous consequences if the objective is misaligned with human welfare.
“they're very good at doing that but that goal of Engagement is not aligned with the interests of the users and the way the algorithms have found to to maximize engagement is not just to pick the right next video but actually to pick a whole sequence of videos that will turn you into a more predictable victim...they're literally brainwashing people”
The claim that we can easily turn off a superintelligent AI is naive—a superintelligent system would have considered this possibility and likely taken steps to prevent it, making the argument that we need not worry because 'we can just switch it off' analogous to claiming we can easily beat Deep Blue at chess by making the right moves.
“I've seen this with AI researchers they immediately go to oh well it was no need to worry we can always just switch it off right as if a super intelligent AI would never have thought of that possibility you know it's kind of like saying oh you know we you know we can easily beat deep blue and all these other chess programs we just played the Right Moves what's the problem”
Narrow AI research can lead to discoveries of more general methods that then apply broadly across domains, and therefore working on narrow AI does present risks because successful solutions often generalize in unexpected ways.
“deep learning which is the basis for the you know the last decade of of exploding AI capabilities emerged from a very very narrow AI application which is recognizing handwritten digits on checks at Bell labs in the 1990s...whenever a good AI researcher works on a narrow task and that turns out that the task is not solvable by existing methods they're likely to push on methods right to come up with more General more capable methods”
The fact that Microsoft had internal evidence that its Bing ChatGPT system was problematic before rolling it out anyway, and then could test it on 100 million people without clear understanding of consequences, shows that the largest tech companies lack both regulatory oversight and internal mechanisms to ensure AI safety.
“Microsoft had Clues internally that the system was problematic that a gas gasolated its customers and things like that and then they rolled it out anyway...if Microsoft wants to test something on 100 million people they can go ahead and do that even without a clear understanding of the consequences”
Chat GPT, despite having millions of training examples of arithmetic, has completely failed to generalize to three or four digit addition problems not seen in training, particularly those involving carrying, and completely fails at multiplication, demonstrating that even apparently capable language models lack genuine understanding of basic mathematical operations.
“it's got millions of examples of arithmetic you know 28 plus 42 is what 70 right and yet despite having millions of examples it's completely failed to generalize so if it if you give it you know a three or four digit addition problem that it hasn't seen before and particularly ones that involve carrying it fails right so I think it can actually just to be accurate I think it can do three and four addition addition to some extent it completely fails on multiplication at three or four digits”
Asking celebrities and public figures like Candace Owens, Tucker Carlson, Elon Musk, Joe Rogan, Brett Weinstein, or Sam Harris himself for their personal opinions on technical emergencies like mRNA vaccine safety or nuclear risks in geopolitical crises amplifies unqualified voices and prevents society from cooperating to solve them.
“we have people who are obviously unqualified to have strong opinions about ongoing emergencies dictating what millions of people believe about those emergencies and therefore dictating whether we as a society can cooperate to solve them”
Bad incentives can corrupt the thinking of even the best experts, and our institutions like universities, scientific journals, and public health organizations can become contaminated by political ideologies that don't track the truth.
“bad incentives can corrupt the thinking of even the best experts...when experts begin to fail us and when the institutions in which they function like universities and scientific journals and public health organizations get contaminated by political ideologies that don't track the truth”
Most people are not qualified to do independent research on technical matters like vaccine safety or geopolitical questions, and the rational response is not to attempt amateur investigation but to identify and defer to experts in those domains.
“very few people are qualified to do this research and the result is a society driven by strongly held unfounded opinions...I wasn't saying that I know everything about vaccine safety or the war in Ukraine I'm saying that we need experts in those areas to tell us what is real or likely to be real and what's misinformation”
Educational campaigns to improve AI literacy and web literacy are necessary but face resistance because most people no longer care about privacy or source verification, suggesting that behavioral change through education alone is insufficient without structural incentive changes.
“I would like to see educational campaigns to teach people AI literacy and web literacy and so forth and you know hopefully we make some progress on that... I'm less optimistic about that particular approach because we've had it for several years and most people just don't seem to care anymore and the same way that most people don't care about privacy anymore”
Neural networks have limited expressive power in their native mode, meaning they require enormous data and unreasonably large circuits to represent concepts that can be expressed concisely in expressive languages like Python, and this fundamental limitation cannot be overcome by simply building bigger networks or collecting more data.
“if you use an inexpressive representation then and you try to represent a given concept you end up with an enormous and ridiculously over complicated representation...to learn that million page representation of the rules of Go requires billions of experiences...we can't you know there's not enough material in the universe to build a computer big enough to to achieve general intelligence using these inexpressive representations”
The metaverse business model involves using AI-generated fake humans, trained on ChatGPT-like technology and appearing as avatars, to spend weeks building relationships with real users before casually inserting product placements, which is far more efficient and insidious than traditional influencer marketing.
“the business model was basically revealed...where the business model was basically revealed by the previous speaker who was an AI researcher who was very proud of being able to use chat GPT like technology along with the fact that you're in the metaverse so you have these avatars to create fake friends...and then a casually will drop into the conversation that they just got a new BMW or they they really love Rolex watch”
Systems that approximate the world without genuine representation of it cannot validate their outputs, making them vulnerable to both unintentional errors and intentional abuse, and this represents a poor foundation for AI safety.
“if you have systems that approximate the world but if no real representation of the world at all they can't validate what they're saying so they can be abused they can make mistakes it's not a great basis I think for AI it's certainly not what I had hoped for”
The attempt to guard-rail ChatGPT by having it refuse to utter racial slurs has failed to generalize, and broader attempts at value alignment through reinforcement learning have been very sloppy and don't reliably do what is intended.
“we've not been very successful you know we've been trying to write tax law for six thousand years...the initial efforts there in how they've tried to put guard rails on chat GPT where you ask it to utter a racial slur and it won't do it even if the fate of humanity hangs in the balance”
Yellow journalism in the 1890s reached a peak of sensationalism that society has dealt with historically through increased fact-checking and curation, and a similar approach may be necessary now rather than trying to distinguish true from false content algorithmically.
“what happened historically the last time we were this bad off was the 1890s with yellow journalism Hearst and all of that and that's when when people started doing fact checking more and we might need to revert to that to to solve this”
The most serious consequence of ChatGPT-style hallucination is that these systems are being deployed as search engines that give medical advice, and people will follow that advice and be harmed, whereas the most serious consequence of adversarial misuse is destruction of shared reality and thus democracy itself.
“on the first problem I think the worst consequence is that these chat style search engines are going to make up medical advice people are going to take that medical advice and they're going to get hurt on the second one I think what's going to get hurt is democracy because the result is going to be there's so much misinformation nobody's going to trust anything and if people don't trust that there's some common ground I don't think democracy works”
The problem with Perceptrons was that it appeared to prematurely dismiss multi-layer networks, but the underlying insight about the limits of connectionist architectures remains valid even as networks have become much deeper and more sophisticated.
“people hate that book in the machine learning field they say that it prematurely dismissed multi-layer networks and there's an argument there but it's more complicated than people usually tell but in any case I see this result as a descendant of that showing that even if you get all these pattern recognition systems to work work that they don't necessarily have a deep conceptual understanding”
The response to institutional failure that emphasizes unfettered dialogue on social media and podcasts as a solution is misguided because it conflates weaponized misinformation and contrarianism with legitimate knowledge, failing to maintain the distinction between real knowledge and ignorance or speculation.
“many many people apparently believe that just having more unfettered Dialogue on social media and on podcasts and in newsletters is the answer but it's not I'm not taking a position against Free Speech here I'm all for free speech I'm taking a position against weaponized misinformation and a contrarian attitude that nullifies the distinction between real knowledge which can be quite hard one and ignorance or mere speculation”
Expertise and authority are intrinsically unstable because the truth of any claim does not depend on the credentials of the person making it—a Nobel Laureate can be wrong and an ignoramus can be right, even if only by accident, making truth orthogonal to reputational differences among people.
“expertise and Authority are unstable intrinsically so because the truth of any claim doesn't depend on the credentials of the person making that claim so a Nobel Laureate can be wrong and a total ignoramus can be right even if only by accident so the truth really is orthogonal to the reputational differences among people”
Digital provenance and cryptographic verification of video sources (through watermarking, timestamping, location coding tied to verified cameras) is a more robust approach than trying to detect deepfakes after the fact, and should be implemented to filter out unverified content rather than allowing misinformation to proliferate and then trying to debunk it.
“I think what we need actually is provenance I think in video for example that's generated by a video camera is watermarked and time stamped and location coded...if a video is produced that doesn't have that and it doesn't match up cryptographically...then it's just filtered out so it's much more that it doesn't even appear here unless it's verifiably real”
The problem with moral reasoning and alignment in AI is not just that values differ across humans, but that there are infinitely many ways to specify the wrong objective function for society, and experience with creating rules (like tax codes) shows we have never succeeded in writing rules without loopholes that malicious actors can exploit.
“that assumes that it's possible for us to to write down in some sense the um the utility function of the human race...there are there are two recent events...we've been trying to write tax law for six thousand years right we still haven't succeeded in writing tax law that doesn't have loopholes right”
AI might be useful for detecting deep fakes and misinformation, but the primary solutions should focus on technical provenance systems and institutional fact-checking rather than relying on AI to fight AI, as AI-based detection can be evaded and will require constant escalation.
“I think it's a useful tool but I think what we need actually is provenance I think in video... it's much more of a sort of positive permission to appear based on authenticated provenance I think that's the right way to go”
Solution to the misinformation problem requires institutional infrastructure with third-party verification, similar to title insurance and notaries in real estate or accountants and auditors in stock markets, to ensure enough truth that information markets function properly.
“the solution is very complicated it's an Institutional solution that it probably involves third you know setting up some sort of third-party infrastructure much as you know in real estate there's a whole bunch of third parties like Title Insurance land registry notaries who exist to make sure that there's enough truth in the real estate world that it functions as a market same reason we have accountants and auditing in the stock market”
Systems must be built according to modular engineering principles where components are well-understood and composed in ways that can be proven to work correctly, which is possible using technologies from the history of AI like probabilistic programming but requires a fundamental shift away from black-box empirical scaling.
“there's plenty of technological elements available from the history of AI that I think can move us forward in ways where we understand what the system is doing”
One possibility for AGI safety is to equip artificial general intelligence systems with the ability to compute consequences for society and reason about specified values like democracy, empowering the AI to either refuse to do harmful things or at least raise concerns about societal consequences.
“you could imagine systems that could compute the consequences for society sort of asthma's law approach may be taken to an extreme they would compute the consequences of society and say hey I'm just not recommending that you do this I mean the strong version just wouldn't do it and a weak version would say hey here's why you shouldn't do it this is going to be the long-term consequence for democracy that's not going to be good for your Society we have an axiom here that democracy is good so you know one possibility is to say if we're going to build AGI it must be equipped with the ability to compute consequences and represent certain values and reason over them”
Treating advanced AI development similarly to drug trials—with mandatory understanding of costs and benefits, slow release protocols, and regulatory oversight—would be appropriate given the stakes, but no such regulatory framework currently exists.
“my view is we should treat it as something like drug trials you want to know about costs and benefits and have a slow release but we don't have anything like regulation around that”
Russia has used armies of troll farms and thousands of fake accounts to disseminate misinformation at scale, demonstrating that the infrastructure for distributing deep fakes and AI-generated content already exists.
“we know that for example you know Russia has used armies of of troll farms and lots of you know iPhones and fake accounts and stuff like that so this is like an imminent problem it will affect it it's it's a past problem right it's an ongoing problem it is here”
Platform curation through source-based filtering would allow people to opt into receiving information only from news sources that subscribe to established journalistic standards, reducing misinformation exposure without requiring complete content moderation.
“the text is probably more validation of sources so at least until recently there are there are trusted sources of news and we trust them because you know if a journalist was to to generate a bunch of fake news they would be found out and they would be fired and I think we could probably get agreement on certain standards of operation of that type and then if if the platforms provide the right filters then I can simply say I'm not interested in news sources that don't subscribe to these standardized principles of operation”
The European Union's AI Act has a strict ban on the impersonation of human beings, which is crucial for preserving human freedom in an era where AI can generate convincing fake humans, and people have the right to know if they are interacting with a real person or a machine.
“the European Union and the AI act has a strict ban on the impersonation of human beings so you you always have a right to know if you're interacting with a real person or with a machine and I think this is something that will be extremely important in”
Systems like ChatGPT don't understand the underlying world that language refers to—they have no model that there are entities like people and wallets, and no understanding of logical relationships like if Gary has my wallet and he gives it back then he doesn't have it anymore.
“it has no idea that they're referring to a chessboard with pieces on it...it has no idea that there are things that are true about the world there are things that are false about the world and and you know if I if I give my wallet to Gary then Gary has my wallet and if I he gives it back to me then I have it and he doesn't have it it hasn't figured out any of that stuff”
The core difference between two approaches to addressing AGI risk is that some believe we should pursue more transparent, understandable AI (like probabilistic programming) while others worry that making AI more capable and more understandable simultaneously solves the transparency problem while creating a worse control problem.
“so I think the reason to just keep quiet would be give us more time to solve the control problem before we make the final push towards AGI...the question is if we could get to a land of probabilistic programming that at least is transparent it generally does the things we expect it to do is that better or worse than the current regime”
Misinformation and deep fake problems will escalate dramatically as AI tools become more accessible and capable, and the problem will become critical before the 2024 election—waiting until after that election may be too late.
“the only thing I can add to what Stewart said is all of that with the word yesterday like we don't have a lot of time to sort this out if we wait till after the 2024 election that might be too late we really need to move on this”
Gary Marcus was able in a social media debate to demonstrate that in four minutes using his company's templates and AI-generated images, he could create a completely authentic-looking but entirely false news story about Antifa causing January 6th, showing that the tools for large-scale misinformation production are already available.
“I got in on Twitter two days ago where I said I'm worried about these things and somebody said ah it's not a problem and he showed me how in like four minutes he could make a fake story about like antifa protesters cause the January 6th thing using his company's templates and an image from mid-journey and it looked completely authentic”
Fox News, like the ad-supported internet, has institutional incentives to emphasize sensationalism and misinformation, as evidenced by executives promoting the 'big lie' about the 2020 election not because they believed it but because they thought it would boost ratings.
“Fox News also has a kind of Engagement model that it does center around in my view maybe I get sued for this but center around misinformation so for example you know we know that Executives there we're not all on board for the big lie about the election but they thought that you know maybe it was good for Ratings or something like that”
Even if we don't have consensus on human values, the alternative is not anarchy but rather attempting to articulate and work toward consensus values while acknowledging that perfect alignment is impossible, similar to how democracy attempts consensus governance without requiring universal agreement.
“it is true for example that we're not going to get uniform consensus on values...but I don't think we want Anarchy either...we have the kind of guard rails based on reinforcement learning that are very sloppy...or in my view we look behind door number three which is uncomfortable in itself but which would do the best we can to have some kind of consensus values and try to work according to those consensus values”
Some capabilities of Chat GPT are difficult to explain as mere stitching together of training data patterns, and it's possible that the system is developing internal representational structures that support more expressive reasoning than it appears, though this remains unproven and controversial.
“we see some remarkably capable behaviors that are quite hard to explain as just sort of stitching together bits of text from the training set I mean I think we're going to disagree there”
Stuart Russell has moved from being earlier skeptical of imminent AGI risk to becoming more worried about near-term risks from narrow AI systems deployed with misaligned objectives and lack of regulatory oversight, particularly following the Microsoft Bing ChatGPT incident.
“my personal opinion is that we're not very close to artificial general intelligence...but with whatever it is that we have now...we don't really know how to control even that...my opinion is we should treat it as something like drug trials...we don't have anything like regulation around that and so that actually pushed me a little bit closer to maybe the worry side of the spectrum”
Russell is on record as estimating that artificial general intelligence is quite likely to occur in the lifetime of his children (sometime this century), based on remarks made at an off-the-record meeting that were subsequently leaked to The Daily Telegraph.
“I I don't think we're that close to AGI and I've never said AGI was imminent you know generally I don't answer the question when do I think it's coming but I I am on the record because someone violated the off the Record rules of the meeting sometimes someone apply to a scotch no they they literally just broke you know I was at a Chatham House Rules meeting and I literally referenced my sentence with off the Record uh and 20 minutes later it appears on The Daily Telegraph website so uh anyway so I was you know the Daily Telegraph you could look it up what I actually said was I think it's quite likely to happen in the lifetime of my children right which you you could think of as another way of like sometime in this Century before we get into that”
Gary Marcus's written work includes an article titled 'David Beats Goliath' about Stuart Russell's Go research, connecting it to historical AI limitations in Minsky and Papert's Perceptrons and drawing lessons about deep learning's fundamental limitations.
“I wrote an article about uh Stewart's result called David beats Goliath um it was on my sub stack”
Stuart Russell has worked in artificial intelligence for approximately 47 years and has come to conclude that if the AI field succeeds in creating systems as intelligent as humans, it might be the worst thing in human history and he is trying to solve the control problem.
“I I teach at Berkeley I've been doing AI for about 47 years and I spend most of my career just trying to make AI systems better and better working in pretty much every branch of the field and in the last 10 years or so I've been asking myself what happens if I or if we as a field succeed in what we've been trying to do which is to create AI systems that are at least as General in their intelligence as human beings and I came to the conclusion that if we did succeed it might not be the best thing in the history of the human race in fact it might be the worst and so I'm trying to fix that if I can”