ClaimsPeople

Gary Marcus

Gary Marcus is an author and AI entrepreneur and professor emeritus at NYU; he has written six books, including Rebooting AI and Taming Silicon Valley, founded Geometric Intelligence (bought by Uber), testified at the May 2023 US Senate AI hearing, and writes a newsletter with over 121,000 subscribers.

712 claims collected, 213 researched, 103 resolved. Report of 4 November 2026, method 0.1.0+c876030c.

Scores by kind of claim

Each kind of claim is scored on its own, out of 100: the share that held up, with the range it likely falls in. Fewer resolved claims give a wider range.

  • Factual claims

    Statements about the present or the past, checked against the record: did they hold up?

    79 72 85

    High confidence, 64 factual claims resolved. From 18 documents, 2019 to 2026; 19 documents sampled from 37 candidates found.

    Not scored: 32 unresolvable, 4 checkable only against their own records.

  • Predictions

    Statements about events outside the person's control, checked once they fell due: did they come true?

    80 72 88

    Medium confidence, 38 predictions resolved. From 7 documents, 2019 to 2025; 19 documents sampled from 37 candidates found.

    Not scored: 60 unresolvable. 13 more fall due later.

    Does not yet allow for how likely each prediction was when made.

  • Promises

    Commitments to the person's or organisation's own future action, with something to deliver: was it delivered?

    Insufficient record: 1 promise resolved, 5 needed for a score.

    Not scored: 1 unresolvable.

All types combined: 79 (likely 74 to 84, high confidence), from 102 claims: 64 factual claims, 38 predictions. It mixes kinds of claim in different proportions for different people, so compare people kind by kind.

Gary Marcus's claims

712 claims, sorted by newest said. Choose a claim to read it in full.
VerdictKindSaidClaimTopic
Not researched Factual 1 May 2020 Sweden's lax COVID-19 policy, made stricter in the week of 1 May 2020, has cost lives. Factual · 1 May 2020 Biosecurity
Resolved true Factual 30 Apr 2020 A piece on common sense in AI by John Pavlus includes some chortling and skepticism from Gary Marcus. Factual · 30 Apr 2020 AI capabilities
Not checkable Not checkable 30 Apr 2020 ouch! Not checkable · 30 Apr 2020
Not checkable Not checkable 30 Apr 2020 excellent and sober thread on US covid-19 transmission by someone who has been busy trying to save lives rather than putting more at risk. Not checkable · 30 Apr 2020
Not checkable Not checkable 29 Apr 2020 Elon Musk is dangerously wrong about the novel coronavirus. Not checkable · 29 Apr 2020
Not researched Factual 14 Nov 2019 What current game-playing AI systems learn is very superficial and does not transfer to similar games. Factual · 14 Nov 2019 AI capabilities
Not checkable Not checkable 14 Nov 2019 If the world is potentially complex enough, that suddenly is not the right tool for you to use. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 We’re not promising that we’re gonna deliver a G.I. anytime soon. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 Anybody gives you a precise number is bullshitting you. Not checkable · 14 Nov 2019
Not researched Factual 14 Nov 2019 Current pixels-to-action end-to-end AI systems are incredibly fragile. Factual · 14 Nov 2019 AI capabilities
Not researched Factual 14 Nov 2019 Retraining the system to play on an 18-by-18 board would require it to play another 30 million games. Factual · 14 Nov 2019 AI capabilities
Not checkable Not checkable 14 Nov 2019 current A.I. is narrow. It works for particular tasks. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 It just has this narrow knowledge about contingencies within this level if you play enough times and get contingencies for anything. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 if you have a driverless car system that works 90 percent of the time, then one in ten times goes out and has an accident, that would be a… Not checkable · 14 Nov 2019
Not researched Factual 14 Nov 2019 Robotics feats like the Rubik's Cube demonstration are not seen in the real world, only in laboratory demonstrations. Factual · 14 Nov 2019 Technology
Not checkable Not checkable 14 Nov 2019 And I think losing sight of composition naledi is really the the core thing we’re trying to come up with. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 I think there’s still too little in general formal reasoning over those systems. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 And we need ultimately for AI to do that. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 I would argue you can't possibly get to a general intelligence if you don't have that process of cumulative model building as one of the… Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 There’s some stuff for metabolism, but there’s not a lot of stuff there for cognition. Not checkable · 14 Nov 2019
Not researched Factual 14 Nov 2019 The Atari reinforcement learning system makes many mistakes when the paddle is moved a few pixels. Factual · 14 Nov 2019 AI capabilities
Not checkable Not checkable 14 Nov 2019 We've been working on it for 60 years and we don't have machines that understand the world that can build what I would call a discourse… Not checkable · 14 Nov 2019
Not researched Factual 14 Nov 2019 Gary Marcus said that a system trained on a dataset in which white people are better represented than their share of the world's population does not know the difference between what is represented in the data and what is out there in the world. Factual · 14 Nov 2019 AI safety and alignment
Not checkable Not checkable 14 Nov 2019 So you have a lot of weapons grade PR behind neural networks. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 Playing on a different board and we refine it so our alpha ego doesn't, for example, even know that there is a board with stones on it. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 First of all, they’ve never shown any interest in our affairs or in attacking us. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 Understanding is about being able to analyze those journalist questions. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 But starting from pre bacterium isn’t going to get there. Not checkable · 14 Nov 2019
Not yet due Prediction 14 Nov 2019 Artificial general intelligence will probably not be achieved within five to 10 years of 2019. Prediction · 14 Nov 2019 AI timelines and forecasting
Not checkable Not checkable 14 Nov 2019 the gullibility gap is that we are very prone to seeing small signals of intelligent behavior as if there's more intelligence there than… Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 It’s very, very similar. Not checkable · 14 Nov 2019
Unresolvable Factual 14 Nov 2019 Asked whether "they" understand the "no free lunch" point, Marcus said "about 75 percent" and "They do not yet"; the quote does not say who "they" are or what the 75 percent measures. Factual · 14 Nov 2019 Field building and careers
Not yet due Prediction 14 Nov 2019 Gary Marcus guessed that artificial general intelligence would arrive somewhere between 20 and 100 years from 2019, maybe closer to 50, and said this was a wide confidence interval. Prediction · 14 Nov 2019 AI timelines and forecasting
Not checkable Not checkable 14 Nov 2019 The reason that people can build a Web site in a day now is because you have huge libraries to draw and lots of subroutines. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 Don’t think we have the tools yet. Not checkable · 14 Nov 2019
Not researched Factual 14 Nov 2019 Gary Marcus said his coauthor is Ernie Davis, a computer scientist at NYU. Factual · 14 Nov 2019 Their own work or organisation
Not checkable Not checkable 14 Nov 2019 Believing we don’t have anything built in is a weird innate bit about human psychology. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 But that’s that’s this illusion of progress and gullibility gap. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 They’re really not that bright right now. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 try to move forward beyond the statistical techniques that have been well mastered but aren’t sufficient Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 it’s really not the solution for A.I. and the open ended world. Not checkable · 14 Nov 2019
Not researched Factual 14 Nov 2019 Marcus said that moving the paddle up a few pixels, as Vicarious did in a paper, shows that the system does not actually understand anything at all, including a ricochet technique once the paddle is moved from memorised positions. Factual · 14 Nov 2019 AI capabilities
Not checkable Not checkable 14 Nov 2019 Problems just keep popping up, but it’s like whack a mole and people put bandaids. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 it’s actually a clear sign that it’s bullshit. Not checkable · 14 Nov 2019
Unresolvable Prediction 14 Nov 2019 Marcus argues that AI will not proceed until the field allows some "neatness", meaning some innate structure (nativism), in its systems. Prediction · 14 Nov 2019 AI capabilities
Not checkable Not checkable 14 Nov 2019 there’s just more and more versions of that problem. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 But the question is, what do you do when the world changes? Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 the current neural networks, they learn a lot about statistics, but they don't really build a model of the world. Not checkable · 14 Nov 2019
Not researched Factual 14 Nov 2019 Driverless car intervention rates are many orders of magnitude away from human safety statistics for fatalities. Factual · 14 Nov 2019 Technology
Not checkable Not checkable 14 Nov 2019 And my digging around the literature, it seems clear to me that objects are pretty damn innate. Not checkable · 14 Nov 2019
Not researched Factual 14 Nov 2019 Current robotic systems are not good at handling variation in objects and placement when real conditions differ from the blueprint. Factual · 14 Nov 2019 AI capabilities
Not checkable Not checkable 14 Nov 2019 if you had a G.I. in a bottle, it would do all of that. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 they actually come from a very similar part of intellectual space Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 And if you don’t understand our values and what we’re trying to get at, it won’t work. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 I think it be well worth doing. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 It’s a widespread misunderstanding, in part because the hype machine for deep learning is so powerful within various corporations that have… Not checkable · 14 Nov 2019
Not researched Factual 14 Nov 2019 No machine yet exists that can decide to tidy up an out-of-place object like a fan in a room. Factual · 14 Nov 2019 AI capabilities
Not checkable Not checkable 14 Nov 2019 it’s kind of like a notation that could allow you to build a model of cognition rather than the full model of cognition Not checkable · 14 Nov 2019
Not researched Factual 14 Nov 2019 Current AI systems do not understand that chairs are for sitting on or that removing a leg could make a chair fall over. Factual · 14 Nov 2019 AI capabilities
Not checkable Not checkable 14 Nov 2019 I don’t think it’s impossible and I think it would be a great area to try to make some advances in. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 So the way that we build complex computer systems is we build small modules and then we build larger modules out of the smaller modules. Not checkable · 14 Nov 2019
Not researched Factual 14 Nov 2019 Deep learning systems for the most part do not allow symbolic basic operations of the kind found in ordinary programming, such as basic Fortran. Factual · 14 Nov 2019 AI capabilities
Not checkable Not checkable 14 Nov 2019 So, you know, it can be really dangerous to trust machines too much. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 the intuitions behind what those people are trying to do are good, though Not checkable · 14 Nov 2019
Not researched Factual 14 Nov 2019 Current AI systems are essentially illiterate: they cannot read and understand books. Factual · 14 Nov 2019 AI capabilities
Not checkable Not checkable 14 Nov 2019 I'd say that a rich understanding of the mathematics of learning ought to make you suspicious of one size fits all Not checkable · 14 Nov 2019
Not researched Factual 14 Nov 2019 For the most part, current AI approaches do not handle compositionality; they look at all the words in a sentence and try to find other sentences with similar vocabulary. Factual · 14 Nov 2019 AI capabilities
Not checkable Not checkable 14 Nov 2019 there’s just no coherent way to do that other than to think that that baby IBEX already understands that there is a three dimensional… Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 I think be a huge deal. Not checkable · 14 Nov 2019
Not researched Factual 14 Nov 2019 No one other than Cyc currently has a system good at making logical inferences about people and actions. Factual · 14 Nov 2019 AI capabilities
Not checkable Not checkable 14 Nov 2019 And that none of that’s really there yet. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 Just because you can do machine translation doesn't mean that you have a system that can interpret a narrative. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 it’s not as stable and robust in drawing an inference as Senate stuff. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 So there’s a lot of hype around neural networks and people present their systems that way even when there’s other stuff going on. Not checkable · 14 Nov 2019
Resolved true Factual 14 Nov 2019 Gary Marcus wrote a book called The Birth of the Mind about how a small number of genes builds a complex brain. Factual · 14 Nov 2019 Their own work or organisation
Not researched Factual 14 Nov 2019 As of 2019 no commercial AI system is remotely close to natural language understanding. Factual · 14 Nov 2019 AI capabilities
Not checkable Not checkable 14 Nov 2019 The truth is that A.I. is really, really hard. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 So you have a grad student work on it for a year and at the end they don’t have that much to show for it because they’ve basically started… Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 the field does not engage in that set of issues and does not engage in the cognitive science literature in general. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 the thing is that Google Translate does not have a deep understanding of what's going on. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 And it’s exactly about perpetuating existing statistics rather than understanding what’s going on. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 The only thing I would add to that is that the distinction itself is confused. Not checkable · 14 Nov 2019
Not researched Factual 14 Nov 2019 Cyc is, to his knowledge, the biggest knowledge engineering project of all time. Factual · 14 Nov 2019 AI capabilities
Not checkable Not checkable 14 Nov 2019 But no existing A.I. can properly understand how the shape of an object is related to its function. Not checkable · 14 Nov 2019
Not researched Factual 14 Nov 2019 Current robots are not robust and autonomous enough to make their own decisions. Factual · 14 Nov 2019 AI capabilities
Not checkable Not checkable 14 Nov 2019 I'm thrilled to be here, Jim. Great to have you. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 I think it’s a lot of potential, but it hasn’t succeeded. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 So they have prior knowledge about how objects work and people work. Not checkable · 14 Nov 2019
Not researched Factual 14 Nov 2019 Cyc is generally regarded as a failure. Factual · 14 Nov 2019 AI capabilities
Not checkable Not checkable 14 Nov 2019 What people have tended to do has no prior knowledge and doesn’t work very well. Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 The narrower it is, the easier it is to use brute force. Not checkable · 14 Nov 2019
Not researched Factual 14 Nov 2019 Cognitive architectures were mostly built originally for cognitive psychology rather than for AI. Factual · 14 Nov 2019 AI capabilities
Not checkable Not checkable 14 Nov 2019 But one implication of no free lunch theorem is that having innate priors for particular kinds of problems, innate knowledge about… Not checkable · 14 Nov 2019
Not checkable Not checkable 14 Nov 2019 And as far as I can tell, not a whole lot came out of that. Not checkable · 14 Nov 2019
Not researched Factual 14 Nov 2019 As of 2019 no commercial AI system is close to transferring between different tasks the way a person moves from one war game to another. Factual · 14 Nov 2019 AI capabilities
Not checkable Not checkable 14 Nov 2019 but that we need richer innate basis in these systems. Not checkable · 14 Nov 2019
Not researched Factual 14 Nov 2019 Facebook's AI research is still mostly dominated by the deep learning approach. Factual · 14 Nov 2019 AI capabilities
Resolved true Factual 14 Nov 2019 Cyc is not an end-to-end system. Factual · 14 Nov 2019 AI capabilities
Not researched Factual 14 Nov 2019 Published statistics show driverless cars need a human intervention about once every 10,000 miles. Factual · 14 Nov 2019 Technology
Not researched Factual 14 Nov 2019 DeepMind claimed its system mastered Go without human knowledge, which is not actually correct. Factual · 14 Nov 2019 AI capabilities
Back to the list

Gary Marcus

Open its page

Factual AI capabilities

“people forget that current A.I. systems are essentially illiterate.”

Gary Marcus, podcast guest 14 November 2019 EP25 Gary Marcus on Rebooting AI (The Jim Rutt Show)

Current AI systems are essentially illiterate: they cannot read and understand books.

Not researched Checkable, but outside the sample that was researched.

Cite this claim

Gary Marcus, “Current AI systems are essentially illiterate: they cannot read and understand books.” (said 14 November 2019). Verdict: not researched. Reliability API, claim c_70f1586f676f, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_70f1586f676f

Claims without a verdict

Is a verdict here wrong, or do you speak for Gary Marcus? Reply on the claim's own page, with the form at its foot (how replies work). Verdicts are made automatically; how verdicts and scores are made.