The New Science of Alt Intelligence
We recorded 20 claims from this post. 11 were checked against evidence. 9 can’t be checked. 7 more are held. Read the post at its source.
Claims from this post (20)
In the order they appear in the post. Each quotation is the speaker’s words in this post. A label such as “Supported” or “Can’t be checked” is the claim’s current assessment; its page has the evidence and the reasons.
Assessment: Can’t be checkedSaid
“A large fraction of current AI researchers, or at least those currently in power, don’t (so far as I can tell) give a damn about any of this.”
What we tested: A large fraction of current AI researchers, or at least those currently in power, do not, so far as Marcus can tell, care about AI taking inspiration from natural intelligence.
Assessment: UnresolvedSaid
“Right now, the predominant strand of work within Alt Intelligence is the idea of scaling.”
What we tested: Scaling is currently the predominant strand of work in AI that substitutes data for intelligence.
Assessment: UnresolvedSaid
“Systems like DALL-E 2, GPT-3, Flamingo, and Gato appear to be mind-blowing, but nobody who has looked at them carefully would confuse them for human intelligence.”
What we tested: Systems like DALL-E 2, GPT-3, Flamingo and Gato appear mind-blowing, but nobody who has looked at them carefully would confuse them for human intelligence.
Assessment: SupportedSaid
“But it’s also pretty prone to surprising stumbles, like this uncomprehending reply to “a red cube on top of a blue cube”.”
What we tested: DALL-E 2 is fairly prone to uncomprehending errors such as failing to render 'a red cube on top of a blue cube'.
What happened: The authors discuss the exact red-cube/blue-cube prompt in Figure 14 and report recurring object–attribute mix-ups in reconstructions in Figure 15. This supports a recurring compositional weakness in the original model, without establishing a population-wide error rate. (arXiv; OpenAI, )
Assessment: SupportedSaid
“DeepMind’s newest star, just unveiled, Gato, is capable of cross-modal feats never seen before in AI, but still, when you look in the fine print, remains stuck in the same land of unreliability, moments of brilliance coupled with absolute discomprehension:”
What we tested: DeepMind's Gato achieves novel cross-modal feats but remains unreliable, mixing brilliance with total discomprehension.
What happened: The launch paper reports one model spanning dialogue, captions, games and real robot control, including human-level or better performance on 23 Atari games, while acknowledging frequent poor dialogue responses. Its prior-work comparison supports novelty in that combined coverage, rather than novelty of multimodality itself. (arXiv; DeepMind, )
Assessment: Can’t be checkedSaid
“But anyone who is candid will recognize that these kinds of errors reveal that something is, for now, deeply amiss.”
What we tested: Anyone who is candid will recognise that these kinds of errors by current AI systems reveal that something is, for now, deeply amiss.
Assessment: Can’t be checkedSaid
“To some (such as myself), the abiding problem of discomprehension might—despite the immense progress—signal the need for a fundamental rethink, such as the one that Davis and I offered in our book Rebooting AI.”
What we tested: The persistent problem of discomprehension may signal that AI needs a fundamental rethink.
Assessment: SupportedSaid
“The history missed is the the fact that many neural net enthusiasts, have argued against symbols for decades, and the reality that is missed is the above-documented fact that symbolic descriptions like “red cube on blue cube” still elude the state-of-the-art in 2022.)”
What we tested: Many neural net enthusiasts have argued against symbols for decades, and symbolic descriptions like 'red cube on blue cube' still elude the state of the art in 2022.
What happened: This dated scholarly account documents an explicit anti-symbolic strand well before the post. It is historical synthesis rather than a primary statement by an advocate. (Stanford Encyclopedia of Philosophy, )
Assessment: UnresolvedSaid
“So far, in most domains (not all, to be sure) Big Data has triumphed (for now) over careful knowledge engineering.”
What we tested: So far, in most but not all domains, Big Data approaches have beaten careful knowledge engineering.
Assessment: UnresolvedSaid
“Virtually all of the world’s software, from web browsers to spreadsheets to word processors, still depends on knowledge engineering, and Sutton sells that short.”
What we tested: Virtually all of the world's software still depends on knowledge engineering.
Assessment: UnresolvedSaid
“Virtually none of the critical problems for AI that cognitive scientists such as Steve Pinker, Judea Pearl, the late Jerry Fodor, and myself have been pointing to for decades is actually solved yet.”
What we tested: Virtually none of the critical AI problems that cognitive scientists such as Steve Pinker, Judea Pearl, Jerry Fodor and Marcus have been pointing to for decades is actually solved yet.
Assessment: Can’t be checkedSaid
“We are in early days in AI.”
What we tested: AI is still at an early stage of development.
Assessment: SupportedSaid
“Success on a few problems with a particular strategy in no way guarantees that we can solve all problems in a similar way. It is sheer inductive folly to imagine otherwise, particular when the failure modes (unreliability, bizarre errors, failures in compositionality and discomprehension) have not changed since Fodor and Pinker pointed them out (in separate articles) in 1988.”
What we tested: Success on a few problems with a particular strategy in no way guarantees that all problems can be solved the same way, and it is inductive folly to think otherwise, especially as the failure modes (unreliability, bizarre errors, failures of compositionality and discomprehension) have not changed since Fodor and Pinker pointed them out in 1988.
What happened: Tests involving SCAN and RobustFill found continuing difficulty combining learned components in new ways, even with modified Transformers. This supports persistence of compositional generalization problems while explicitly acknowledging progress. (arXiv; Kensen Shi and colleagues, )
Assessment: Can’t be checkedSaid
“Let us all encourage a field that is open-minded enough to work in multiple directions, without prematurely dismissing ideas that happen to be not yet fully developed.”
What we tested: The AI field should pursue multiple research directions rather than dismissing undeveloped ideas prematurely.
Assessment: Can’t be checkedSaid
“It may just be that the best path to artificial (general) intelligence isn’t through Alt Intelligence, after all.”
What we tested: The best path to AGI may not be through data-driven Alt Intelligence approaches.
Assessment: UnresolvedSaid
“As I have written, I am fine with thinking of Gato as an “Alt Intelligence”— an interesting exploration in alternative ways to build intelligence—but we need to take it in context: it doesn’t work like the brain, it doesn’t learn like a child, it doesn’t understand language, it doesn’t align with human values, and it can’t be trusted with mission-critical tasks.”
What we tested: Gato does not work like the brain, learn like a child, understand language, or align with human values, and cannot be trusted with mission-critical tasks.
Assessment: Can’t be checkedSaid
“It may well be better than anything else we currently have, but the fact that it still doesn’t really work, even after all the immense investments that have been made in it, should give us pause.”
What we tested: Gato may well be better than anything else we currently have, but the fact that it still doesn't really work, even after immense investment, should give us pause.
Assessment: Can’t be checkedSaid
“AI should certainly not be a slavish replica of human intelligence (which after all is flawed in its own ways, saddled with lousy memory and cognitive bias). But it should look to human (and animal cognition) for clues.”
What we tested: AI research should draw clues from human and animal cognition without slavishly replicating human intelligence.
Assessment: UnresolvedSaid
“The bottom line is this, something that AI once cherished but has now forgotten: If we are to build AGI, we are going to need to learn something from humans, how they reason and understand the physical world, and how they represent and acquire language and complex concepts.”
What we tested: If we are to build AGI, we will need to learn something from humans: how they reason and understand the physical world, and how they represent and acquire language and complex concepts.
Assessment: Can’t be checkedSaid
“It is sheer hubris to believe otherwise.”
What we tested: It is sheer hubris to believe that AGI can be built without learning from how humans reason, understand the world, and represent and acquire language and concepts.
Held from this post (7)
These records did not pass every check, so nothing about them is published. Held is not a finding.
- The two reviews read the claim differently
- The criteria are ambiguous. Our reading, not assessed: Marcus does not think any pure deep learning/big data system can match Sumit Gulwani's one-trial Flash Fill feature.
- The two reviews read the claim differently
- The two reviews read the claim differently
- The two reviews read the claim differently
- The record does not meet the current rules. Our reading, not assessed: It is false that neural nets have no issue creating and manipulating symbols.
- The two reviews read the claim differently