AGI will not happen in your lifetime. Or will it?
We recorded 28 claims from this post. 11 were checked against evidence. 17 can’t be checked. 16 more are held. Read the post at its source.
Claims from this post (28)
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: UnresolvedSaid
Observation
“Neither one of us believes for a minute that large language models are remotely close to artificial general intelligence, and what’s more, we mostly agree about what’s missing: semantics, reasoning, common sense, theory of mind and so on.”
Our reading: Large language models are not remotely close to AGI; they lack semantics, reasoning, common sense and theory of mind.
No horizon Evidence as of
Limitation: Historical benchmark results are context, not a complete January 2023 assessment.
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“I think we are wasting funding and bright young minds on an approach that probably isn’t on the right path.”
Our reading: Marcus thinks the field is wasting funding and bright young minds on an approach that probably is not on the right path.
No horizon
Qualifier: I think
Assessment: Can’t be checkedSaid
Conditional forecast
“Even if we need to invent a new chip type as a prerequisite, that could be a thirty year process rather a one hundred year process.”
Our reading: Even if a new chip type is required for AGI, that could take about thirty years rather than a hundred.
No horizon
Only if: if a new chip type is needed as a prerequisite
Assessment: Can’t be checkedSaid
Observation
“What I think we are missing are (a) ideas and (b) large scale databases of machine-interpretable knowledge.”
Our reading: What AGI development is missing is new ideas and large-scale databases of machine-interpretable knowledge.
No horizon Evidence as of
Limitation: The narrower factual question of whether large knowledge databases existed is checkable.
Assessment: Can’t be checkedSaid
Observation
“I take that as a hopeful sign that the scaling-über-alas narrative is losing force, and that more and more people are open to new things.”
Our reading: Marcus takes Meta's Cicero as a hopeful sign that the scaling-über-alles narrative is losing force and that more and more people are open to new things.
No horizon
Qualifier: I take that as a hopeful sign
Assessment: UnresolvedSaid
Observation
“The neurosymbolic AI community is growing fast; conferences that used to be dozens are now thousands.”
Our reading: The neurosymbolic AI community is growing quickly, with conferences going from dozens to thousands of attendees.
No horizon Evidence as of
Limitation: The only attendance figure comes from a single virtual IBM event, reported by IBM itself. It may not be comparable with in-person neurosymbolic workshops.
Assessment: SupportedSaid
Observation
“ChatGPT is (at least in many ways) a far inferior intelligence, that, despite having some undeniable talents, cannot at all be trusted.”
Our reading: ChatGPT is in many ways a far inferior intelligence and cannot be trusted.
No horizon Evidence as of
Limitation: The evidence does not measure the 'far inferior intelligence' part. Neither source compares ChatGPT's overall intelligence with human intelligence.
Assessment: Can’t be checkedSaid
Forecast
“Troll farms may well starting using it to create misinformation at unprecedented scale”
Our reading: Troll farms may start using ChatGPT to create misinformation at unprecedented scale.
No horizon
Assessment: SupportedSaid
Forecast
“we can also expect more and circles of fake web sites in order to sell advertisements”
Our reading: ChatGPT will lead to more circles of fake websites created to sell advertising.
Settled before its deadline Evidence as of
Limitation: Most sites are attributed to AI chatbots in general, not to ChatGPT specifically.
Assessment: UnresolvedSaid
Forecast
“In a recent Wired article, I predicted that 2023 see the first death attributable to large language models, and I stand by that prediction.”
Our reading: 2023 will see the first death attributable to large language models.
Deadline passed Evidence as of
Limitation: The strongest evidence is original journalism, not a public death-investigation record.
Assessment: SupportedSaid
Observation
“Already we have seen minor abuses; plagiarism, for example, will never be the same.”
Our reading: Minor abuses of ChatGPT have already been seen, and plagiarism, for example, will never be the same.
No horizon Evidence as of
Limitation: The incident is documented through reporting and the professor's testimony, not a published disciplinary file.
Assessment: Can’t be checkedSaid
Forecast
“Bias is likely to be huge issue, too.”
Our reading: Bias is likely to be a huge issue with large language models like ChatGPT.
No horizon
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“ChatGPT is a dress rehearsal, and I wouldn’t say we as a society have gone through it with flying colors.”
Our reading: ChatGPT is a dress rehearsal for more potent AI, and Marcus would not say society has gone through it with flying colors.
No horizon
Qualifier: I wouldn't say
Assessment: Can’t be checkedSaid
Conditional forecast
“I think that better AI, in which we could directly instill values, might actually make us safer than the world in which we inhabit now, in which so much of what happens is down to the vagaries of what happens to be in this or that training set”
Our reading: Marcus thinks better AI, in which we could directly instill values, might actually make us safer than the current world, in which so much depends on the vagaries of what happens to be in this or that training set.
No horizon
Only if: better AI in which we could directly instill values
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“it terrifies me that so much of current AI is dependent on random details of training corpora that are not available to scientists for inquiry. This seems like a really bad idea.)”
Our reading: Current AI should not depend on random details of training corpora that scientists cannot inspect; this is a bad idea.
No horizon
Assessment: Can’t be checkedSaid
Observation
“I don’t think that kind of flexibility in any way rests on or demands on sentience.”
Our reading: The flexible, general intelligence that defines AGI does not require sentience.
No horizon
Assessment: SupportedSaid
Observation
“These characters (especially perhaps the latter two) might or might not be argued to be sentient, but it seems to be that they are flexible, resourceful, reliably (in a way that LLMs certainly aren’t), and adaptive.”
Our reading: LLMs are certainly not reliable in the way general intelligences like the Star Trek computer would be.
No horizon Evidence as of
Limitation: 'Reliably' is qualitative. The comparison is to fictional characters, so only the plain reading (frequent errors and inconsistency) is testable.
Assessment: Can’t be checkedSaid
Forecast
“I am also with you in doubting that there will be a single magic singularity moment.”
Our reading: There is unlikely to be a single magic singularity moment for AGI.
No horizon
Assessment: UnresolvedSaid
Forecast
“Machines will just get more and more reliable and more general.”
Our reading: AI systems will gradually become more reliable and more general.
No deadline stated Evidence as of
Limitation: Benchmark performance is an incomplete measure of generality.
Assessment: Can’t be checkedSaid
Forecast
“AGI is unlikely to be exactly humanlike, but I don’t see being a replica is a necessity; adaptivity is key, and that is probably achievable by multiple means”
Our reading: AGI is unlikely to be exactly humanlike; adaptivity is key and probably achievable by multiple means.
No horizon
Assessment: Can’t be checkedSaid
Observation
“With all that considerable agreement noted, I am still not seeing a principled argument that constructing flexible, adaptive systems with commonsense reasoning, deeper semantics, and so on will take us past the end of the current century.”
Our reading: Marcus is still not seeing a principled argument that constructing flexible, adaptive systems with commonsense reasoning, deeper semantics and so on will take us past the end of the current century.
No horizon
Qualifier: I am still not seeing
Assessment: Can’t be checkedSaid
Conditional forecast
“Of course nothing is guaranteed; we might not get there because we blow ourselves up, or because we waste so much time on large language models that other important challenges—like semantics, causality, commonsense reasoning, abduction and so forth—never get the attention they deserve.”
Our reading: Nothing is guaranteed: we might not get to AGI because we blow ourselves up, or because we waste so much time on large language models that other important challenges such as semantics, causality, commonsense reasoning and abduction never get the attention they deserve.
No horizon
Only if: we blow ourselves up
Assessment: Can’t be checkedSaid
Conditional forecast
“Or because we can’t come up with something better than the current systems that are so greedy in both data and energy consumption that they could asymptote before anything really general emerges.”
Our reading: We might not get to AGI if we can't come up with something better than current systems, which are so greedy in data and energy consumption that they could asymptote before anything really general emerges.
No horizon
Only if: we can't come up with something better than the current systems
Assessment: UnresolvedSaid
Observation
“A decent fraction of the AI community has already recognized that the primary forte of large language models is in generating bullshit; many also see that the current path to autonomous vehicles is not making the progress that some expected.”
Our reading: A decent fraction of the AI community has already recognized that the primary forte of large language models is generating bullshit, and many also see that the current path to autonomous vehicles is not making the progress that some expected.
No horizon Evidence as of
Limitation: The survey predates the post and covers NLP researchers rather than all AI researchers.
Assessment: Can’t be checkedSaid
Forecast
“But certainly the hunger is there, the funding is there, and my guess is that patience for LLMs (which I see as a detour) is not infinite.”
Our reading: LLMs are a detour, and the field's patience for them will likely run out.
No horizon
Assessment: UnresolvedSaid
Forecast
“So I don’t think we will waste more than 5 years before a substantial fraction of AI effort is looking outside the limited streetlight that currently are sucking up so much mental energy.”
Our reading: Within 5 years, a substantial fraction of AI effort will look beyond the current LLM-focused approaches.
Not yet due Evidence as of
Limitation: Research activity, staffing, and funding are different measures of effort.
Assessment: UnresolvedSaid
Forecast
“I think that in achieving AGI we will have no choice but to find ways of integrating whatever we can get out of deep learning with explicit knowledge and explicit cognitive models.”
Our reading: Achieving AGI will require integrating deep learning with explicit knowledge and explicit cognitive models.
No deadline stated Evidence as of
Limitation: The available architecture disclosures are insufficient for a definitive architectural test.
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“And I also agree that we need many other voices, not just technologists, speaking up in this journey.”
Our reading: AI development should include many voices beyond technologists.
No horizon
Held from this post (16)
These records did not pass every check, so nothing about them is published. Held is not a finding.
- The criteria are ambiguous. Our reading, not assessed: AI physical reasoning will improve before psychological reasoning, and AGI will not arrive on a single identifiable day.
- 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: Humanity will eventually reach AGI, quite possibly before the end of the 21st century.
- 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: Integrating AI components into a whole is a huge unsolved problem.
- The two reviews read the claim differently
- The record does not meet the current rules. Our reading, not assessed: Marcus has long been concerned that a fixation on Big Data has sucked the oxygen from many other ideas that might be better.
- The two reviews read the claim differently
- The criteria are ambiguous. Our reading, not assessed: Most AI money right now is going mostly to large language models, which to Marcus (as to Booch) just seem like an approximation to intelligence rather than the real thing, and so a frustrating distraction.
- The record does not meet the current rules. Our reading, not assessed: ChatGPT has apparently been used to create malware and was used by CNET to produce error-filled news stories.
- The two reviews read the claim differently
- The two reviews reached different assessments. Our reading, not assessed: We do not yet have AI as flexible and reliable as AGI.