Dear Elon Musk, here are five things you might want to consider about AGI
We recorded 21 claims from this post. 16 were checked against evidence. 5 can’t be checked. 4 more are held. Read the post at its source.
Claims from this post (21)
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: SupportedSaid
Observation
“You said, for instance in 2015, that (truly) self-driving cars were two years away; you’ve pretty much said the same thing every year since. It still hasn’t happened.”
Our reading: Musk said in 2015 that truly self-driving cars were two years away and has pretty much said the same thing every year since, and truly self-driving cars still had not arrived.
No horizon Evidence as of
Limitation: Several predictions are documented through contemporary reporting because readable original transcripts were unavailable.
Assessment: SupportedSaid
Observation
“But for the kinds of AI techniques that are currently popular, and which rely more on big data than on reasoning, it’s a very serious issue.”
Our reading: The long-tail problem of rare events is a very serious issue for the currently popular AI techniques, which rely more on big data than on reasoning.
No horizon Evidence as of
Limitation: Evidence is strongest for visual recognition.
Assessment: UnresolvedSaid
Causal claim
“The thing is, it’s easy to convince yourself that AI problems are much easier than they are actually are, because of the long tail problem.”
Our reading: The long tail problem makes it easy to believe AI problems are much easier than they actually are.
No horizon Evidence as of
Limitation: The closest empirical evidence uses a simulated robot task.
Assessment: Can’t be checkedSaid
Forecast
“Unexpected circumstances have been and continue to be the bane of contemporary AI techniques, and probably will be until there is a real revolution.”
Our reading: Unexpected circumstances have been and continue to be the bane of contemporary AI techniques, and probably will remain so until there is a real revolution.
No horizon
Qualifier: until there is a real revolution
Assessment: SupportedSaid
Forecast
“And it’s why I absolutely guarantee you won’t be shipping level 5 self-driving cars this year or next, no matter what you might have told Chris Anderson at TED.”
Our reading: Tesla will not ship level 5 self-driving cars in 2022 or 2023.
Deadline passed Evidence as of
Limitation: The assessment concerns delivered capability, not whether Tesla sold an option named Full Self-Driving.
Assessment: Can’t be checkedSaid
Conditional forecast
“Seven years is a long time, but the field is going to need to invest in other ideas if we are going to get to AGI before the end of the decade. Or else outliers alone might be enough to keep us from getting there.”
Our reading: Unless the field invests in ideas beyond current techniques like deep learning, outliers alone may prevent reaching AGI before the end of the decade.
No horizon
Only if: unless the field invests in ideas beyond deep learning
Assessment: SupportedSaid
Observation
“But you and I both know that they continue to be a major problem that as yet has no known robust solution.”
Our reading: Outliers remain a major AI problem with no known robust solution.
No horizon Evidence as of
Limitation: The absence of a broadly validated solution is an inference from the inspected public research record, not proof that every possible solution was absent.
Assessment: UnresolvedSaid
Observation
“Deep learning is pretty decent at, for example, recognizing objects, but not nearly as good at planning, or reading, or language comprehension.”
Our reading: Deep learning is decent at object recognition but much worse at planning, reading and language comprehension.
No horizon Evidence as of
Limitation: Benchmark success does not establish broad real-world comprehension.
Assessment: UnresolvedSaid
Observation
“We still don’t have anything like stable or trustworthy solutions for common sense, reasoning, language, or analogy. Truth is, I have been using this same pie chart for about 5 years, and the situation has scarcely changed.”
Our reading: AI has no stable or trustworthy solutions for common sense, reasoning, language or analogy, and this has scarcely changed in about five years.
No horizon Evidence as of
Limitation: The studies use different tasks, models and evaluation methods.
Assessment: SupportedSaid
Observation
“Even within perception 3D perception remains a challenge, and scene understanding is by no means solved.”
Our reading: 3D perception and scene understanding remain unsolved for current AI.
No horizon Evidence as of
Limitation: The studies cover particular tasks rather than every form of perception or scene understanding.
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“Four years later, many people continue to hope that deep learning will be a panacea; that still seems pretty unrealistic to me; I still think we need a whole lot more techniques.”
Our reading: Many people continue to hope deep learning will be a panacea, which seems pretty unrealistic to Marcus, and he still thinks a whole lot more techniques are needed.
No horizon
Qualifier: seems pretty unrealistic to me
Assessment: Can’t be checkedSaid
Forecast
“Being realistic, 7 years might well not be enough to (a) invent those tools (if they are don’t already exist) and (b) take them from the lab into production.”
Our reading: Seven years might well not be enough to invent the further techniques needed, if they do not already exist, and to take them from the lab into production.
No horizon
Qualifier: if they don't already exist
Assessment: UnresolvedSaid
Observation
“I don’t know what you have planned for Optimus, but I can promise you that the AGI that you would need for a general-purpose domestic robot (where every home is different, and each poses its own safety risks) is way beyond what you would need for a car that drives on roads that are more or less engineered the same way from one town to the next.”
Our reading: The AGI needed for a general-purpose domestic robot, where every home is different and poses its own safety risks, is way beyond what is needed for a car driving on roads engineered more or less the same from town to town.
No horizon Evidence as of
Limitation: Household robotics evidence includes simulation and constrained kitchen tasks.
Assessment: UnresolvedSaid
Observation
“The fourth thing to realize is that we, and by this I mean humanity as a whole, still don’t really have an adequate methodology for building complex cognitive systems.”
Our reading: Humanity lacks an adequate methodology for building complex cognitive systems.
No horizon Evidence as of
Limitation: The empirical study covers a sample of practitioners and systems.
Assessment: MixedSaid
Causal claim
“Debugging with deep learning is wicked hard, because nobody (a) really understands how it works, and (b) knows how to fix problems other to collect more data and add more layers and so forth.”
Our reading: Debugging deep learning is wicked hard because nobody really understands how it works or knows how to fix problems other than collecting more data, adding more layers and so forth.
No horizon Evidence as of
Limitation: The evidence concerns neural language models, not every deep learning application.
Assessment: Can’t be checkedSaid
Conditional forecast
“Things like programming verification might eventually help, but again, in deep learning we don’t yet have the tools for writing verifiable code. If you are to win the bet, we are probably going to have to solve that problem too, and soon.”
Our reading: Deep learning lacks tools for verifiable code, and AGI by 2029 would probably require solving that soon.
No horizon
Only if: if AGI is to arrive by 2029
Assessment: UnresolvedSaid
Forecast
“In 2029, AI will not be able to reliably construct bug-free code of more than 10,000 lines from natural language specification or by interactions with a non-expert user.”
Our reading: In 2029, AI will not be able to reliably build bug-free code over 10,000 lines from a natural-language spec or from talking with a non-expert user, not counting glued-together library code.
Not yet due Evidence as of
Limitation: The cited demonstrations are developer reports, not independent evaluations satisfying the whole contract.
Assessment: UnresolvedSaid
Forecast
“In 2029, AI will not be able to read a novel and reliably answer questions about plot, character, conflicts, motivations, etc.”
Our reading: In 2029, AI will not be able to read a novel and reliably answer questions about plot, character, conflicts, motivations, etc., where the key is going beyond the literal text.
Not yet due Evidence as of
Limitation: The studies do not establish current frontier performance across the complete task or what systems will achieve in 2029.
Assessment: UnresolvedSaid
Forecast
“In 2029, AI will not be able to take arbitrary proofs from the mathematical literature written in natural language and convert them into a symbolic form suitable for symbolic verification.”
Our reading: In 2029, AI will not be able to take arbitrary proofs written in natural language from the mathematical literature and convert them into a symbolic form suitable for symbolic verification.
Not yet due Evidence as of
Limitation: The research and developer reports were read, but their generated proof artifacts were not independently rerun in this assessment.
Assessment: UnresolvedSaid
Forecast
“In 2029, AI will not be able to work as a competent cook in an arbitrary kitchen (extending Steve Wozniak’s cup of coffee benchmark).”
Our reading: In 2029, AI will not be able to work as a competent cook in an arbitrary kitchen.
Not yet due Evidence as of
Limitation: The cooking demonstrations and unfamiliar-home results concern different systems and tasks.
Assessment: UnresolvedSaid
Forecast
“In 2029, AI will not be able to watch a movie and tell you accurately what is going on (what I called the comprehension challenge in The New Yorker, in 2014).”
Our reading: In 2029, AI will not be able to watch a movie and accurately describe what is going on.
Not yet due Evidence as of
Limitation: The search does not establish that no qualifying system exists.
Held from this post (4)
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: In 2029, no AI system will be able to reliably build bug-free code of more than 10,000 lines from a natural-language specification or through interaction with a non-expert user, not counting code glued together from existing libraries.
- The two reviews read the claim differently
- The criteria are ambiguous. Our reading, not assessed: So far as Marcus knows, Tesla's cars are still mostly doing a kind of glorified hash-table-like lookup (deep learning), and as a result still have a lot of problems with unexpected situations.
- The record does not meet the current rules. Our reading, not assessed: Marcus doesn't think AGI can be reached without solving the methodological problem of building complex cognitive systems, and he doesn't think anybody yet has a good proposal for solving it.