According to 20 AI safety experts, what will the first human-level AI look like?
Dec 31
Human-level AI is a stupid concept / your definition is unhelpful
Basically GPT-4/Gemini but bigger and with some tweaks, scaffolding, and more modes
Current AI systems + 1-5 transformer-sized breakthroughs
Current AI systems + some breakthroughs + embodiment
Radically different to what we have now

I have been conducting an informal survey of AI safety experts to elicit their opinions on various topics. I will end up with responses from around 20 people, including researchers at DeepMind, Anthropic, Redwood, FAR AI, and others. The sample was pseudo-randomly selected, optimising for a) diversity of opinion, b) diversity of background, c) seniority, and d) who I could easily track down.

One of my questions was: "What is your modal guess of what the first human-level AI will look like? I define human-level AI as an AI system that can carry out 100% of economically valuable cognitive tasks more cheaply than a human." I asked participants to answer from their inside view as much as possible.

Which theme of answer came up most often?

I will resolve this question when the post for this survey is published, which will happen some time between March and June. Thanks to Rubi Hudson for suggesting turning this into a prediction market.

Get Ṁ600 play money
Sort by:

I wish you didn't force the answers to add up to a 100% xor had them be mutually exclusive.

also 100% of economic tasks more cheaply is pretty hard. 99% is way more feasible.

@Jono3h I know that % is still weird and arbitrary, you could consider asking instead about what the AI will look like after it was attributed to doubling the worlds GDP.

Is "embodiment" ~= robots?


More related questions