Elizabeth Barnes, METR founder and CEO, outlined two possible readings of ambiguous AI capability terminology distinguishing near-term potential from realized task outcomes
Matthew Barnett put rapid capability gains below 5% probability.
@tamaybes @RyanPGreenblatt I realized 2 is a grammatical ambiguity, I didn't intend the parsing of "develop AI capable of [possibly causing z within a few years]", meant "possibly develop [AI capable of causing z] within a few years"
Elizabeth Barnes@BethMayBarnes@tamaybes @RyanPGreenblatt Genuinely unsure what you mean here. Some guesses: 1. ◊X ="AI capable of z" and X = "AI that will in fact succeed at doing z" (or "z will happen")? 2. ◊X = "AI capable of [causing z within a few years]", X="AI capable of z"?
@MatthewJBar @BronsonSchoen @tamaybes @GuiveAssadi It seems very reasonable to me to think capability progress this rapid is pretty unlikely (e.g. <15% or something) but I'd find it hard to be very confident it's not possible (e.g. <1%) - curious how confident you are here?
@MatthewJBar @BronsonSchoen @tamaybes @GuiveAssadi Ah, I'm not sure we actually disagree a ton on substance then - maybe just a communication issue? "quite possibly" is pretty compatible with 5% in my mind. Is this a "brits and americans use 'quite' differently" thing maybe?
Matthew Barnett@MatthewJBar@BethMayBarnes @BronsonSchoen @tamaybes @GuiveAssadi I think it's logically possible but very unlikely (<5%).
@MatthewJBar @BronsonSchoen @tamaybes @GuiveAssadi Or maybe a combination of somewhat different view on risk level, and different sense of what our obligations in public communication are? Ie level of concern about "alarmism" vs "failure to disclose risk"?
Elizabeth Barnes@BethMayBarnes@MatthewJBar @BronsonSchoen @tamaybes @GuiveAssadi Ah, I'm not sure we actually disagree a ton on substance then - maybe just a communication issue? "quite possibly" is pretty compatible with 5% in my mind. Is this a "brits and americans use 'quite' differently" thing maybe?
@MatthewJBar @BronsonSchoen @tamaybes @GuiveAssadi Agree this matters a lot! Although seems to me that it's still useful to know that it's not <1% and not >90%, and get everyone on the same page about that, and take some of the actions that make sense if all you know is you're in that range.
Matthew Barnett@MatthewJBar@BethMayBarnes @BronsonSchoen @tamaybes @GuiveAssadi I think in these contexts it's important to communicate clearly and precisely. It matters a lot whether AI systems are 5% vs. 75% likely to be capable of making humanity go extinct within 3 years.
@BethMayBarnes @BronsonSchoen @tamaybes @GuiveAssadi I think it's logically possible but very unlikely (<5%).
Elizabeth Barnes@BethMayBarnes@MatthewJBar @BronsonSchoen @tamaybes @GuiveAssadi It seems very reasonable to me to think capability progress this rapid is pretty unlikely (e.g. <15% or something) but I'd find it hard to be very confident it's not possible (e.g. <1%) - curious how confident you are here?
@BethMayBarnes @BronsonSchoen @tamaybes @GuiveAssadi I think in these contexts it's important to communicate clearly and precisely. It matters a lot whether AI systems are 5% vs. 75% likely to be capable of making humanity go extinct within 3 years.
Elizabeth Barnes@BethMayBarnes@MatthewJBar @BronsonSchoen @tamaybes @GuiveAssadi I dunno, seems quite possible that "fairly likely" is compatible with 5% :P (I feel a bit torn about how much to use explicit probabilities; obviously desirable to avoid miscommunications, but can feel unrepresentative of actual epistemic state)