The adversarial Turing test seems like an odd definition to forecast on. Nuno’s linked blogpost makes one side of the argument well: There could be ways to identify an AI as different from a human long after AI becomes economically transformative or capable of taking over the world. On the other side, AI that passes an adversarial Turing test could still fail to have economic impact (perhaps because of regulation, or maybe it’s too expensive to replace human labor) or pose a meaningful existential risk (because it’s not goal directed, misaligned, or capable of overpowering humanity).
I’d be more interested in your forecasts on a few other operationalizations of AI timelines:
Economic impact, as measured by GDP growth rate or AI as % of inputs to GDP, seems like an important aggregate to track and forecast. It has the important quality of being easily verifiable and continuous over time, making forecasts easy to validate with each passing year. On the other hand, economic impact will likely lag cutting edge capabilities, which might pose the most x-risk.
X-Risk is what I actually care about. With all the debate over whether AI x-risk is disjunctive or conjunctive, I wouldn’t want to use a model split into “Will we get AGI, and if so, will x-risks be realized?” that has clear cases where x-risk could occur without first meeting the AGI definition. A tougher question is whether to forecast the exact date of human disempowerment, or a preceding “point of no return”, or another set of ideas. But all of these seem more directly aimed at the most important question of x-risk.
A particularly clean decomposition is “In what year will world energy consumption first exceed 130% of every prior year?” from Matthew Barnett’s Metaculus question. This is designed to forecast transformative AI while accounting for the possibility that AI will overpower humanity, causing GDP to collapse as AI seizes all available resources for its own goal. Forecasting both this question and the economic impact question might show your x-risk estimate in the difference, unless you think that AI could overpower humanity without transformative industrial capacity.
Your thinking on these questions has been pretty persuasive to me, especially Nuno’s recent blog and Eli Lifland’s writeup of thinking through the full case. It’s nice to get a perspective that’s just a bit outside of the constant AI hype bubble. But these forecasts just felt a bit less informative than they could otherwise be, driven by edge cases around the definition. Curious if you would disagree with the importance of those edge cases, or think other forecasting targets have important flaws.
Thank you! We agree and [...], so hopefully, it’s more informative and is not about edge cases of Turing Test passing.
We chose to use an imperfect definition and indicated to forecasters that they should interpret the definition not “as is” but “in spirit” to avoid annoying edge cases.
Fair enough. I think people conceive of AGI too monolithically, and don’t sufficiently distinguish between the risk profiles of different trajectories. The difference between economic impact and x-risk is the most important, but I think it’s also worth forecasting domain-specific capabilities (natural language, robotics, computer vision, etc). Gesturing towards “the concept we all agree exists but can’t define” is totally fair, but I think the concept you’re gesturing towards breaks down in important ways.
Note that GDP is a bad metric here because it doesn’t captura value created (e.g., Wikipedia doesn’t particularly increase GDP in proportion to its value), and e.g., we might also expect GPT-3 and descendants to fall prey to Jevons paradox.
The adversarial Turing test seems like an odd definition to forecast on. Nuno’s linked blogpost makes one side of the argument well: There could be ways to identify an AI as different from a human long after AI becomes economically transformative or capable of taking over the world. On the other side, AI that passes an adversarial Turing test could still fail to have economic impact (perhaps because of regulation, or maybe it’s too expensive to replace human labor) or pose a meaningful existential risk (because it’s not goal directed, misaligned, or capable of overpowering humanity).
I’d be more interested in your forecasts on a few other operationalizations of AI timelines:
Economic impact, as measured by GDP growth rate or AI as % of inputs to GDP, seems like an important aggregate to track and forecast. It has the important quality of being easily verifiable and continuous over time, making forecasts easy to validate with each passing year. On the other hand, economic impact will likely lag cutting edge capabilities, which might pose the most x-risk.
X-Risk is what I actually care about. With all the debate over whether AI x-risk is disjunctive or conjunctive, I wouldn’t want to use a model split into “Will we get AGI, and if so, will x-risks be realized?” that has clear cases where x-risk could occur without first meeting the AGI definition. A tougher question is whether to forecast the exact date of human disempowerment, or a preceding “point of no return”, or another set of ideas. But all of these seem more directly aimed at the most important question of x-risk.
A particularly clean decomposition is “In what year will world energy consumption first exceed 130% of every prior year?” from Matthew Barnett’s Metaculus question. This is designed to forecast transformative AI while accounting for the possibility that AI will overpower humanity, causing GDP to collapse as AI seizes all available resources for its own goal. Forecasting both this question and the economic impact question might show your x-risk estimate in the difference, unless you think that AI could overpower humanity without transformative industrial capacity.
Your thinking on these questions has been pretty persuasive to me, especially Nuno’s recent blog and Eli Lifland’s writeup of thinking through the full case. It’s nice to get a perspective that’s just a bit outside of the constant AI hype bubble. But these forecasts just felt a bit less informative than they could otherwise be, driven by edge cases around the definition. Curious if you would disagree with the importance of those edge cases, or think other forecasting targets have important flaws.
Thank you! We agree and [...], so hopefully, it’s more informative and is not about edge cases of Turing Test passing.
Fair enough. I think people conceive of AGI too monolithically, and don’t sufficiently distinguish between the risk profiles of different trajectories. The difference between economic impact and x-risk is the most important, but I think it’s also worth forecasting domain-specific capabilities (natural language, robotics, computer vision, etc). Gesturing towards “the concept we all agree exists but can’t define” is totally fair, but I think the concept you’re gesturing towards breaks down in important ways.
Note that GDP is a bad metric here because it doesn’t captura value created (e.g., Wikipedia doesn’t particularly increase GDP in proportion to its value), and e.g., we might also expect GPT-3 and descendants to fall prey to Jevons paradox.