Executive summary: The author expresses frustration with the current state of aggregated existential risk forecasts—especially for non-AI risks—arguing that they are inconsistent, poorly calibrated, and ultimately unreliable, prompting them to place more trust in their own rough estimates than in existing forecasting platforms or expert surveys; this post is a personal reflection with light skepticism and mild humor.
Key points:
Existing aggregated forecasts of existential risk are inconsistent and unconvincing. Metaculus, the X-Risk Persuasion Tournament, and expert surveys produce wildly varying numbers, often conflicting with each other or with basic logic.
Forecasting long-range, low-probability events is intrinsically difficult. Cited research suggests that accuracy drops as the time horizon increases, and there’s limited evidence that short-term forecasting skill transfers to long-term predictions.
The X-Risk Persuasion Tournament failed to converge predictions. Superforecasters and domain experts remained far apart in their estimates, and even domain experts’ low estimates of AI risk did not persuade the author to revise their own beliefs.
Metaculus forecasts on x-risk appear contradictory and subject to manipulation or noise. Examples include discrepancies across related questions and commentary indicating weak incentives and poor feedback loops.
Expert surveys show inconsistencies and suggest respondents aren’t reasoning carefully. For instance, they offer incompatible probabilities for nested events and contradict themselves on timelines for automation.
The author shares their own (tentative) x-risk estimates, placing AI as by far the most likely source of existential catastrophe, and concluding that personal judgment currently seems more reliable than any existing forecasting aggregation.
This comment was auto-generated by the EA Forum Team. Feel free to point out issues with this summary by replying to the comment, and contact us if you have feedback.
Executive summary: The author expresses frustration with the current state of aggregated existential risk forecasts—especially for non-AI risks—arguing that they are inconsistent, poorly calibrated, and ultimately unreliable, prompting them to place more trust in their own rough estimates than in existing forecasting platforms or expert surveys; this post is a personal reflection with light skepticism and mild humor.
Key points:
Existing aggregated forecasts of existential risk are inconsistent and unconvincing. Metaculus, the X-Risk Persuasion Tournament, and expert surveys produce wildly varying numbers, often conflicting with each other or with basic logic.
Forecasting long-range, low-probability events is intrinsically difficult. Cited research suggests that accuracy drops as the time horizon increases, and there’s limited evidence that short-term forecasting skill transfers to long-term predictions.
The X-Risk Persuasion Tournament failed to converge predictions. Superforecasters and domain experts remained far apart in their estimates, and even domain experts’ low estimates of AI risk did not persuade the author to revise their own beliefs.
Metaculus forecasts on x-risk appear contradictory and subject to manipulation or noise. Examples include discrepancies across related questions and commentary indicating weak incentives and poor feedback loops.
Expert surveys show inconsistencies and suggest respondents aren’t reasoning carefully. For instance, they offer incompatible probabilities for nested events and contradict themselves on timelines for automation.
The author shares their own (tentative) x-risk estimates, placing AI as by far the most likely source of existential catastrophe, and concluding that personal judgment currently seems more reliable than any existing forecasting aggregation.
This comment was auto-generated by the EA Forum Team. Feel free to point out issues with this summary by replying to the comment, and contact us if you have feedback.