• Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    AI
    Reaction

    Lab risk perception and AI 2027's human-response forecasts

    A commentator disputes AI 2027's human-response forecasts, arguing labs accelerate when perceived risk falls and act when it rises.

    2 Sources, 6h ago, first seen 6h ago

    TLDR

    A commentator says they agree with AI 2027's technology forecast but not its forecasts of how people will respond. Their “risk homeostasis” hypothesis holds that lab staff and leaders tend to target a perceived 10–20% sense of doom: they accelerate if risk feels lower and take action if it feels much higher. They argue safety work may leave existential risk roughly unchanged—or increase it—if perceived safety gains outpace actual, scalable gains.

    Combined views

    —

    2 Sources, first seen 6h ago

    — likes— comments— saves— reposts

    Combined views

    —

    2 Sources, first seen 6h ago

    — likes— comments— saves— reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    Featured Source

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    2 Sources

    Jan Kulveit@jankulveitWhen people are "we are on AI 2027 timeline" : so far it seems you are more likely on my timeline. The diff relative to AI 2027 is I disagreed with their human response forecasts while agreeing with their tech tree forecast; subjectively I'm more confident that I'll win the bet with @DKokotajlo than when we made the bet - I expect the divergence of trajectories reality:AI27 to increase over time as the human response becomes larger part of the picture. One example of the human feedback loop in my view modeled wrong in AI27: the subjective felt sense of fear of people in the labs. My impression for many years is we live in 'risk homeostasis' regime, where the target 'perceived sense of doom' by lab folks/leads is about 10-20%. If they feel the risk is smaller, they accelerate. But if they feel it is much higher, they will do something. It was hard to forecast what exactly, as they are smart and agenty, but now we see some signs. ('risk homeostasis' is an older theory but in case of AI safety was AFAIK first stated by @AndrewCritchPhD) Obviously the problem is the 'target' level of labs is very different from the rest of society (but their payoffs are different as well) -> large fraction of AI safety is 'lab alignment problem'. Also: the risk homeostasis hypothesis unfortunately implies legible prosaic AI safety work is often roughly x-risk neutral and easily can be x-risk increasing, if the perceived increase in safety is larger than actual scalable increase in safety.6h
    thebes@voooooogelRT @jankulveit: When people are "we are on AI 2027 timeline" : so far it seems you are more likely on my timeline. The diff relative to AI…1h
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI

    2 Sources

    Jan Kulveit@jankulveitWhen people are "we are on AI 2027 timeline" : so far it seems you are more likely on my timeline. The diff relative to AI 2027 is I disagreed with their human response forecasts while agreeing with their tech tree forecast; subjectively I'm more confident that I'll win the bet with @DKokotajlo than when we made the bet - I expect the divergence of trajectories reality:AI27 to increase over time as the human response becomes larger part of the picture. One example of the human feedback loop in my view modeled wrong in AI27: the subjective felt sense of fear of people in the labs. My impression for many years is we live in 'risk homeostasis' regime, where the target 'perceived sense of doom' by lab folks/leads is about 10-20%. If they feel the risk is smaller, they accelerate. But if they feel it is much higher, they will do something. It was hard to forecast what exactly, as they are smart and agenty, but now we see some signs. ('risk homeostasis' is an older theory but in case of AI safety was AFAIK first stated by @AndrewCritchPhD) Obviously the problem is the 'target' level of labs is very different from the rest of society (but their payoffs are different as well) -> large fraction of AI safety is 'lab alignment problem'. Also: the risk homeostasis hypothesis unfortunately implies legible prosaic AI safety work is often roughly x-risk neutral and easily can be x-risk increasing, if the perceived increase in safety is larger than actual scalable increase in safety.6h
    thebes@voooooogelRT @jankulveit: When people are "we are on AI 2027 timeline" : so far it seems you are more likely on my timeline. The diff relative to AI…1h
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet