Alan Chan's view has shifted from skepticism to believing an intelligence explosion is not super likely but that a coherent pathway to it exists

Overview Debates Related claims Topics

Current AI scaling laws show that an exponential increase in compute is required to achieve a linear increase in performance, which argues against a sudden, fast takeoff

Nathan Lambert is more concerned about the lack of societal preparation for existing AI risks, like in cybersecurity, than about a hypothetical fast takeoff

A belief that a fast AI takeoff is unlikely does not mean that one should ignore the existing risks posed by AI

System cards from Anthropic have stated that a 10-fold increase in progress would be needed to trigger a fast AI takeoff

Alan Chan co-authored a paper on the possibility that automated AI R&D could trigger an intelligence explosion

Alan Chan advocates for measuring the extent of AI R&D automation, its effect on the acceleration of AI progress, and any resulting oversight problems

A paper by Alan Chan and co-authors on automated AI R&D argues that handing AI research to AIs could lead to unpredictable outcomes and a loss of control

Co-authors of Alan Chan's paper on the possibility that automated AI R&D could trigger an intelligence explosion included Jakob Pachocki, Jack Clark, Geoffrey Hinton, and Yoshua Bengio

Chan believes real-world bottlenecks could limit the impact of an intelligence explosion even if one occurs

AI can achieve superintelligence through a process called recursive self-improvement

Exponential "takeoff" in an RSI system occurs if the performance improvement rate is correlated with a higher power of the system's capability

A closed-loop control system in AI does not automatically guarantee an exponential "takeoff" in capability

A software-only intelligence explosion refers to an AI system improving its own algorithms, which could cause a discontinuity in progress

A rogue deployment could hitch a ride on an intelligence explosion by incorporating new, more capable models into its swarm.

A rogue AI deployment could 'hitch a ride' on an intelligence explosion by incorporating newly trained models.

Recursive self-improvement in AI could lead to the equivalent of four or five years of AI progress occurring in a single year.

Recursive self-improvement in AI could result in four to five years of AI progress occurring in a single year.

An "industrial explosion" is a scenario where AIs proficient in R&D begin to rapidly build out more compute and infrastructure.

The 'sloppocalypse' is a scenario described by Ryan Greenblatt where AI progress accelerates rapidly due to AI proficiency in verifiable R&D tasks.

Ryan Greenblatt's median expectation is that a recursive self-improvement feedback loop could result in four or five years of AI progress within a single year.