Superintelligence (ASI)

Superintelligence, or artificial superintelligence (ASI), refers to an intellect that greatly exceeds human performance across most domains of interest. Nick Bostrom popularised the term. It is a hypothetical future capability, not an existing system.

The idea, from I. J. Good, is that a system able to improve its own design could start a feedback loop of rapid capability gains. Whether this is likely, and how fast it would be, is debated. Chalmers and MIRI analyse the argument.

Alignment research aims to make AI systems pursue the goals their designers and users intend. Active topics include reward modelling, scalable oversight, debate, constitutional methods, weak-to-strong generalization and interpretability.

Scalable oversight asks how humans can reliably supervise AI systems on tasks too hard for people to evaluate directly. Proposals include debate, recursive reward modelling and using weaker models to supervise stronger ones.

Researchers disagree. Some, including signatories of the Statement on AI Risk, argue that advanced AI could pose extinction-level risks. Others consider the risk speculative or exaggerated. Structured arguments and surveys help compare positions.

Proposals include international oversight bodies, compute governance, licensing and evaluation of frontier models, and binding safety commitments. Existing instruments include the EU AI Act, the NIST AI RMF and voluntary commitments from the Bletchley and Seoul summits.

Links were checked in November 2026. Some sites block automated checks and may need a manual click. This page is a resource list, not advice or a forecast, and essays reflect their authors' views.