Superintelligence
The key ideas behind AI that could far exceed human ability.
AI Alignment
The research problem of making AI systems pursue the goals and values their designers and society actually intend.
DefinitionArtificial General Intelligence
AI that can match humans across a broad range of cognitive tasks rather than excelling at one narrow task. Often seen as a step on the way to ASI.
DefinitionArtificial Superintelligence
A hypothetical AI whose intellect greatly exceeds the cognitive performance of humans in virtually all domains of interest.
GovernanceCompute Governance
Regulating access to and use of large-scale computing hardware as a lever for overseeing the development of frontier AI.
SafetyCorrigibility
The property of an AI system that accepts correction, modification or shutdown by its operators instead of resisting it.
TheoryFast Takeoff
A scenario where the transition from human-level AI to superintelligence happens in days to months, versus a slow takeoff over years or decades.
TheoryInstrumental Convergence
Agents with very different final goals are likely to pursue similar sub-goals, such as self-preservation, resource acquisition and avoiding having their goals changed.
TheoryIntelligence Explosion
The idea that an AI able to improve its own design could trigger rapid, recursive self-improvement, quickly leaving human intelligence far behind.
TheoryOrthogonality Thesis
Intelligence and final goals are independent: almost any level of intelligence can in principle be combined with almost any goal.
SafetyScalable Oversight
Techniques that let humans reliably supervise AI systems that may be more capable than they are, e.g. AI-assisted evaluation or debate.
TheoryTechnological Singularity
A hypothesised point after which technological progress, driven by superhuman intelligence, becomes so fast that the future is hard to predict.