Superintelligence
Superintelligence is intelligence whose general cognitive abilities substantially exceed human capabilities across many or most cognitive domains.
Sutskever's List
An artificial intelligence whose general cognitive abilities substantially exceed those of humans.
- How it is used
- Frames the problem of steering a potentially more capable agent toward safe goals.
- How deep it goes
- Explains instrumental subgoals through a thought experiment and surveys control approaches without demonstrating a solution.
- Examples
- Uses a math-solving agent that converts Earth into computing hardware and OpenAI’s Superalignment team as contrasting illustrations.
- What you can do after
- Evaluate/understand why capable goal pursuit can create safety risks and why proposed controls remain research questions.
Introduction to Generative AI, Second Edition
Superintelligence is intelligence that surpasses human-level performance across many or all cognitive tasks.
- How it is used
- Frames debate about AI capabilities beyond human level and preparations for their possible disruption.
- How deep it goes
- Defines the term and explains why judging its achievement is difficult, without offering a technical test.
- Examples
- Contrasts AI’s strength on discrete tasks with failures of human-common-sense evaluations.
- What you can do after
- Evaluate/understand claims of superintelligence without treating task-specific success or confident predictions as proof.
Read how Introduction to Generative AI, Second Edition explains superintelligence →
Inside AI
An ultra-intelligent machine is a hypothetical machine whose intellectual abilities greatly exceed those of any human.
- How it is used
- Frames technological singularity as a speculative scenario while separating historical predictions from established AI capability.
- How deep it goes
- Defines the machine and sketches intelligence explosion and control concerns, but does not explain implementation or control methods.
- Examples
- Uses Good’s machine-design scenario and historical predictions about computers matching or exceeding human intellectual work.
- What you can do after
- Evaluate superintelligence claims by distinguishing speculative scenarios from demonstrated capabilities and recognizing control risks.
Related concepts
- intelligence explosionproduces · An intelligence explosion is a hypothesized rapid increase in artificial intelligence caused by systems designing increasingly capable successors.
- artificial general intelligencecontrasts with · Artificial general intelligence is a hypothetical form of machine intelligence capable of learning and performing a broad range of intellectual tasks across domains with human-like generality or greater.
- machine learningkind of · Machine learning is a field of computational methods in which systems learn patterns or relationships from data to make predictions, decisions, or other outputs.
- safetycontrasts with · Safety is the property of a system or process whereby specified harmful, hazardous, invalid, or otherwise forbidden outcomes are avoided under defined conditions.
- superalignmentaddressed by · Research aimed at ensuring that highly capable or superintelligent AI systems remain aligned with human intentions and values.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.