Superintelligence
AI that vastly exceeds human intelligence — a speculative scenario that dominates discussion in AI safety and long-term risk.
Think of it like
If humans are to ants, superintelligence might be to humans: incomparably more powerful, making our concerns seem quaint.
Example
Kurzweil and others predict superintelligence within decades; critics argue the prediction conflates narrow progress with general reasoning.
How it actually works
Superintelligence encompasses ASI and scenarios just beyond human-level AGI. The concept is baked into AI safety work: if superintelligence is achievable and misaligned, humanity could face an existential threat (the "alignment problem"). Conversely, aligned superintelligence could solve all human problems. This framing is powerful and shapes research priorities, but critics argue it's speculative — we don't know if superintelligence is physically possible, and the probability model is largely guesswork. The term itself is less precise than AGI; it's used to invoke urgency and high stakes.
For product teams
A high-stakes future scenario; AI safety funding and strategy are built around preventing or ensuring alignment.
For engineers
A hypothetical AI system with cognitive capabilities dramatically exceeding human performance across all domains.
Related
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome