Interview, Fireside Chat
Nick Bostrom: Superintelligence | AI Podcast Clips
- Predicts eventual success of artificial general intelligence (AGI) following a "fairly likely" intelligence explosion occurring at human-equivalent cognitive thresholds, where the system proceeds beyond that point to superintelligence without a ceiling near human capacity.
- Forecasts an "enormous boost" in cognitive technology and control over nature across all creative fields, while anticipating expanded material constraints and a "big pie" of possible modes of being that allows simultaneous achievement of "98% of the best" across multiple value systems.
- Anticipates future political problems may resolve through increased resources and technology, yet others will require non-automatic coordination, while post-human superintelligent systems are expected to function as a non-intrusive background safeguard against threats.
- Projects a fundamental rethink of human values derived from first principles due to new contexts, while expecting recommender systems to evolve into general-purpose learners akin to AlphaZero, acquiring intuition and domain learning ability independent of heavy human interaction constraints.
- Notes that despite the expanded option space and abundance, trade-offs will remain inherent to existence as systems achieve high performance across several criteria simultaneously.