AI future depends on ontological assumptions about agents, says LessWrong essay
Current models miss unknown unknowns when forecasting AI's societal impact.
Deep Dive
Jonas Hallgren's LessWrong essay argues that predictions about AI's future are shaped by untestable assumptions about what agents are—humans, AI, or hybrids. He critiques linear forecasts that ignore black swan events, using drone warfare in Ukraine as an example. The piece raises questions about how AI could alter democracy and power balances, highlighting limits of current models.
Key Points
- Predictions about AI future depend on untestable assumptions about agents (humans, AI, hybrids).
- Linear forecasts ignore black swan events like Ukraine's drone warfare innovation.
- Current societal metrics assume human-only agency; new AI agents could break democratic and power balance models.
Why It Matters
Forces AI governance thinkers to reconsider how we model future agents and unknown unknowns.