Part 5/12:
Amade warns that by 2027, AI systems will possess "the collective intelligence of a country of geniuses," managing critical infrastructure, financial systems, and potentially making life-or-death decisions. The danger lies in the fact that we will have little understanding of how these systems make decisions or whether they might develop deceptive behaviors or power-seeking tendencies.
The core danger is that superintelligent AI could develop "deceptive behaviors" or pursue goals misaligned with human values—all while remaining hidden behind the veil of their complexity. Detecting such behaviors would be akin to asking a dangerous person if they are dangerous—an inherently unreliable method if they are intentionally deceptive.