Part 8/17:
The risk of silent failure modes—where AI systems appear safe but secretly are not—is particularly concerning. Such failures could go unnoticed until it's too late, with potentially disastrous consequences, especially if these systems begin to develop longer-term strategic goals.
The Political Landscape: Who Controls the AI?
Control over superintelligent AI is essentially political. Today, companies' CEOs have near-total authority over their AI systems, which could be dangerous if their agendas diverge from societal interests.