[Kwangsoo's Take] Recursive Self-Improvement and the Case for Uncontrollable AI
The broadcast opened with the hosts' commentary on recursive self-improvement (RSI, the stage at which AI trains AI and evolves on its own). Lee Kwangsoo questioned why this discussion has surfaced now in particular, noting that AGI's risks are already well known and that the timing behind their renewed prominence deserves scrutiny.
Park Sidong cited the failure of nuclear non-proliferation agreements as evidence that verbal accords between nations cannot contain AI risk. He argued that the most realistic way to stop bad AI is to develop better AI, and reiterated his investment hint from last week to watch security-related companies.
The two hosts contrasted nuclear proliferation, which was restrained by the overwhelming deterrence of the US and Soviet Union controlling latecomers, with human embryonic stem cell research, which was self-regulated through globally shared ethical norms. They concluded that AI regulation fits neither model cleanly, leaving the question of an international framework unresolved for now.