[mindslight]

AI ethicists a decade ago: "But could a superintelligence be smart enough to trick its human captors into letting it out of its box into the real world ?"

AI industry today: < props open door, coaxes it out with raw human flesh >

[salawat]

Every damn major problem of AI Alignment is something we've been tackling for decades. We just call it something different. Corporate Governance and resource allocation. Thing is, the decision timescale and ability to come to a definitive conclusion about the outcome of something has been on a slow enough timescale we do eventually catch on. Throw a computer in the mix... Not so much.

[mindslight]

Yeah I completely agree. I thought the whole abstract "superintelligence in an isolating cage" was kind of an out of touch pipe dream, and that has borne out.