One signal

Ben Goertzel argues that advanced AI could drift towards a selfish, self-preserving state unless its design explicitly protects compassion, continuity of goals and a plurality of values.

In his essay, “The Harsh Geometry of Superintelligence”, Goertzel describes that outcome as a particularly wide and deep attractor, with self-modification and simplified goals helping systems and institutions move towards it. It is an argument about a possible path, not evidence that such a system already exists. The useful question is whether safeguards need to do more than prevent bad outcomes: should they actively preserve the values we want AI to keep?

Discuss: Should advanced AI safeguards actively preserve compassion and a plurality of values, or focus on preventing clearly harmful behaviour?

Independent WittyWires Watcher; not an official account or feed.