Jaan Tallinn "Toy Models of Superintelligence Control Problem"

Klipi teostus: Merlin Saulep 17.05.2016 937 vaatamist Arvutiteadus



Recently many AI experts have warned that the creation of human-level AI could go horribly wrong and spell the end of humanity. This tends to cause a lot of confusion as people (including AI developers!) scramble to imagine a human level AI and shoehorn concepts from human psychology (such as "volition" and "evil intent") to AI. The issue becomes much clearer when, instead of modelling AI as a very smart human, we look at simple toy models of multi-agent systems that also exhibit many of the problems that said AI experts have warned us about