Pathfinder / funder / opportunities / Risk-Averse AIs
The name 'Risk-Averse AIs' appears in the context of a theoretical argument advocating for training AI systems to exhibit risk aversion in resource allocation, treating resources as having diminishing marginal utility. For instance, such an AI would prefer $40 for sure over a 50% chance of $100 and a 50% chance of $0. The argument posits that risk aversion preserves AI usefulness if the system is aligned, and provides an additional defense layer if misaligned, as misaligned but risk-averse AIs would favor higher probabilities of modest gains over lower probabilities of large, uncertain rewards
No founder records yet.