Sign In

Communications of the ACM

ACM Opinion

Building Artificial Intelligence: Reward Is Not Enough

View as: Print Mobile App Share:
Rewards Are Not Enough

A recent claim from the DeepMind team, stating that simple rewards are enough for agents in rich environments to develop multi-attribute intelligence of the sort needed to achieve artificial general intelligence, "may sound like a bold claim, but, in fact

Herbert Roitblat is a data scientist and author of Algorithms Are Not Enough.

In an essay for TechTalks, Herbert Roitblat responds to a recent paper by DeepMind scientists, which claims that rewards and reinforcement learning are all you need to build all the components of intelligence.

"Rewards are not enough" to build artificial general intelligence, writes Roitblat. "A general intelligence agent must be able to autonomously formulate its own representations. It has to invent its own approach to solving problems, selecting its own goals, representations, methods, and so on. So far, that is all the purview of human designers [but] we cannot achieve general intelligence until we can remove the dependency on humans to structure problems.

"Reinforcement learning, as a selective process, cannot do it."


No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account