Communications of the ACM
Reward Is Enough for Generalized AI
Artificial agents may be required to maximize a variety of reward signals in future environments, resulting in new forms of intelligence with abilities as distinct as laser-based navigation, communication by email, or robotic manipulation.
In a paper, researchers posit that reward is enough to drive behavior that exhibits abilities studied in natural and artificial intelligence, including knowledge, learning, perception, social intelligence, language, generalization, and imitation. This is in contrast to the view that specialized problem formulations are needed for each ability, based on other signals or objectives.
The paper goes on to suggest that agents that learn through trial and error experience to maximize reward could learn behavior that exhibits most if not all of these abilities, and powerful reinforcement learning agents could constitute a solution to artificial general intelligence.
From the Journal of Artificial Intelligence
View Full Article
No entries found