| Summary: | Motivation is a powerful force that drives human action and behavior. It drives us to pursue our goals and aspirations and can significantly impact our decision-making processes. In the field of artificial intelligence, the most common method for modeling human action and decision-making is through reinforcement learning, which relies on external reward-based learning mechanisms to influence the agent’s behavior. While rewards are a primary incentive for learning both in the brain and in machines, recent studies have shown that reward signals in the brain influence motivated behavior in a way that is distinct from learning. In this paper, we design a motivated agent that makes decisions based on individual motivation, rather than learning. To do this, we set out to demonstrate that a motivated agent can outperform a learning agent in a sparse reward environment. We also propose a framework for a goal sustaining mechanism based on dopamine firing, and demonstrate how this component immediately impacts the agent’s behavior in a grid environment without relying on learning. In summary, our work aims to contribute to the understanding of motivation and its role in decision-making, both in humans and in artificial intelligence. By designing a motivated agent that can make decisions based on individual motivation, we hope to shed light on how this fundamental aspect of human psychology can be modeled and utilized in artificial intelligence.
|
|---|