The paper "Is it an Agent, or just a Program?" explores the distinction between AI agents and traditional programs over 5,000 citations. Russell and Norvig define agents as entities that perceive and act in an environment. This concept is critical in reinforcement learning (RL), where agents learn to optimize actions toward goals. RL has enabled breakthroughs like AlphaGo, which outperformed human champions in the game of Go. Additionally, large language models utilize RL, particularly RL from human feedback, to better align outputs with human expectations, while LLMs are typically classified as programs despite their agent-like training elements.
"Agents exhibit autonomy when interacting with their environment, using actions to achieve established goals, setting them apart from typical programs."
"Reinforcement learning agents learn to optimize actions to achieve a goal, often discovering strategies better than human approaches and focusing on long-term gains."
Collection
[
|
...
]