goal-directed behavior

An RL agent isn't just reacting moment to moment; it's pursuing a goal. Because it's judged on reward accumulated over the long run, it learns to act with the future in mind—taking actions that may not pay off now but set up something better later. This purposeful, forward-looking quality is what makes RL feel like decision-making rather than mere pattern-matching.

Crucially, no one programs the goal as explicit steps; the goal lives entirely in the reward. The agent invents its own strategy for reaching it, sometimes finding moves its designers never imagined. That is the promise and the peril: get the reward right and you get genuinely clever goal-seeking; get it wrong and the agent will pursue your literal reward with unsettling single-mindedness.