Understanding Goal Selection Through Human Limitations

From climbing mountains to playing video games, people have a striking ability to make a goal out of just about anything. However, people do not choose their goals haphazardly. Instead, they often rely on previously chosen goals to inform the ones they choose in the future. If people can make a goal of anything, what then governs structure in human goal selection? We frame human goal selection as a balancing act between choosing novel, valuable goals and managing the limited time, cognitive resources, and experience used to identify those kinds of goals in the first place. We then identify three computational problems that people face when selecting goals, which are particularly difficult for novel goals. We hypothesize that structure in human goal selection can arise from a tradeoff between exploiting familiar (but possibly less valuable) goals and expending limited computational resources to come up with, evaluate, and plan for potentially better new goals. Overall, we advocate for an augmented perspective to study goal selection. Instead of viewing goal selection through the broad lens of reward maximization (e.g., characterizing goal selection in terms of the human reward function), or turning to evolutionary timescales as sculptors of our intrinsic reward functions, we view goal selection as special case of decision-making.