Plan forward and backward in time
Whether in solving complex math proof problems or in playing simple maze navigation games, backward planning can be an extremely useful strategy, where one starts their reasoning from the goal state and plans in a direction opposite to later execution. In fact, adopting bi-directional planning at multiple levels can help us arrive at a convergent solution faster. Given that both forward planning and backward planning are useful strategies, how do we use them in constructing our solution? To begin exploring bi-directional thinking, we conducted two experiments using a shortest-path search task in a maze setting. By manipulating the myopic constraints near the starting location and the future constraints near the goal location, we created situations where forward and backward planning will result in differential tendencies in satisfying the myopic or the future constraints. Importantly, satisfying the myopic or the future constraints sometimes led to different choice of path, even as early as the first action taken in approaching the goal. In both experiments, we found that overall, participants showed flexible evaluation of the myopic and future tradeoff. However, there was consistently a myopic, forward bias at the group level. In the second experiment, we introduced a subgoal bottleneck. We found that the bias and flexibility in the path choice patterns with respect to the goal did not fully transfer to the subgoal, with a reduction in the myopic, forward bias at baseline and a reduction in the sensitivity to the myopic and future tradeoff. I will end the talk by discussing some thoughts on exploring forward and backward planning using an abstract task.