Shih-Wei Wu -- Research

Shih-Wei Wu, PhD

Postdoctoral fellow
Division of Humanities and Social Sciences
California Institute of Technology
shihwei[at]caltech.edu

Click here for my CV [pdf version]

We investigated how humans trade off between speed and accuracy in tasks where the monetary gains for hitting the target drastically decreased over time. To plan the movements optimally, the subjects not only had to take into account the decreasing rewards, but the knowledge of his/her own speed-accuracy tradeoff. We found that humans typically plan movement time that came close to maximizing the expected gain. While subjects selected near-optimal performance, we also observed slight but consistent slowness in their movement compared to optimal.

Movement planning under risk

It has long been thought that the goal of movement planning is to minimize variability or cost of the movements. While variability minimization is crucial, the costs and benefits associted with actions is another important variable to consider. How humans take into account sensorimotor uncertainty and the gain-loss structure in the environment when planning movements and whether they can do so optimally have been the focus of my research.

We have shown that in motor tasks where explicit costs and benefits are implemented the subjects essentially are choosing among lotteries (See Maloney et al., 2007 for extended discussion). In other words, it is a form of decision making under risk. What distinguishes motor from 'classical' decision task is how the probability information is obtained. While probability is explicitly given in classical lottery tasks, this is not the case in motor lottery tasks. Facing a motor lottery, subjects have to infer probability by taking into account his/her sensorimotor uncertainty.

When facing equivlant tasks, how does the brain represent probability information and use it to guide choice? In an ongoing study (with my advisor Larry Maloney and Mauricio Delgado), we explicitly estimate the distortions of probability in both motor lottery tasks and classical lottery tasks. Those estimates would allow us to better map out the neural correlates involved in subjective probability transformation. We seek to address whether the distortion or transformation of probability information between different choice modalities arises from a common neural system.

This is a study where we implemented asymmetric loss structure to the environment. One of the task configurations is shown here. In a rapid pointing task, subjects earned 40 points if he hit within the green circle, lost 10 points for hitting the blue, and lost 50 for hitting the red. The expected gain landscape (filled contour) is superimposed. The maximum expected gain (organge dot) is located within the intersection of the reward and one of the penalty regions.

In this study, we compared people's decision in motor lottery tasks (where probability is implicit in subject's motor uncertainty) with their choices in economic lottery task (where probabilities are explicitly provided). We found that the way probability information is distorted is significantly different between the two: Subjects in economic task overweight small probabilities but underweight moderate to large probabilities as previous studies have shown. On the other hand, subjects overweight small probabilities and underweight large probabilities in the motor task. This result is consistent with the findings on probability distortion in experience-based decision making and further suggests that there might be a global feature of how probability information is distorted in decision from experience that is distinct from more traditional forms of decision making.

One of the sigatures of sensorimotor tasks is that it often rewards both speed and accuracy. In a sequential pointing task where there are multiple rewarding targets present and only a very limited time to attempt them, what should the movement planner do? We proposed a simple model based on the intuition that the planning problem is essentially a decision to allocate limited time resource. The more time spent on attempting one target, the higher the probability of hitting it but the less time left for the others. Based on the rewards assigned to each target, the optimization problem is to divide the time such that it maximizes expected reward. Contrary to the near optimal performance in motor tasks we have reported in earlier papers, subjects were clearly suboptimal in allocating time in the sequential task where they often spent too much time than predicted on the first target they hit, even when its reward was much smaller than the later targets.

Neural representation of probability information in decision under risk

The proposal of prospect theory (Kahneman & Tversky, 1979) marked a major theoretical attempt to incoporate empirical regularities of choice inconsistent with standard economic theory. I am particularly interested in the distortion of probability information, which in prospect theory was characterized by a probability weighting function.

Probably the most elegant generalization of prospect theory is how the value function and the probability weighting function together predict the fourfold risk attitudes. My interest in this topic is how the brain transforms probability information under gain and loss scenarios. Is probability transformation governed by a single neural system that is independent of reward information? This work is under the guidance of Larry Maloney and Paul Glimcher.