Figure 8.

Pooled cost function for human subjects. We estimate the cost function for each of the six human subjects separately and pool these estimates weighted according to their reliability. As for the monkeys (Fig. 7), we perform this procedure for various settings of the unknown internal reward. For either choice of this parameter, the cost estimate (shown with ±2 SDs) remains indistinguishable from constant until up to 200 ms, the dips slightly below zero for around 150 ms after which it rises almost linearly. This rise is significantly different from zero but much less pronounced than for the monkeys (Fig. 7; note the different scale).