Problem Solving as Probabilistic Inference with Subgoaling: Explaining Human Successes and Pitfalls in the Tower of Hanoi
Fig 8
Comparison of real and simulated data on the ToH.
(a) Success rate mean and variance, “lesioned patients” and “control subjects” (as reported in [67, 77]) and their simulated equivalents. Results are for 25 simulations. (b-c) Matrices of solutions of 25 simulations of lesioned patients (particle percent: perceptual = 85 and look-ahead = 15) and control group (particle percent: perceptual = 15 and look-ahead = 85), respectively. The greyscale indicates the number of votes received by the solutions, i.e., the sum of θ values of Eq 13; the lighter the grey, the higher the number of votes, the better the result. The x-axis shows that the number of solutions found to the problem P3 by the control group (c) is greater than the number of solutions found by the lesioned patients (b). The possibility to find more solutions suggests a greater flexibility of the look-ahead strategy, which in turn might explain its better performance. Furthermore, it is possible to appreciate that the most voted solutions of the control group receive more votes than the most voted solutions of the lesioned patients (compare their greyscale), indicating a faster decision making process.