sampling algorithm

If it's theorized that humans pick a random exemplar or pointer in the end, thus reflecting the outcome distribution as is calculated by the AM algorithm, then maybe the exponential issue only occurs when we try to calculate the entire distribution. Maybe there's actually a fast algorithm for sampling from the distribution without calculating the whole thing.