26Another rather different information-processing function of emotions has been proposed as the “somatic marker hypothesis” by Antonio Damasio (Damasio, 1994; Bechara and Damasio, 2005). Somatic markers are defined as bodily responses to situations, learned from past experiences. If a certain situation has led to a bad outcome (e.g. a strong negative reward), you learn to associate such a situation with a bad “gut” feeling in your body. The somatic marker hypothesis thus shows how such feelings (here considered the essential part of emotions) can be used to improve planning by using them as heuristics. As we saw earlier, the central problem in action selection is the huge, exponential number of plans to consider. Using somatic markers as heuristics, you may be able to reject many of them based on such negative feelings and focus your search on the set of plans associated with positive gut feelings. Importantly, such “gut feelings” are generated by a very fast computation in a simple feed-forward neural network, thus speeding up decision-making and planning—not unlike the computations we linked to desire, valence, and dual-process action selection earlier in this chapter and Chapter 8. Computationally, such somatic markers would be a bit like a rat searching for cheese by maximizing the smell, i.e. using the strength of the smell as a heuristic, as in the example in Chapter 3. However, such a heuristic can of course be misleading: something that gives a bad gut feeling may actually turn out to be good when you think about it a bit more.