Monday, July 23, 2007

Intuition, reason, and trees

Everyone is used to decision trees, such as this one. These trees are key underlying mechanisms in many areas of decision-making; a significant part of operations research, game theory, symbolic artificial intelligence and management science.

They are deeply associated with the rational actor, and the with supposition of rationality. At each step, the tree branches out some possible paths. In due course, all possible courses of action are (theoretically) enumerated in the tree, such that a rational actor would maximize an utility function U, over all possible decision courses of action, in order to select the best one.

Consider, for example, the game of chess. On an average position of the game, there will be 30 possible moves. This means that the tree will expand thirtyfold (on average) at each ply one analyzes. Standing at a crossroads with 30 alternatives, all theories in management science, operations research, game theory, computer science, would assume the following:

The Choice Assumption. If there are N choices at a branchpoint, they will be compared against each other.

In this way, the best will be chosen. Beautiful to model mathematically, easy to implement computationally, this choice assumption is a non-brainer in the decision sciences. Moreover, it brings real results; theoretical and industrial. Perhaps this explains why it took quite a while for psychologists to start questioning it. Gerd Gigerenzer was, to my knowledge, the first one to do it, by presenting what he calls the "fast and frugal tree": a tree with N nodes, and N+1 leaves. At each node, the decision is binary, akin to questioning oneself "is this solution satisfactory"? A yes will bring it to a leaf, a no will bring it to a further decision node: "how about this one"?

This is where intuition enters the question. Intuition is the information processing between any two nodes.

Note that the topology of the "intuition tree" is quite distinct (See, for instance, figure 3 of this paper).

Where is the psychological evidence that people face a "take-it-or-leave-it" decision at each point, instead of a comparison of multiple choices? Besides Gigerenzer, Gary Klein's recognition-primed decision model also brings up this model (in conversation with Gigerenzer, he confided to me that he couldn't understand what Klein had been up to; 'but hey, maybe, that's what he means'). Moreover, Barry Schwartz discusses many examples in which the
choice assumption does not seem to hold.

So, if people do not compare options at each point, how is a good course of action selected? Gary Klein, Barry Schwartz, and Gigerenzer converge on the same solution: it's not the best alternative; it's just one that seems Ok. It's not about optimizing, it's about satisficing.

So here are some questions: first, what do the states of this intuition tree look like? How can they explain heuristics and biases? What is the process of moving from one state to another like? Finally, what's the test at each node?

For the last one, a refutable, scientific, answer is ready at hand. In FARG terms: Temperature is down to an acceptable level.