British Academy: The UK's National Academy for the Humanities and Social Sciences
Enquiry, Evidence and Facts: An Interdisciplinary Conference
Motivated Causal Reasoning: Choice as Evidence
Professor Steven Sloman
Brown University, Cognitive and Linguistic Sciences
Box 1978, Brown University, Providence, RI 02912, U.S.
An abstract presented to the conference
‘Enquiry, Evidence and Facts: An Interdisciplinary conference’
at the British Academy, London, on 13 December 2007
Biography
Steven Sloman is a professor of Cognitive and Linguistic Sciences at Brown University. He has a B.Sc. in Psychology (1986) from the University of Toronto and a Ph.D. in Psychology from Stanford University (1990). He was a postdoctoral fellow at the University of Michigan (1990-1992). Currently he is an editor of the journal Cognition. His work ranges from topics in categorization like how people choose names for objects and how they make inductive inferences across categories to work in causal reasoning and learning to questions about human decision making. He has published in many leading journals of cognition and in a number of books, including his own, published by MIT Press and edited with Lance Rips, Similarity and Symbols in Human Thinking. He has also written Causal Models: Thinking about the World and its Alternatives, 2005, Oxford University Press.
Abstract
The circumference of our waistlines and the amount of grey in our hair are strongly correlated. Unfortunately however, coloring our hair will not reduce our girth. The correlation arises from a common cause – age – so manipulating our hair color doesn’t solve the real problem. Good decision-making generally depends on how actions are causally related to outcomes. Surprisingly, the canonical normative model of decision-making, expected utility theory, is not directly concerned with causal relatedness. Make choices it says in order to maximize the probability of getting the most valued outcomes, but the theory is silent about how to determine those probabilities. The counsel to maximize expected utility would, on the face of it, suggest that all you need to know is the likelihood of events and your own preferences to make rational decisions. But outcome probabilities cannot always be determined by the distribution of previously observed outcomes. The probabilities that are relevant to decision making must reflect the likelihood of outcomes given that the relevant options are actively chosen and not merely observed. Determining these probabilities in general requires a causal model; a probability distribution is not enough.
Moreover, choices do not merely identify one option among a set of possibilities; choosing is an intervention, an action that changes the world. As a result, good decision making generally requires a model specifying how actions are causally related to outcomes. Interventions license different inferences than observations because an event whose state has been determined by intervention is not diagnostic of the normal causes of that event. The claim that choice is an intervention however is complicated by the fact that choices have causes and that, at least some of the time, we learn about those causes by observing the act of choice. We obviously learn about the causes of other people’s choices by observing them and we even learn about ourselves by observing our own choices. I integrate these points into a causal framework for decision making based on causal model theory. The basic idea is that the model we use to make choices is momentarily different than the model we use to understand the world both before and after choice.
The facts about how people choose also provide divergent indicators of whether people treat choice as an intervention. Data on such things as self-deception suggest that people do not; data that I have collected in collaboration with York Hagmayer paint a more rational portrait of human choice. We have shown that people are willing to take actions that affect the probability of a desired outcome only slightly, but refuse to take actions that have no causal impact even if strongly correlated with the outcome. Our studies of Newcomb’s Paradox have found that once the causal structure underlying the decision is clarified, choices are more likely to be consistent with causal relative to evidential expected utility. I will try to reconcile these different data sets by distinguishing what we are and are not aware of when we choose.
To test these ideas, I have run experiments in which people are asked to make hypothetical choices and then to make inferences about their own preferences and other causes of their choices. I have also run experiments in which people either make choices themselves in the context of a two-player game or observe someone else’s choices. I will report preliminary data from these experiments.