While recovering from my successful hip operation last month, I read Daniel Kahneman's bestselling new book, Thinking, Fast and Slow. The book is an engaging, informative review of a lifetime of influential research. In addition to reporting his epochal work with Amos Tversky on cognitive psychology and behavioral economics, Kahneman describes a project to determine when intuitions are reliable. Because of his many experiments that showed that people's intuitive judgments about statistical matters are frequently in conflict with logical norms, Kahneman was skeptical about claims that we should trust the intuitions of experts such as firefighters.
The question of the reliability of intuitions has philosophical relevance, because many philosophers think that they can use intuitions generated by thought experiments to justify conclusions about the fundamental nature of knowledge, reality, and morality. I'll soon be attending a workshop on thought experiments at the University of Toronto where most of the speakers are enthusiastic about the use of thought experiments to obtain knowledge; but my presentation, consistent with what I argued in The Brain and the Meaning of Life, has the title: Thought Experiments Considered Harmful.
Kahneman describes an "adversarial collaboration" with Gary Klein, who has done fascinating work on expertise described in his book Sources of Power. Their key question was: When can you trust a self-confident professional who claims to have an intuition? They concluded that their disagreement about the reliability of intuitions was partly due to their experience with different experts. Whereas Klein had studied firefighters, nurses, and other professionals with real expertise, Kahneman had studied clinical psychologists, stock pickers, and political scientists whose forecasts were unreliable. Klein and Kahneman came to agree that the confidence that people have in their intuitions is not a good guide to their validity, because it is easy to achieve high confidence by ignoring failures.
Rather, acquiring a skill for making good intuitive judgments requires both an environment that is sufficiently regular to be predictable and an opportunity to learn these regularities through prolonged practice. Kahneman advocates this rule: intuition cannot be trusted in the absence of stable regularities in the environment. There are, however, domains where real expertise is possible, for example in chess where knowledge of regularities is available and acquired by large amounts of practice, so that expert intuitions can sometimes be taken seriously.
The implication of these findings for philosophical intuitions is clear: don't trust them. The common philosophical method of making up a story to pump people's intuitions is not based on stable regularities or prolonged practice, just verbal fluency and cultural influence. Thought experiments in philosophy strike me as no more reliable than faith-based methods such as reading sacred texts. In the philosophy of mind, there is a long history of thought experiments that I think yield the wrong answers, for example Searle's Chinese room thought experiment that supposedly shows that computers cannot represent the world. This argument is refuted by the existence of robot cars that are now successfully navigating complex environments (see my paper on Robosemantics with Chris Parisien).
My skeptical conclusion does not mean that thought experiments are useless. Kahneman and Tversky began all their investigations with thought experiments that they used to probe their own intuitions about statistical matters, but they followed good scientific practice in following them up with rigorous real experiments that showed that many other people had the same erroneous intuitions. Great scientists like Galileo and Einstein undoubtedly used thought experiments to advance their research, but acceptance of their ideas depended on the accumulation of experimental results arising from careful observations of interactions with the world. Philosophers should follow scientists in restricting their use of thought experiments to the generation and clarification of hypotheses, and to the demonstration of inconsistencies in rival hypotheses. Real experiments produce evidence that is repeatable and robust in that it can be obtained by different people in different ways through interaction with the world. In contrast, the intuitions produced by thought experiments are not evidence at all.