Neil W. Rickert (1998) Intelligence is not Rational. Psycoloquy: 9(51) Social Bias (3)

Volume: 9 (next, prev) Issue: 51 (next, prev) Article: 3 (next prev first) Alternate versions: ASCII Summary
PSYCOLOQUY (ISSN 1055-0143) is sponsored by the American Psychological Association (APA).
Psycoloquy 9(51): Intelligence is not Rational

Commentary on Krueger on Social-Bias

Neil W. Rickert
Department of Computer Science
Northern Illinois University
DeKalb, IL 60115


Krueger is concerned with psychological experiments which purport to show that humans are irrational. He argues that the experimental results are partly the outcome of unwise experimental design. While granting Krueger's point, I suggest that the problem is more basic. The very idea of rational behavior, as it is usually conceived, is incompatible with what we understand as human intelligence.


creativity, intelligence, norm, rationality.
1. A series of research reports, such as those described by Kahneman, Slovic & Tversky (1982), have been interpreted by some to suggest that human behavior is not completely rational. For example, Sutherland (1992) considers the research as evidence of problems that should seriously concern us. Others, for example Stich (1990) suggest that part of the problem is with our concept of rationality. Still others suggest that part of the problem is in the research protocols themselves. The target article by Joachim Krueger (1998) falls into this last category.

2. Krueger attributes the results to two aspects of the research procedures. One of these is in the normative theories used as standards of rationality. The other is in the unwise ways in which statistical hypothesis testing has been used. I shall be concerned mainly with the former: the use of normative theories in the research.

3. Krueger's objection to normative theories is that the research has used norms that are too narrow. The problem he sees is a simple statistical one. Human behavior is varied, so we do not normally expect everyone to behave in the same way. By testing behavior against a set of norms, we can identify the proportion of the population which fails to meet those norms. If we make the norms sufficiently narrow, a high proportion of the population will fall outside those standards. Thus the use of overnarrow norms is almost guaranteed to reveal what looks like a problem.

4. While I agree with Krueger's analysis, my concerns are somewhat different from his. I shall be suggesting that there are serious problems inherent in the very question of identifying occasional irrational behavior. Here is a brief outline of the argument I shall present.

    i. Our theories of rationality are normative.

    ii. Creative behavior is behavior which is not norm governed.

    iii. Creative behavior is at the core of what we value as

It follows from this that intelligent behavior is not rational behavior. It need not be irrational. It could be arational behavior, not in accordance with the norms, but not seriously violating them either.

5. The idea that intelligence is not rational may seem absurd. And perhaps it is. But that might only mean that there is a serious problem with our theories of rationality. I shall assume only traditional ideas as to what constitutes rationality.


6. As usually understood, a theory of rationality is a philosophical theory and is usually considered to be part of epistemology. Most philosophers consider epistemology to be a normative discipline. As such, they likewise treat their theories of rationality as normative. That is, a theory of rationality is a theory of how people ought to behave, rather than a theory of how they actually do behave.

7. To say that a discipline is normative, is to say that it has a system of rules or norms by which to judge. A normative theory of behavior is a set of rules for judging behavior. If you did not have rules or specifications, you would not have norms, and the theory could not be normative.

8. Most commonly the rules which establish the norms of rationality are taken to be those of logic. It is assumed that a person, or a rational agent, has a system of beliefs and a set of goals or desires or intentions. In that case, the norms of rationality require that the agent only make decisions which are in accord with its beliefs and its current perceptual inputs. It is allowed that the agent combine rules and beliefs according to the principles of deductive logic. This is about the view of Sutherland (1992).

9. Some philosophers relax the requirements. Cherniak (1986) has done so in his theory of minimal rationality. His idea is that the computational complexity of applying logic to a system of beliefs may be such as to make the problem of perfect rationality an intractable one. In order to allow for this, his theory of minimal rationality permits the use of heuristic solutions which can be expected not to be too far from giving a logically correct solution. In addition, many epistemologists consider induction to be be a rational, if fallible, means of acquiring new beliefs.

10. In what follows, it will not be important whether "rationality" is being used to refer to perfect rationality, or to minimal rationality. What we shall mainly use is the fact that rationality is judged in terms of conformity to some set of rules or specifications.


11. We use the term "creative" to describe a behavior or idea that is unexpected or surprising. However, we would not normally be surprised at a decision or an idea that was arrived at by following a set of rules or specifications. In the circumstances, creative behavior or creative decision making is likely to be outside the range of what would normally be judged as rational behavior. That need not make it irrational.

12. To say that a behavior is irrational is to say that it violates the norms of rational behavior. To say that it is not rational is merely to say that it does not follow the norms. There might be circumstances, for example, in which the norms give insufficient guidance. If I have to choose between a red wine and a white wine, and if I have no beliefs which would distinguish them, then choosing the red wine is neither contrary to the norms of rationality nor does it follow those norms. We might say that it is arational, to indicate that it is outside what is specified by the norms of rationality.

13. What we would expect, then, is that in most cases creative behavior would not be rational, although it might possibly be arational rather than irrational. It might seem to be only a trivial concern if a particular creative act were technically arational, but not irrational. However, the examples of creativity given in the next section will be ones that do violate the usual norms of rationality.

14. There can be cases we would consider creative yet in accordance with the norms. This could happen where a problem is computationally intractable, so there is no practical way of following the norms. Yet a solution might be found by some other method (perhaps even by guessing) which is in accordance with the norms. We might expect this to be possible in cases of high complexity, such as playing chess or solving complex mathematical problems. Some AI programs for playing chess or solving mathematical problems might be considered creative in this sense. However, most creative acts are not of this form, so as a general rule the principle holds that creativity is not rational.


15. Here I shall discuss two cases which have been taken as examples of intelligence. Both were highly creative.

16. For our first case, we examine Einstein's theory of special relativity. The theory grew out of problems in Maxwell's electrodynamics. At the time, the standard belief of physicists had been that light was a wave motion carried by the luminiferous ether. Maxwell, theorizing about electrical and magnetic fields, had been able to derive the differential equation of wave motion. Moreover, the computed velocity of wave propagation was close to the measured velocity of light. The difficulty was that whereas under accepted theory the velocity of light was relative to the ether drift, the theoretically derived velocity of Maxwell's electro-magnetic waves was independent of the ether.

17. There was a perfectly rational solution to this problem. Physicists could simply have insisted that light was not electro-magnetic waves. The norms of logic would have supported them in this belief, on the basis that electromagnetic waves and light waves had different properties. However, Einstein instead gave us special relativity. In order to do so, he had to persuade us to change our concepts of time and space. After those conceptual changes, the ether was no longer relevant to the propagation of light and there was no longer any conflict in assuming that light was electro-magnetic waves. Einstein's creation was outside the norms of rationality, for those norms make no provision for conceptual change. And Einstein's solution failed to follow the perfectly rational solution described above.

18. For a second example, we look to mathematics, and to the Theory of Distributions of L. Schwartz (1957, 1959). Until that time, physicists had been doing mathematics in a way which violated the accepted standards (or norms). For example, they were using the Dirac delta "function." This "function" was supposed to be zero everywhere except at 0, and to be infinite at 0, in such a manner that its integral was 1. The trouble was that there was no such function. What Schwartz gave us was a theory which allowed an extended idea of function, to be called a "distribution," in which it now became perfectly sound mathematics to do what physicists had been doing. As with the case of relativity, the theory introduced significant conceptual change. Once again we see something highly creative, and yet something that was not possible according to the accepted norms. It was outside of what was considered rational, and arguably it was contrary to the accepted norms.


19. It is now almost 50 years since Alan Turing (1950) suggested that computers could be intelligent. In the interim there has been a great deal of research on artificial intelligence. Most of the AI research has been based on rational agency models. Over that time period there have been many advances in what computers can do, and AI researchers have been behind quite a few of those advances. Yet the feeling persists that what computers do is not truly intelligent -- that any intelligence they display is no more than a demonstration of the intelligence of the programmers. We still talk of "dumb computers" and describe them as stupidly following their rules -- rules which are based upon the norms of rationality -- even when it makes no sense to do so. The AI pioneer Terry Winograd has come to recognize that the rational agency model is the wrong one for producing an intelligent system (Winograd & Flores, 1987).


20. I have argued that intelligence is, in principle, contrary to the accepted norms of rationality. In the circumstances it is a mistake to expect that intelligent people will be rational in all of their activities. It is more realistic to expect them to be creative in their attempts to solve problems. And if they are creative, we should not be surprised if occasionally their creative solutions to a problem should turn out to be wrong. Kahneman et al. (1982) have catalogued examples of human behavior which have been considered irrational. They might be better treated as examples of behaviors that were creatively wrong, most commonly in circumstances where the costs for being wrong were quite small.


Cherniak, C. (1986). Minimal Rationality. The MIT Press.

Kahneman, D., Slovic, P., and Tversky, A., editors (1982). Judgment under Uncertainty: Heuristics and Biases. Cambridge University Press.

Krueger, J. (1998). The bet on bias: A forgone conclusion? Psycoloquy 9(46)

Schwartz, L. (1957). Theorie des Distributions, volume I. Hermann, Paris.

Schwartz, L. (1959). Theorie des Distributions, volume II. Hermann, Paris.

Stich, S. P. (1990). The Fragmentation of Reason. The MIT Press.

Sutherland, N. S. (1992). Irrationality: the enemy within. Constable.

Turing, A. (1950). Computing machinery and intelligence. Mind, 59:433-460.

Winograd, T. and Flores, F. (1987). Understanding Computers and Cognition. Addison Wesley.

Volume: 9 (next, prev) Issue: 51 (next, prev) Article: 3 (next prev first) Alternate versions: ASCII Summary