Mises Wire

Home | Blog | The Origin of Games

The Origin of Games

October 11, 2005

In light of this year's Nobel for economics, it is worth recalling the reasons that lead Oskar Morgenstern to develop (in conjunction with von Neumann) the theory of games. In 1928 Morgenstern published a book called "Wirtschaftsprognoses," -- "Economic Forecasting." He gave a detailed critique of statistical theory (as it had developed at that time, obviously), and concluded that it was inherently impossible to predict the economic future with quantitative methods. Among the reasons he gave were: (a) actual historical events are distinctly unique in too many of their essential properties to be reduced to sufficiently homogeneous classes of events; (b) actual historical events were not sufficiently independent of each other for much of standard probablility theory; (c) the statistical approach focused on aggregative relationships, while economic phenomena studies relationships among the interdependencies among individuals and their market decisions; (d) human events were not reducable to purely quantitative characteristics, since in the social sciences one is studying human actions that contain meanings and valuations.

On the latter point, years later Morgenstern emphasized: "Yet social phenomena are different: people are acting sometimes against each other, sometimes cooperatively with each other; they have different degrees of information about each other, their aspirations lead them to conflict or cooperation. Inanimate nature shows none of these traits. Atoms, molecules, stars may coagulate, collide, and explode but they do not fight each other; they do not collaborate. Consequently, it was dubious that the methods and concepts developed for the physical sciences would succeed in being applied to social problems." ('Forward' to Morton D. Davis, "Game Theory: a Montechnical Introduction" [New York: Basic Books, 1970], pp. ix-x).

In "Wirtschaftsprognoses" Morgenstern first presented his "thought experiment" of Sherlock Holmes and Professor Moriarity, in which he asked the question, suppose the actors have "perfect foresight" of the future, i.e., that no matter what the actors (in a conflict situation) attempt to plan to foil the actions of the other, each can perfectly anticipate what the other is intending to do, what is implied in terms of their conduct? He concluded that, under conditions of such perfect foresight, the actors would be reduced to robots who can do nothing but what is "preordained," since from the beginning point of the exercise each would know all actions planned by themselves and the other.

Indeed, from the start each would know who, at the end of the day, must surrender to the other. In this situation, there is no longer any meaning to "action" or "decision-making," in the ordinary, everyday sense in which we normally think of it, and act in terms of. Why? Because real actors think and plan with less than perfect foresight of their own future situations, knowledge, and decisions, and that of any others with whom they might in some way interact. Thus there exists "choice," which means (unpredictable) uncertainty.

So how, then, can people anticipate the possible conduct of others, to construct their own plans of action, since what others may do can influence the success of their own intended course(s) of action? Notice how "Austrian" is the nature of the way Morgenstern thought about the problem for which he was searching for a "solution": imperfect knowledge, intentionality, planning, and "action"; the mutual meanings and interpretations of interdependent situtions.

His "answer," as expressed in the game theoretic approach was to say, imagine that we assume that the actors do not have perfect foresight, but in the "matrix" know the number of possible (interdependent) outcomes, the likely pay-offs, the probabilities of selecting one of these outcomes given how the other agent is making his evaluation, etc., etc. Here the agents are given types and degrees of "knowledge sets." And if the analyst -- standing "outside" and "all knowing" about what the actors, respectively know, believe and evaluate -- knows all of this "data," he can then "predict" in this "constrained" situation what the outcomes must be (or in a probabliistic sense may most likely be).

Hence, the analyst is able to devise for himself a "point" or a "range" of determinate outcomes, given the knowledge and related assumptions he has "built into" the actors. Morgenstern's game theoretic approach may be contrasted with that of another "Austrian" who also was interested in thinking about the nature of interdependent human actions: Alfred Schutz and his development of Max Weber's "ideal type" method. It would take too long in this email to restate the entire Schutzian perspective and method. But let me say that Schutz's approach is more open-ended and "non-deterministic."

That is, he is interested in explaining how actors evolve and utilize situation and "action" ideal types of each other in social settings (at a moment in time and through time), and how these typifications provide degrees of social institutional "predictability" concerning what might be expected from others in various circumstances and interactive situations. But how the individual agents will apply this "social knowledge" is open to their free decision-making.

If I may suggest an analogy in contrasting these two approaches. James Buchanan and others have made the distinction between "end-dependent" and "end-independent" rules. End-dependent rules are designed to assure that the actions of the agents tend to and result in a particular outcome or set of outcomes. End-independent rules are meant to provide general guidelines of what type of conduct is or is not permissable in various situations; but what outcomes are likely or will occur is not knowable, because that will depend on the particular goals and plans of the individual agents. This distinction in these type of "rules" has sometimes been called "end-state" vs "process."

Morgenstern's method of studying action under conditions of limited, but defined knowledge, was concerned with explaining the logic of a set of decisions that then leads to a particular outcome, or a narrow set of possible and anticipatable outcomes. I.e., end-state oriented. Schutz's alternative approach is concerned with the analyzing the social and institutional knowledge conditions, in the context of which we may better understand how and why it is possible for individuals to act (with degrees of success) in inherently interdependent situations, regardless of the particular plans and desired outcomes that motivate the actual actions of others. I.e., process-oriented.

Thus, both Morgenstern and Schutz began with the "Austrian" problem-set (imperfect knowledge, action, intentionality, interdependent minds in social and market situations). But their method of attack differed, due to the type of answers they considered important to obtain.

Follow Mises Institute

Add Comment