Satisficing Games and Decision Making: With Applications to Engineering and Computer Science

Citation

Andrew, A.M. (2005), "Satisficing Games and Decision Making: With Applications to Engineering and Computer Science", Kybernetes, Vol. 34 No. 6, pp. 904-906. https://doi.org/10.1108/03684920510595571

The term “satisficing” is attributed to Herbert Simon and is usually understood to mean much the same as optimisation except that the search is terminated once a “good enough” solution has been found. The effort required to find a better solution may not be justified by the further improvement. In this book, however, it is argued that “satisficing” should not be seen merely as truncated optimisation, but that it offers the key to a fresh and useful approach to many problems of decision‐making. For one thing, it may be possible where the data do not allow optimisation.

In the theory of games and economic behaviour as developed notably by von Neumann and Morgenstern, it is assumed that players act purely in self‐interest and that for each there is a pay‐off to be maximised. With more than two players there are complications that have led to the introduction of principles such as Pareto optimality and Nash equilibrium. A shortcoming of these is that they may indicate a solution (or a set of solutions) that is in some sense optimal, but not how it might have been reached by the interacting players. Also, there are situations that do not lend themselves to such analysis, one such arising when the preferences of one player depend on a choice made by another. A final and crucial shortcoming is that for many situations the results of the analysis do not correspond well with the observed behaviour of actual players.

Reference is made to the “prisoner's dilemma” situation often postulated as a model of international conflict, and to the studies by Axelrod and Rapoport suggesting how altruism might arise in evolution and how situations of the “prisoner's dilemma” kind need not produce deadlock. Altruism is a feature of human nature, and groups of people form a group identity even when all decisions are made individually. One of the advantages of the approach based on satisficing is that it can readily embody multiple objectives, and so can deal with the case where each player responds in accordance with group, as well as individual, preferences.

It is worth noting, although the point is not made in the book, that people do not always act to maximise expected return even when pursuing self‐interest, otherwise neither casinos nor insurers would do business. The question received attention in the 18th century (Pedoe 1963) when it was pointed out, as the “St Petersburg Paradox” that players were not prepared to pay a large sum to enter a game in which the return would be decided by a series of throws of a coin, with a payout of one unit of currency if “heads” appears on the first throw, two units if it appears for the first time on the second throw, four if for the first time on the third throw, and so on with successive doublings. The calculated expected return is infinite so the reluctance of players to invest shows they do not act to maximise expected return. These observations lend support to the “satisficing” viewpoint and its acceptance of multiple objectives. In the St Petersburg case, the decisions by players were probably, and very sensibly, influenced by scepticism about an unbreakable bank for the larger payouts. Attitudes to casinos and insurers are readily accounted for if it is assumed that the attractiveness of a large jackpot payout, and the fear of, say, a disastrous house fire, are treated separately from the goal of immediate financial prosperity.

Returning to the theory as presented, it is obvious that there are many unanswered questions since for instance, the means by which group identity arises is little understood. Consequently the approach based on satisficing requires many rather arbitrary choices by the analyst. It is argued, however, that this is inevitable when a little‐understood process is modelled and that other approaches such as that based on standard games theory are no better in this respect. The need to study group behaviour has been recognised in early work in cybernetics, for instance in a paper by Rapoport (1962).

The approach and viewpoint based on satisficing is cogently defended and will prompt further studies by psychologists and sociologists. Such studies are needed to fill in some of its details, particularly where group identity is involved. My own feeling is that a good case has been made, and that, faced with a meta‐decision about which version of decision theory to employ in a social or economic study, I would be inclined to opt for this one.