This article was translated to French by Anna Chekovsky (cache); Romania by Irina Vasilescu
Many economic theories are built upon the assumption that people are perfectly rational. In general, investors are assumed to prefer higher return and lower risk, as long as they know what the returns and risks are. Suppose a merchant receives two offers to buy one of his commodities: one offers £10 and the other offers £20. With everything else being equal, the merchant, being "rational", will sell it for £20.
Rationality involves knowledge
But suppose the merchant is offered the choice between a payment of £100 today, and a payment of £10 per month over 12 months, what would he do? With basic mathematical and finance training, and the knowledge of his cost of capital, a "rational" merchant should have no problem to choose between the two offers.
Rationality involves computation
The above example shows that knowledge is required in making some "rational" decisions. In fact, "being rational" involves more than basic knowledge. It also requires computation. For example, supposed a merchant has to visit his customers, who are located far apart. Travelling from one customer to another involves a cost, which may vary depending on the time to travel. Some customers may not be available at all times. Suppose the merchant wants to design an itinerary that visits all his 100 customers, with the objective to minimizes travelling costs and satisfying all the customers' availability constraints.
A "rational" merchant would attempt to find the optimal itinerary in this problem. This is a complicated version of problem known as the "travelling salesman problem", which has been studied extensively in operations research and computer science. Clever heuristics have been invented to tackle the travelling salesman problem, but basically they involve heavy computation. Unfortunately, the problem is in nature NP-hard, which means that the time required to find the optimal solution grows exponentially as (in our example) the number of customers increases. Given a fixed amount of planning time, one may not be able to find the optimal itinerary (i.e. the itinerary with minimal travelling cost). In that case, one would have to settle for an itinerary that involves the least travelling cost.
But then what does "being rational" mean? Is one no longer rational if one cannot find the optimal itinerary? Herbert Simon pointed out that most people are only partly rational. He suggested that people are "bounded rational", which means that they can only make the best decisions within their knowledge and resources. Although most economists would accept that perfect rationality is not a realistic assumption, it is not clear how most of the economic theories can be revised to reflect bounded rationality. Concretely quantifying what bounded rationality means remains a grand challenge to the research community. As decision makers have to make decisions about how and when to decide, Ariel Rubinstein proposed to model bounded rationality by explicitly specifying decision making procedures. This puts the study of decision procedures on the research agenda.
A computation point of view
From a computational point of view, decision procedures can be encoded in algorithms and heuristics. In the travelling salesman problem above, some heuristics find better solutions than others. Our knowledge of algorithms, heuristics and our computational power determines how optimal our solutions can be. If rationality is measured by optimality, then our computational knowledge determines how rational we are. Therefore, designing better algorithms and heuristics helps to extend the rationality boundary. Computational Intelligence Determines one's Effective Rationality -- we refer to it as the CIDER Theory.
Where do decision procedures come from?
The General Problem Solver was an attempt in Artificial Intelligence to mimic human intelligence. Decision problems are tackled in constraint satisfaction, which brings together research in Artificial Intelligence, Logic Programming and Operations Research. Instead of designing decision procedures, Evolutionary algorithms attempt to evolve them. Besides, Evolutionary algorithms (including Estimation of distribution algorithms) can be used to model reinforcement learning by bounded rational beings. Therefore, they should have a serious role to play in advancing bounded rationality.
For more details of the CIDER theory, see
Tsang, E.P.K., Computational intelligence determines effective rationality, International Journal on Automation and Control, Vol.5, No.1, January 2008, 63-66 (Early version: Working Paper WP015-07, Centre for Computational Finance and Economic Agents, University of Essex, December 2007) (mirror)
If one takes the CIDER theory further, and study the procedures of bounded rational beings,
one could be in a position to model the micro structure of financial markets and economic systems.
Then one stands a chance to catch a glimpse of the emperor beyond the new clothes of
classical economics, as described in:
Olsen, R., Classical economics: an emperor with no clothes, Wilmott Magazine Volume 15, January 2005, p84-85
Edward Tsang 2007.11.19; revised 2014.12.14