[1] University of Trento and Venezia. The experiments reported here have been conducted at the Laboratory of Computable and Experimental Economics in Trento . The game and the data can be found at location http://www-ceel.economia.unitn.it on INTERNET. Preliminary versions of this paper have been presented and discussed at a number of seminars and conferences. We gratefully acknowledge all the discussants who helped us to clarify numerous hidden problems. Particular thanks are due to Paolo Patelli whose competence was of great help in conduct of the experiments.

[2]In this definition the word computing routine is used in the meaning given to it by the theory of computation, as synonymous with Program or Algorithm. It can be shown that computing routines can be represented by a set of condition-action rules or an equivalent computing device. (Cutland, 1988, Minsky 1967)

[3] Also called (Egidi, 1994) "Transform the Target".

[4]The theory of computation traditionally considers "routines" to be synonymous with "programs" , i.e. a list of instructions in an (artificial) language. Since the two progenitors, Turing machines and Lambda Calculus, a large variety of mathematical representations of "computing machines" have been proposed: URM machines, Post systems, Production Systems, etc. All of these have been proved equivalent to a Turing machine. (Minsky 1967, see for a comment Egidi's Appendix in Cohen M. D. Burkhart R., Dosi G. Egidi M., Marengo L., Warglien M., Winter S. (1955))

[5] The data refer to the behavior of 32 pairs of subjetcs over a 40-run tournament in which only 27 runs were second-level boards, with the Target area occupied at the beginning by either 3 or 4.

[6]120 subjects (i.e. 30 pairs of players in each group), recruited from first year students at the University of Trento, undertook the experiment.

[7] In the last 27 games only 24 were second-level boards, therefore 24 was the highest value of the sum of games played as a 422 and/or a 442.

[8] Two pairs are left out because they often had problems to solve the hands and the data were not comparable.

[9] Heiner's model suggests that if an agent is imperfectly able to select the best strategy to reach a given goal, the more uncertain are the outcomes of his actions, the more limited becomes the repertoire of actions he chooses. In Heiner's view, this means that in conditions of raising uncertainty, agent's behavior becomes more and more routinized. Even though we did not conducted a specific experiment on this aspect of players' behaviors, the findings of our experiment seems quite consonant with Heiner position, insofar there are many situations in which players competence does not allow to decode and use all relevant information, and this seems to induce players to limit their repertoire of actions to the strategy they learned first. rd (2) he needs. Even though this behavior is rather