Defining a human-like problem solving strategy for a robot

We developed a system the purpose of which is obtaining a robot able to emulate the strategies used by a human facing a problem-solving task. We have been able to solve this problem within a very particular psychological setting, in which the human behaviour can be interpreted as ‘observables’ of his/her problem-solving strategy. Our solution encompasses the one of yet another problem, namely, how to close up a loop starting with the behaviour of several humans, its analysis and interpretation in terms of human observables, the definition of what was are the strategies used by the humans (including inefficient ones), the interpretation of the human observables in terms of movements of the robot, the definition of what is a “robot strategy" in terms of human strategies. The loop is closed with a programming language enabling us to program these robot strategies, making them observables in the same way as the human strategies are observables at the beginning of the loop. This paper is devoted to the detailed explanation of one of the above steps, that is, how we have been able to define in an objective way what we call a robot strategy. We shall see that our solution merges two different factors. The one aims at avoiding very ‘inhuman’ behaviours and is based on the mean of the behaviour of the set of humans we observed. The other provides ‘humanity’ to the robot by allowing it to deviate from this mean by n times the standard deviation observed, paralleling to deviations from the mean of the human it is supposed to emulate. Completely new human-like behaviours are also easy to program.