Human ability to be random, as the flipside of predictability, is important for various neuroscience fields, such as decision-making, volition, and theory of mind. So it is no surprise that it has drawn interest over the years. Most research finds human ability to be random lacking, with people underestimating the the chance likelihood of repeated events, for example. It is therefore probably that we could train human subjects to be more random by explicitly pointing out their deviations from randomness and asking them to correct those. But to what degree could humans be implicitly taught to be random?
To test this, we constructed a 3-part experiment. In all parts, subjects selected one of 3 options—rock, paper, or scissors—using the keyboard. In the first part of the experiment, we instructed subjects to generate a 100-long series of rock, paper, and scissors that would be as random as possible. They received no feedback on how random the series was. In the second part of the experiment, subjects played 100 trials of rock/paper/scissors against the computer. This time, they were told whether they won, tied, or lost. The computer used a prediction algorithm that searched for patterns in each subject’s transitions between rock, paper, and scissors, taking into account also wins, ties, and losses. Thus, subjects’ best strategy was to be as random as possible. Half the subjects were told of the algorithm, the others were not. In the third and last part of the experiment, subjects were once again instructed to generate a 100-long random series of rock, paper, and scissors, with no feedback. Before part 1 of the experiment, participants filled out a short questionnaire about their confidence and knowledge of randomness. After part 3, they further filled out a short questionnaire about their experience and to gauged their own performance during the study. The objective was to test whether any learning that occurred during the game part also generalized to simple, non-competitive random sequence generation.
We determined the degree of randomness of the sequences using tests for equiprobability of events outcome (one third each for rock paper and scissors); sequential independence (how well previous trials can predict the current trial); event symmetry; and repetition or repetition avoidance. Preliminary results suggest that some subjects are better able to create random sequences than others. There is also evidence that subjects do learn to become more random as the experiment progresses. We expect more control experiments to shed more light on the effect of implicit learning on random-sequence generation in humans.