Dados do Trabalho

Título

The influence of age, gender, and personality on complex skill learning

Introdução

Many cognitive laboratory tasks pose relatively simple challenges to human subjects. In contrast, many human-inspired artificial intelligence (AI) algorithms are able to master much harder problems. What are the differences between human and AI learning in these more complex conditions? We developed a cognitive task of intermediate complexity that is both challenging to subjects and amenable to quantitative analysis. In this task, subjects must discover a sequence of actions that allows them to collect sufficient rewards through trial and error, without the benefit of verbal instructions.Unlike modern reinforcement learning agents known as deep-Q networks (DQNs), we find that humans learn more quickly and discontinuously, displaying ‘leaps of insight’ that generalize to previously unencountered problems. However, there remain substantial sources of variation among human subjects that remain unexplained.

Objetivo

To independent measures such as age, gender, and personality correlate with observed differences in learning across subjects.

Método

1778 participants performed our task and answered a survey to report their age, gender, and a brief measure of the ‘big five’ personality traits (TIPI). We defined these measures as our independent variables. Our dependent variables consisted of measures of behavioral performance in our task: whether or not the subject successfully found a solution to the problem; number of attempts before solving/giving up; average time to complete each puzzle attempt (a proxy for reaction time). The analysis was done by pair correlation.

Resultados

Finding the solution was positively correlated with male gender (r = 0.22; p<0.01) and the personality factor "open to experience" (r=0.11; P<0.05), and was negatively correlated with age (r = -0.3; p<0.01). Age also showed a strong positive correlation with the other behavioral variables. Compared to people aged 20, people aged 60 needed more than twice as many attempts to overcome the level.

Conclusão

Of the measured independent variables, age was most strongly correlated with all aspects of performance and can be associated with both cognitive decline and familiarity with online games. Future studies may clarify this aspect.

Palavras-chave

Skill Learning; Personality; Games.