Enlarge / In the study, participants assessed what constitutes an equitable monetary offer. (credit: manusapon kasosod)
Artificial intelligences often learn from materials created or selected by humans, which poses significant challenges in preventing these systems from inheriting human biases and societal prejudices. This issue is especially critical as AIs are increasingly relied upon for making pivotal medical and financial judgments.
A team of researchers at Washington University in St. Louis has uncovered an intriguing layer to this complexity: when individuals who train AIs are aware that their actions might shape AI decision-making in the future, they may alter their own behaviors accordingly. Notably, these behavioral shifts can persist beyond just training contexts.
The Dynamics of Game Theory Engagement
The researchers organized a straightforward game theory experiment involving volunteers, where each pair received a total amount of $10 to distribute between themselves. One person was tasked with proposing how much money to offer to the other participant, who had the option to accept or decline the proposed sum. If the latter chose not to accept the offer, neither participant would receive any funds.
Read 11 remaining paragraphs | Comments
