SAFE Working Paper No. 336

Artificial Intelligence, Ethics, and Pivotality

With Big Data, decisions made by machine learning algorithms depend on training data

generated by many individuals. In the ethical domain, how does this feature affect the

prosociality of the decisions that serve to train the AI? In an experiment in which we

manipulated the pivotality of individual human decisions used to train an artificially

intelligent algorithm, we show how the diffusion of responsibility weakened revealed

social preferences and led to the design of algorithmic models favoring selfish decisions.

This does not result from a change in the structure of incentives, and it is independent

from the existence of externalities. Rather, our results show that Big Data offers an

excuse for selfish behavior through lower responsibility for one’s and others’ fate.