Nate Lambert

Four studies measured or manipulated beliefs in free will to illuminate how such beliefs are linked to other aspects of personality. Study 1 showed that stronger belief in free will was correlated with more gratitude, greater life satisfaction, lower levels of perceived life stress, a greater sense of self-efficacy, greater perceived meaning in life, higher commitment in relationships, and more willingness to forgive relationship partners. Study 2 showed that the belief in free will was a stronger predictor of life satisfaction, meaning in life, gratitude, and self-efficacy than either locus of control or implicit person theory. Study 3 showed that experimentally manipulating disbelief in free will caused a reduction in the perceived meaningfulness of life. Study 4 found that inducing a stronger belief in free will caused people to set more meaningful goals for themselves. The possible concern that believers in free will simply claim all manner of positive traits was contradicted by predicted null finding..

Do philosophic views affect job performance? The authors found that possessing a belief in free will predicted better career attitudes and actual job performance. The effect of free will beliefs on job performance indicators were over and above well-established predictors such as conscientiousness, locus of control, and Protestant work ethic. In Study 1, stronger belief in free will corresponded to more positive attitudes about expected career success. In Study 2, job performance was evaluated objectively and independently by a supervisor. Results indicated that employees who espoused free will beliefs were given better work performance evaluations than those who disbelieve in free will, presumably because belief in free will facilitates exerting control over one’s actions.

Foundation models such as GPT-4 are fine-tuned to avoid unsafe or otherwise problematic behavior, so that, for example, they refuse to comply with requests for help with committing crimes or with producing racist text. One approach to fine-tuning, called reinforcement learning from human feedback, learns from humans' expressed preferences over multiple outputs. Another approach is constitutional AI, in which the input from humans is a list of high-level principles. But how do we deal with potentially diverging input from humans? How can we aggregate the input into consistent data about "collective" preferences or otherwise use it to make collective choices about model behavior? In this paper, we argue that the field of social choice is well positioned to address these questions, and we discuss ways forward for this agenda, drawing on discussions in a recent workshop on Social Choice for AI Ethics and Safety held in Berkeley, CA, USA in December 2023.