Foresight for ethical AI.

There is growing expectation that artificial intelligence (AI) developers foresee and mitigate harms that might result from their creations; however, this is exceptionally difficult given the prevalence of emergent behaviors that occur when integrating AI into complex sociotechnical systems. We argue that Naturalistic Decision Making (NDM) principles, models, and tools are well-suited to tackling this challenge. Already applied in high-consequence domains, NDM tools such as the premortem, and others, have been shown to uncover a reasonable set of risks of underlying factors that would lead to ethical harms. Such NDM tools have already been used to develop AI that is more trustworthy and resilient, and can help avoid unintended consequences of AI built with noble intentions. We present predictive policing algorithms as a use case, highlighting various factors that led to ethical harms and how NDM tools could help foresee and mitigate such harms.

Neurocognitive assessment on a tablet device: Test-retest reliability and practice effects of ANAM Mobile.

New technology has allowed for the transition of computerized neurocognitive assessments to increasingly user-friendly mobile platforms. While this increased portability facilitates neurocognitive assessment in a wider variety of settings, this transition necessitates further examination of the psychometric properties of tests that have been previously validated on alternate platforms. The present study evaluated the test-retest reliability and practice effects for a new version of the Automated Neuropsychological Assessment Metrics (ANAM), ANAM Mobile, designed to be administered on a tablet computer. A total of 108 undergraduate students completed ANAM Mobile and returned after one week for repeated testing. Observed test-retest reliabilities were consistent with previously established estimates across similar time intervals and ranged from .55 (Simple Response Time) to .87 (Matching to Sample). Modest practice effects were observed. Base rates of reliable decline were low and suggested that declines on two or more tests are uncommon among healthy college students. The present study demonstrates that ANAM Mobile subtests have good-to-excellent reliability across a 7-day retest interval with minimal practice effects. Future research should explore within-subject reliability across repeated ANAM administrations on a tablet device and further examine retest reliabilities over varying time courses and in more diverse samples.