The consequences of AI training on human decision-making.

AI is now an integral part of everyday decision-making, assisting us in both routine and high-stakes choices. These AI models often learn from human behavior, assuming this training data is unbiased. However, we report five studies that show that people change their behavior to instill desired routines into AI, indicating this assumption is invalid. To show this behavioral shift, we recruited participants to play the ultimatum game, where they were asked to decide whether to accept proposals of monetary splits made by either other human participants or AI. Some participants were informed their choices would be used to train an AI proposer, while others did not receive this information. Across five experiments, we found that people modified their behavior to train AI to make fair proposals, regardless of whether they could directly benefit from the AI training. After completing this task once, participants were invited to complete this task again but were told their responses would not be used for AI training. People who had previously trained AI persisted with this behavioral shift, indicating that the new behavioral routine had become habitual. This work demonstrates that using human behavior as training data has more consequences than previously thought since it can engender AI to perpetuate human biases and cause people to form habits that deviate from how they would normally act. Therefore, this work underscores a problem for AI algorithms that aim to learn unbiased representations of human preferences.

Optimal Carotid Plaque Features on Computed Tomography Angiography Associated With Ischemic Stroke.

Background Stenosis has historically been the major factor used to determine carotid stroke sources. Recent evidence suggests that specific plaque features detected on imaging may be more highly associated with ischemic stroke than stenosis. We sought to determine computed tomography angiography (CTA) imaging features of carotid plaque that optimally discriminate ipsilateral stroke sources. Methods and Results In this institutional review board-approved retrospective cross-sectional study, 494 ipsilateral carotid CTA-brain magnetic resonance imaging pairs were available for analysis after excluding patients with alternative stroke sources. Carotid CTA and clinical markers were recorded, a multivariable Poisson regression model was fitted, and backward elimination was performed with a 2-sided threshold of P<0.10. Discriminatory value was determined using receiver operating characteristic analysis, area under the curve, and bootstrap validation. The final CTA carotid-source stroke prediction model included intraluminal thrombus (prevalence ratio, 2.8 [P<0.001]; 95% CI, 1.6-4.9), maximum soft plaque thickness (prevalence ratio, 1.2 [P<0.001]; 95% CI, 1.1-1.4), and the rim sign (prevalence ratio, 2.0 [P=0.007]; 95% CI, 1.2-3.3). The final discriminatory value (area under the curve=78.3%) was higher than intraluminal thrombus (56.4%, P<0.001), maximum soft plaque thickness (76.4%, P=0.007), or rim sign alone (69.9%, P=0.001). Furthermore, NASCET (North American Symptomatic Carotid Endarterectomy Trial) stenosis categories (cutoffs of 50% and 70%) had lower stroke discrimination (area under the curve=67.4%, P<0.001). Conclusions Optimal discrimination of ipsilateral carotid sources of stroke requires information on intraluminal thrombus, maximum soft plaque thickness, and the rim sign. These results argue against the sole use of carotid stenosis to determine stroke sources on CTA, and instead suggest these alternative markers may better diagnose vulnerable carotid plaque and guide treatment decisions.