Amy Summerville

High-consequence domains such as medical triage for critical casualties often involve decisions for which there is no single right answer, and for which expert decision makers will disagree. These decisions thus represent a particular challenge for trustworthy and ethical AI in these domains, as simply training AI to make “the right decision” is not possible. In the current research, we offer a framework for aligning AI systems to the key, domain-specific attributes of decision-makers in order to produce trust in human experts. In Study 1, we investigate six possible attributes and find that the degree of alignment predicts rated trust ratings and delegation using short vignettes for assessment and delegation. In Study 2, we extend these results to demonstrate that alignment predicts trust in decisions from two different AI systems across two distinct attributes. The results thus offer a dimensionality-reducing approach to AI trustworthiness in high-consequence domains.

As AI systems become both more powerful and prevalent, ensuring that their actions align with human values is paramount. The challenge of AI alignment is thus an interdisciplinary one that involves not only a technical challenge for computer science but one with important ties to the psychology of moral values, decision-making, and trust. Early work identified a static set of universal values, without considering the key questions of to whom and to which values AI should be aligned. This perspective paper challenges the notion of universal alignment and instead argues for dynamic, context-specific alignability across different domains, tasks, and users. Specifically, we emphasize the need to go beyond traditional pluralism and rethink how AI alignment can be achieved through a qualitative and quantitative research process that involves identifying context-specific values, developing alignable AI algorithms using limited human feedback, and evaluating alignment through assessing both an AI’s values and actions, while considering how humans trust and delegate to the AI. We discuss several paths forward for our proposed framework, including the potential ethical and societal implications of context-specific alignability, and draw on examples ranging from chatbots to value-aligned decision-making in the medical triage domain.

Aligning AI systems to the attributes of human decision-makers offers a promise of greater AI trustworthiness. The current research offers a proof-of-concept experiment investigating whether alignment to six different high-level attributes of decision-making in critical care medical triage and military field medicine is related to greater trust in more aligned target decision-makers. We find that greater alignment on these values is associated with reports of the target as trustworthy, with ratings that the target makes decisions in the same way as the participant, and with willingness to delegate to the target. Additionally, alignment predicted a forced-choice decision to delegate to one of two target decision makers. This research thus offers evidence that creating AI systems capable of alignment on these characteristics could offer an approach for trustworthy AI in medical triage.

Counterfactuals, thoughts about “what might have been,” play an important role in causal judgment, emotion, and motivation, and spontaneously arise during daily life. However, current methods to measure spontaneous counterfactual thinking are cumbersome and subjective. The current research adapts a paradigm from the Spontaneous Trait Inference literature to develop the Spontaneous Counterfactual Inference measure (SCFI), which uses false recognition of counterfactual statements as a measure of spontaneous counterfactual thought. Studies 1a and 1b demonstrate that the SCFI is sensitive to precursors of counterfactual thinking: norm violation and counterfactual closeness. Study 2 demonstrates that the SCFI converges with the generation of counterfactual statements in an open-ended writing task. The SCFI also predicts two important consequences of counterfactual thought, blame (Study 3), and intention endorsement (Study 4). The SCFI thus offers a new tool for researchers interested in counterfactual thinking.

This research examines how counterfactual potency (CP), the multiplicative effect of the likelihoods of the “if” and “then” clauses of counterfactuals, determines the effects of counterfactuals on behavioral intentions. In Study 1, we found that participants who read highly (vs. minimally) mutable vignettes perceived the counterfactuals as more likely and endorsed relevant intentions more. However, CP did not mediate the effect of mutability on intentions. In Studies 2 and 3, we found that CP directly affected intentions and also mediated the effects of mutability on intentions when mutability was specifically manipulated via controllability (Study 2) or norm violation (Study 3). Finally, Study 4 used archival reaction time data to show that more concrete counterfactuals were perceived as more likely and subsequently facilitated intentions. Taken together, the current research provides evidence that more likely counterfactuals facilitate behavioral intentions.

Counterfactual thinking is a ubiquitous feature of daily life with links to causal reasoning. Therefore, we argue that cultures that vary in perceptions of what controls important life outcomes may also vary in counterfactual thought. Investigating White American and United Arab Emirates-based Arab participants' counterfactual potency and spontaneous counterfactual thinking, we found that Arab participants endorsed counterfactual thoughts less than White Americans, and were unaffected by the routine nature of action when negative outcomes were severe. Differences in counterfactual endorsement in response to severe negative outcomes were linked to greater beliefs in divine control and fate in Arab participants, and not to religiosity, reinforcing an important role of perceptions of control in counterfactual thought. However, although reporting less counterfactual endorsement overall, Arabs showed a similar pattern of counterfactual thought to White Americans when negative outcomes were mild, or when reporting spontaneous thought. Arabs likewise showed a similar pattern of regret as White Americans regardless of event severity, reporting more regret when outcomes resulted from unusual action. These patterns suggest a dissociation between affect and cognition, and between what kind of outcomes are subject to counterfactual scrutiny in Arab participants. (PsycInfo Database Record (c) 2022 APA, all rights reserved).