Scott Carter

Anna Kawakami, Luke Guerdan, Yanghuidi Cheng et al. · cmu

A growing body of research has explored how to support humans in making better use of AI-based decision support, including via training and onboarding. Existing research has focused on decision-making tasks where it is possible to evaluate "appropriate reliance" by comparing each decision against a ground truth label that cleanly maps to both the AI's predictive target and the human decision-maker's goals. However, this assumption does not hold in many real-world settings where AI tools are deployed today (e.g., social work, criminal justice, and healthcare). In this paper, we introduce a process-oriented notion of appropriate reliance called critical use that centers the human's ability to situate AI predictions against knowledge that is uniquely available to them but unavailable to the AI model. To explore how training can support critical use, we conduct a randomized online experiment in a complex social decision-making setting: child maltreatment screening. We find that, by providing participants with accelerated, low-stakes opportunities to practice AI-assisted decision-making in this setting, novices came to exhibit patterns of disagreement with AI that resemble those of experienced workers. A qualitative examination of participants' explanations for their AI-assisted decisions revealed that they drew upon qualitative case narratives, to which the AI model did not have access, to learn when (not) to rely on AI predictions. Our findings open new questions for the study and design of training for real-world AI-assisted decision-making.

Sumanta Bhattacharyya, Francine Chen, Scott Carter et al.

Large Language Models (LLMs) excel academically but struggle with social intelligence tasks, such as creating good compromises. In this paper, we present methods for generating empathically neutral compromises between two opposing viewpoints. We first compared four different prompt engineering methods using Claude 3 Opus and a dataset of 2,400 contrasting views on shared places. A subset of the gen erated compromises was evaluated for acceptability in a 50-participant study. We found that the best method for generating compromises between two views used external empathic similarity between a compromise and each viewpoint as iterative feedback, outperforming stan dard Chain of Thought (CoT) reasoning. The results indicate that the use of empathic neutrality improves the acceptability of compromises. The dataset of generated compromises was then used to train two smaller foundation models via margin-based alignment of human preferences, improving efficiency and removing the need for empathy estimation during inference.

Nikos Arechiga, Francine Chen, Rumen Iliev et al.

Identifying personalized interventions for an individual is an important task. Recent work has shown that interventions that do not consider the demographic background of individual consumers can, in fact, produce the reverse effect, strengthening opposition to electric vehicles. In this work, we focus on methods for personalizing interventions based on an individual's demographics to shift the preferences of consumers to be more positive towards Battery Electric Vehicles (BEVs). One of the constraints in building models to suggest interventions for shifting preferences is that each intervention can influence the effectiveness of later interventions. This, in turn, requires many subjects to evaluate effectiveness of each possible intervention. To address this, we propose to identify personalized factors influencing BEV adoption, such as barriers and motivators. We present a method for predicting these factors and show that the performance is better than always predicting the most frequent factors. We then present a Reinforcement Learning (RL) model that learns the most effective interventions, and compare the number of subjects required for each approach.

Scott Carter, Pernilla Qvarfordt, Matthew Cooper et al.

Tutorials are one of the most fundamental means of conveying knowledge. Ideally when the task involves physical or digital objects, tutorials not only describe each step with text or via audio narration but show it as well using photos or animation. In most cases, online tutorial authors capture media from handheld mobile devices to compose these documents, but increasingly they use wearable devices as well. In this work, we explore the full life-cycle of online tutorial creation and viewing using head-mounted capture and displays. We developed a media-capture tool for Google Glass that requires minimal attention to the capture device and instead allows the author to focus on creating the tutorial's content rather than its capture. The capture tool is coupled with web-based authoring tools for creating annotatable videos and multimedia documents. In a study comparing standalone (camera on tripod) versus wearable capture (Google Glass) as well as two types of multimedia representation for authoring tutorials, we show that tutorial authors have a preference for wearable capture devices, especially when recording activities involving larger objects in non-desktop environments. Authors preferred document-based multimedia tutorials because they are more straightforward to compose and the step-based structure translates more directly to explaining a procedure. In addition, we explored using head-mounted displays for accessing tutorials in comparison to lightweight computing devices such as tablets. Our study included tutorials recorded with the same capture methods as in our access study. We found that although authors preferred head-mounted capture, tutorial consumers preferred video recorded by a camera on tripod that provides a more stable image of the workspace.