Multi-agent debate has been shown to improve reasoning in large language models (LLMs).However, it is compute-intensive, requiring generation of long transcripts before answering questions.To address this inefficiency, we develop a framework that distills multi-agent debate into a single LLM through a two-stage fine-tuning pipeline combining debate structure learning with internalization via dynamic reward scheduling and length clipping.Across multiple models and benchmarks, our internalized models match or exceed explicit multiagent debate performance using up to 93% fewer tokens.We then investigate the mechanistic basis of this capability through activation steering, finding that internalization creates agent-specific subspaces: interpretable directions in activation space corresponding to different agent perspectives.We further demonstrate a practical application: by instilling malicious agents into the LLM through internalized debate, then applying negative steering to suppress them, we show that distillation makes harmful behaviors easier to localize and control with smaller reductions in general performance compared to steering base models.Our findings offer a new perspective for understanding multi-agent capabilities in distilled models and provide practical guidelines for controlling internalized reasoning behaviors.1

Videos are dynamic and multi-modal compared to other types of content, making automatic filtering difficult, which is why content moderators play a crucial role. However, video content moderators are exposed to more profound emotional labor because videos contain rich visual information, sometimes including even harmful content, such as violent or terrifying scenes. In this work, we explore the effect of six intervention techniques on alleviating negative emotions during video content moderation tasks. We conducted one online crowdsourcing experiment and two controlled user studies to find out that (i) interleaving with positive videos or (ii) cartoonization could significantly reduce negative emotions in the moderators. Participants reported that the advantages of these approaches are in helping reduce negative emotions at the time of moderation while existing approaches focus on post-task activities (e.g., relaxation, talking with others, or getting a hobby). We discuss the applicability of our findings to broader tasks, including improvement in intervention techniques.

Consumers leave digital footprints through large volumes of heterogeneous data which is a wealth of commercial value for firms, waiting to be mined. While there are initial success stories, this area is still under-explored. Further research and communication between the ML community and business community are needed to better align the objectives and create more successful applications. While machine learning is equipped to handle a variety of raw data for predictive tasks, without the theoretical insights from economics and consumer behavior to guide ML models, extracting generalizable insights with clear managerial implications and formulating impactful policies remain elusive. This workshop aims to promote further communication between these disciplines to foster synergistic development of impactful research that could benefit one another.

Machine learning algorithms have gained widespread usage across a variety of domains, both in providing predictions to expert users and recommending decisions to everyday users. However, these AI systems are often black boxes, and end-users are rarely provided with an explanation. The critical need for explanation by AI systems has led to calls for algorithmic transparency, including the "right to explanation'' in the EU General Data Protection Regulation (GDPR). These initiatives presuppose that we know what constitutes a meaningful or good explanation, but there has actually been surprisingly little research on this question in the context of AI systems. In this paper, we (1) develop a generalizable framework grounded in philosophy, psychology, and interpretable machine learning to investigate and define characteristics of good explanation, and (2) conduct a large-scale lab experiment to measure the impact of different factors on people's perceptions of understanding, usage intention, and trust of AI systems. The framework and study together provide a concrete guide for managers on how to present algorithmic prediction rationales to end-users to foster trust and adoption, and elements of explanation and transparency to be considered by AI researchers and engineers in designing, developing, and deploying transparent or explainable algorithms.

We investigate the moderating effect of product attributes and consumer reviews on the efficacy of a collaborative filtering recommender system on an e-commerce site. We run a randomized field experiment on a top North American retailer's website with 184,375 users split into a recommender-treated group and a control group with 37,215 unique products in the dataset. By augmenting the dataset with Amazon Mechanical Turk tagged product attributes and consumer review data from the website, we study their moderating influence on recommenders in generating conversion. We first confirm that the use of recommenders increases the baseline conversion rate by 5.9%. We find that the recommenders act as substitutes for high average review ratings with the effect of using recommenders increasing the conversion rate as much as about 1.4 additional average star ratings. Additionally, we find that the positive impacts on conversion from recommenders are greater for hedonic products compared to utilitarian products while search-experience quality did not have any impact. We also find that the higher the price, the lower the positive impact of recommenders, while having lengthier product descriptions and higher review volumes increased the recommender's effectiveness. More findings are discussed in the Results.