Events

Professor Xueming Luo
Charles Gilliland Distinguished Chair Professor of Marketing
Fox School of Business
Temple University

ABSTRACT

Marketing, customer, employee, firm disclosure data have become increasingly multimodal, simultaneously conveying visual, auditory, and textual information. Most prior studies adopt a “signal-by-signal” AI approach, using LLM to extract visual, textual, and audio features separately. Yet consumers integrate facial expressions, gestures, vocal tone, and contextual cues to construct meaning. Ignoring these cross-modal interactions can therefore produce biased estimates and incomplete theoretical understanding. Furthermore, prior research often treats extracted visual, textual, and audio features as exogenous, overlooking endogeneity and unobserved confounders in unstructured data. Because creators strategically choose language, imagery, and vocal expressions, focal treatments may be confounded by latent factors such as design aesthetics, persuasion, authenticity, and audience targeting. Consequently, even LLM-based predictive reasoning AI models cannot reliably identify causal treatment effects from unstructured data. To address these limitations, I will introduce methodological frameworks that integrate advances from multisensory AI, multimodal foundation models, and emerging causal inference techniques for generative AI systems. Specifically, I will discuss state-of-the-art methods for modeling cross-modal interactions through attention mechanisms, multimodal representation learning, and modality fusion architectures, alongside recent developments in causal machine learning, representation-based deconfounding, and generative more credible causal inference from large-scale video, image, audio, and language data. Whether you are a PhD student, faculty member, or industry practitioner, this session will provide actionable insights into the next frontier of AI-powered research and practice for business decision making.