A review of Division 5 programming at the 2026 APA convention

Psychological methods is a broad and pluralistic discipline that encompasses the philosophy of science, qualitative inquiry, statistics, measurement, assessment, modeling, and program evaluation. Over the past year, Division 5 initiatives, events, and programs have sought to advance methodology rather than any particular method. Stated differently, our efforts have emphasized the principles, values, and reasoning that guide the use of our scientific tools across diverse contexts. This perspective builds on the vision advanced by past Division 5 Presidents Fred Wertz and Alex Beaujean, who highlighted the importance of methodological reflection, intellectual openness, and respect for diverse approaches as catalysts for innovation and the generation of new knowledge in psychological science and practice. 

Recognizing the transformative nature of Artificial Intelligence (AI) on research and practice, 2026 APA Division 5 Program Chairs Ken Kelley and Zack Fisher organized five timely, informative, and well-attended convention sessions, each offering continuing education credit. These sessions addressed critical questions facing the field today: How is AI affecting data quality? In what ways can AI enhance assessment processes? How can AI support scientific discovery and qualitative inquiry? What might be potential pitfalls? And more broadly, how are psychological methods evolving in response to advances in AI? Together, these discussions highlighted both the opportunities and challenges that AI presents for methodological research and practice.

The five sessions were:

I. AI, Survey Responding, and Data Quality

This session examined how advances in AI are reshaping survey research and raising new questions about data quality. Andrew Gordon (Prolific) reported proprietary research platforms (e.g., Prolific, CloudResearch Connect, and Verasight) generally yield higher quality data than hybrid panels (e.g., Prodege, Dynata, Qualtrics) or aggregator panels (e.g., Clint, PureSpectrum, Primepanels). Although fully autonomous AI agents can generate survey responses, their large-scale use remains limited by cost and technical barriers. More concerning is the rise of AI-assisted human responding, as low-quality human data currently presents a greater threat to research validity than fully AI-generated responses.  Nick Stagnaro (MIT) reached similar conclusions based on his work developing tools to detect AI-generated survey answers. Marsha Krupenkin (University of Maryland) demonstrated that features of open-ended survey responses, such as device type and response mode (voice vs. text), can influence the quality of large language model (LLM) analyses. These differences can affect substantive conclusions and might be amplified by automated text-analysis procedures.  

II. Using generative AI to create or analyze survey or test items

This session examined the growing role of generative AI in assessment. Kevin Williams (ETS) discussed how AI is being integrated into assessment workflows, including content generation, item bank maintenance, automated scoring, and simulation-based assessment and training. Although Alison Cheng (University of Notre Dame) was unable to present in person, Cheng’s materialsopens in new window provided a historical overview of generative AI (GenAI) in assessment, highlighting its roles as an item developer, respondent, judge/rater, and interactive agent.

Both presentations emphasized that GenAI can support assessment development, training, and evaluation, but increased model sophistication alone (e.g., use of LLMs in place of traditional psychometric models) does not guarantee better outcomes. Data quality remains paramount. A central theme was that AI should augment, rather than replace human expertise. Human oversight is essential to ensure validity, transparency, fairness, interpretability, privacy, security, and meaningful variability in assessment. As discussant, Ken Kelley (University of Notre Dame) noted that AI has long played a role in assessment, particularly through computerized adaptive testing, and that recent advances hold considerable promise for further strengthening psychological science and practice.

III. Harnessing AI for Scientific Discovery: Quantitative and Qualitative Perspectives

This session explored how AI is expanding opportunities for scientific discovery across both qualitative and quantitative research. Gabriel Velez (Marquette University) discussed the growing use of GenAI for coding, memoing, summarizing, and interpretation in qualitative research. Drawing on a relational perspective that views GenAI as a third presence between researcher and participant, he highlighted risks such as cognitive offloading, premature interpretive closure, threats to confidentiality, and erosion of community trust. To address these concerns, he advocated for human-led approaches, routine checks for bias and hallucinations, and clear standards for data governance.

From a quantitative perspective, Peter Kvam (The Ohio State University) focused on the scientific potential of deep learning beyond LLMs. He argued that machine learning (ML) methods excel at prediction, classification, and dimension reduction, making them particularly useful for estimating and comparing highly complex models. Grounded in the universal approximation theorem, neural networks can approximate any smooth relation between predictors and outcomes when sufficiently parameterized. As a result, ML methods offer powerful tools for uncovering novel patterns and representations in data, facilitating theory development and scientific discovery.

IV. Psychological Methodology Meets Artificial Intelligence: Modern Innovations and Future Advancements

The keynote session highlighted the growing convergence of psychological methodology and AI. Ahmed Abbasi (University of Notre Dame) described three AI-enabled paradigms relevant to psychological science: (a) using pretrained models to infer psychological constructs from text, (b) prompting GenAI models to adopt specific roles and perspectives, and (c) evaluating AI performance using psychometric methods. He emphasized that realizing the full potential of these approaches will require closer collaboration between psychological methods and computational scientists.

Kai Larsen (University of Colorado – Boulder) demonstrated how AI can translate theory into practice through PsyProxyopens in new window, a platform that analyzes text to identify theoretically grounded constructs linked to outcomes of interest. Drawing on frameworks from psychology and related social and behavioral sciences, the platform can be applied to diverse text sources, including interviews, reviews, patient narratives, and AI-generated conversations. Such tools can help uncover hidden patterns in text, support business and health analytics, and distinguish between human and AI-generated content.

V. Presidential Address: Reflections on Psychological Methodology and AI

Drawing together themes from across the program, Jolynn Pek (The Ohio State University) reflected on the intertwined histories of psychological methodology and AI. She noted that modern AI emerged from advances in statistics, ML, Bayesian computation, and data science, while psychology has long contributed to AI through research on learning, cognition, and human behavior. Learning theory helped shape ML, neural networks were inspired by biological neural systems, and cognitive science – an interdisciplinary field integrating psychology, computer science, linguistics, neuroscience, and philosophy – played a central role in shaping modern AI. 

Pek observed that AI is transforming every stage of the scientific process. Viewed through the distinction between method and methodology, ML approaches have largely been treated as a method for data analysis. GenAI, however, increasingly functions as an active collaborator in research workflows, contributing to construct development, measurement, data generation, coding, interpretation, and evaluation, as reflected throughout the convention program. This shift raises important methodological questions about how human-AI partnerships should be structured and governed in psychological science and practice.

Looking ahead, AI presents new opportunities to advance longstanding methodological debates, including prediction versus explanation, nomothetic versus idiographic approaches, and the relation between observed data and latent psychological constructs. At the same time, AI raises fundamentally psychological questions about agency, identity, cognition, and lived experience in human-AI interactions. For psychological methodologists and scientists, these developments represent not only a technological frontier but also a rich agenda for future inquiry.