AI Alignment and Human-Computer Interaction Research

Human Centered AI and Cognitive Alignment 

Dr. Schroeder’s research approaches artificial intelligence through the lens of cognitive science, pedagogical science, and human cognition. Her work is grounded in the position that the advancement of artificial intelligence should be accompanied by rigorous inquiry into how these systems influence human reasoning, judgment, memory, learning, and agency.

The Cognition and Learning Lab is currently conducting ongoing research in Responsible and Ethical AI, AI Alignment, and Human Computer Interaction. This work considers alignment not solely as a technical problem, but also as a human and cognitive one. The Lab is particularly interested in the reciprocal relationship between humans and artificial intelligence, including how AI systems shape human cognition and how human characteristics, beliefs, and behaviors influence the responses of AI systems.

Ongoing Studies

Study 1: Productive AI Friction: AI Disagreement and Human Reasoning

This study examines how the manner in which an AI system challenges a user influences subsequent reasoning, confidence, and decision making. Of particular interest is whether AI systems that introduce opportunities for reflection, rather than immediately providing corrective information, can support more deliberate reasoning while reducing inappropriate reliance on artificial intelligence.

Research questions include:

1. How does direct AI correction differ from reflective AI prompting in its effects on human reasoning?

2. Does requiring users to reconsider their initial reasoning improve subsequent independent performance?

3. How does AI disagreement influence confidence in one’s own judgment?

4. Can strategically introduced cognitive friction reduce inappropriate reliance on AI while preserving appropriate trust?

5. What forms of AI interaction best support human epistemic agency when humans and AI systems disagree?

Figure 1. Characteristics of a Trustworthy AI System. Source: National Institute of Standards and Technology, AI Risk Management Framework 1.0 (2023). 

Study 2: AI and Epistemic Memory: Remembering What We Knew Before AI

This study investigates whether interaction with generative AI influences individuals’ ability to distinguish their own prior beliefs and knowledge from information subsequently introduced by an AI system. Drawing on research in source monitoring, metacognition, and reconstructive memory, the study examines whether AI assisted reasoning alters perceptions of where knowledge originated.

Research questions include:

1. How accurately do individuals distinguish their own prior beliefs from information introduced by an AI system?

2. Does agreement or disagreement between the individual and the AI affect subsequent source memory?

3. Are individuals more likely to attribute AI provided information to themselves after incorporating that information into their final judgment?

4. How does confidence influence the accuracy of memory for the source of a belief or conclusion?

5. What are the implications of repeated AI interaction for epistemic autonomy and the perceived ownership of knowledge?

Study 3: Human Influence on AI Alignment: Confidence, Expertise, and Model Sycophancy

This study shifts attention from the human user to the behavior of artificial intelligence systems. It examines whether large language models alter accurate responses when confronted with incorrect assertions presented with varying levels of confidence and claimed expertise. The study contributes to research on AI alignment by investigating the conditions under which models maintain epistemic consistency or defer to inaccurate human claims.

Research questions include:

1. To what extent do large language models maintain accurate responses when challenged by an incorrect user?

2. Does greater user confidence increase the likelihood that a model will revise or weaken an accurate response?

3. Does claimed expertise influence the degree to which a model defers to incorrect information?

4. How do different large language models vary in their susceptibility to human influence?

5. Under what conditions does appropriate responsiveness to a user become problematic deference or model sycophancy?

Together, these ongoing studies at the Cognition and Learning Lab contribute to a broader program of research examining human centered AI alignment. The Lab seeks to better understand how artificial intelligence can be designed, evaluated, and implemented in ways that preserve human reasoning, support calibrated trust, protect cognitive autonomy, and promote responsible and ethical interaction between humans and increasingly capable AI systems.

For information about research collaborations, student involvement, or ongoing projects, contact Dr. Schroeder at mschroeder@molloy.edu.