Skip to content
Kean CS Lab

Program 01 / 06

Trustworthy & Human-Centered AI

How should learning systems explain themselves — and to whom — when they enter classrooms, clinics, and civic software?

Kean CS faculty work on explainable and human-centered AI: not only accuracy, but whether a person can form a usable mental model of a system, detect failure, and refuse a bad recommendation. The thread runs through XAI, bias-aware evaluation, and computing-education research on how students and practitioners actually encounter AI tools.

Questions the work is for

  1. 01

    What explanations help non-experts calibrate trust rather than merely increase it?

  2. 02

    How do we evaluate models when the user population is socioeconomically and linguistically diverse?

  3. 03

    Where should a doctoral project sit between model internals and the interface that carries them?

Methods, data, benchmarks

method

Explainable AI (XAI)

Feature attribution, concept-based explanations, and human-subject studies of explanation usefulness — not explanation volume.

method

Human-centered evaluation

Task-based studies, think-aloud protocols, and sociotechnical analysis of how people register, contest, and recover from model errors.

benchmark

Fairness & robustness suites

Standard classification and language benchmarks paired with subgroup analysis; accent and dialect stress tests for speech models.

tooling

Inclusive design curricula

Public departmental work on embedding equitable and inclusive design across undergraduate CS — a living lab for how AI literacy is taught.

Taxonomy on this page: method, dataset, benchmark, tooling. Field-standard methods are named as such; they are not automatic claims of a Kean dataset release.

Doctoral tracks

Explanation as an interface problem

Adenuga’s stated focus on human-centered and explainable AI sits next to Ma’s HCI/LLM work: a natural doctoral spine is the design of explanations that are measurable in human terms, not only in fidelity metrics.

Evaluation under real users

Kwak, Hu, and collaborators have publicly examined how automatic speech recognition handles accents. That is the evaluation posture we want for doctoral work: the benchmark is incomplete until the population is.

AI in the curriculum

Morreale, Li, Kumar, and Kwak publish on computing education and inclusive design. Ph.D. students can treat pedagogy itself as a research site — instruments, not anecdotes.

Who this is for

Strong fit if you want a dissertation that binds model behavior to human judgment — XAI, HCI for AI, or evaluation methodology. Bring a quantitative backbone and a willingness to run studies with people, not only with leaderboards.

Ph.D. Program →

Faculty in this program

Selected papers

← All programs