I’m an assistant professor at Clemson, where I direct the BIG CAT Research Group and co-direct CU-CHAI. My work focuses on human-AI teams, trust, training, and performance.
A way to design and evaluate AI that starts with people, work, and consequences, not just model performance.
The developer? The interface designer? The person using it? The organization? Someone else?
Turns inputs into a score, label, prediction, or generated response.
Adds an interface, thresholds, defaults, explanations, and controls.
Includes the people, policies, incentives, infrastructure, and communities involved.
Reports, records, sensors.
Score, classify, generate.
Display, alert, rank.
Accept, edit, defer.
Allocate, deny, dispatch.
Benefit, burden, harm.
Little help and little control over outcomes.
The system acts while people absorb the failures.
Strong control with limited computational help.
People can direct, inspect, correct, and recover.
Detect a pattern in text, image, audio, or behavior.
Estimate a future state, risk, demand, or outcome.
Rank options and direct human attention.
Create content, plans, code, or candidate designs.
Call tools, transact, allocate, communicate, or control.
A person sees an output.
The system orders attention.
Inaction becomes a decision.
The system changes the world.
The question is whether that person can notice, judge, and act.
Agency requires real options, enough understanding to choose, and controls that make the person’s intention matter.
A dispatcher, auditor, affected resident, and model engineer need different evidence about the same system.
Reject useful support despite evidence that it fits the task.
Use or reject based on capability, context, evidence, and stakes.
Accept output beyond the system’s competence or authority.
Voluntary engagement when automation is suitable.
Over-reliance or use outside the competence boundary.
Under-reliance that rejects beneficial automation.
Automation imposed without adequate regard for human consequences.
Know that AI influenced the outcome.
Access the reason and relevant evidence.
Repair data, context, or assumptions.
Reach an empowered, independent reviewer.
Reverse harm and improve the system.
It limits action before an incident and makes recovery possible afterward.
Thresholds, workflow, alert burden.
Learning evidence, access, integrity.
Metrics, proxies, policy choices.
Purpose, inference, correction.
Ecology, allocation, resilience.
Context, language, dependency.
Boundaries, rebound, lifecycle.
Authority, audit, shutdown.
Purpose: What human goal justifies the system?
Evidence: What supports this use in this context?
Power: Who can direct, question, correct, and stop it?
Consequences: Who benefits, who carries risk, and who can recover?