Richard Lush

I'm interested in understanding what happens when people begin to trust AI systems, particularly when those systems are helping them make important decisions.

Human-in-the-loop is often presented as a safeguard for AI, with the assumption that keeping a person involved means decisions remain under human control. But is that really the case? If someone trusts an AI recommendation, are they still critically evaluating it, or are they simply accepting what the system tells them?

My research explores how trust influences human decision-making and the effectiveness of oversight in AI-assisted systems. I'm particularly interested in the psychological and neurobiological mechanisms behind trust, and how factors such as time pressure, task complexity and individual experience affect our willingness to accept or challenge AI recommendations.

I'm approaching this from a cybersecurity perspective, drawing on more than 25 years of experience in security operations and leadership. Cybersecurity provides an interesting environment for studying these questions because decisions often need to be made quickly, with incomplete information and potentially significant consequences.

Ultimately, I want to understand what makes human oversight effective, how we can measure it, and whether keeping a human in the loop genuinely improves AI safety.