Dr Benjamin Krarup - University of Lincoln

The challenge

In law enforcement, defence and intelligence, experienced professionals make consequential decisions quickly, under uncertainty and with incomplete information – relying on intuition to do so. Those intuitions often reflect real expertise, but they can also carry bias. The AI systems being deployed to support these decisions engage with neither possibility. They operate on explicit models and cannot represent the implicit knowledge that intuition encodes. As such, when a plan and a practitioner's instinct conflict, there is no principled way to resolve it. The practitioner overrides the system or defers to it, and neither response is accountable. The opportunity is to make intuition something an AI system can recognise, question and use. 

The research

Dr Benjamin Krarup's work investigates how experienced professionals use intuition when they plan, and whether a machine can tell when they are relying on it. Working with practitioners in policing, defence and intelligence, it asks what separates an intuitive judgement from a deliberate one, and when instinct proves reliable. The aim is a framework, instinct in the loop, that recognises when a professional is drawing on intuition and separates hard-won expertise from bias. Built into a new approach to human-in-the-loop planning, it would let an AI act on the first and question the second, so that experience shapes the plan and bias does not.

The researcher

Dr Krarup's research has always centred on making AI usable: systems that users can understand, question, and trust. His work at King's College London established that the explanations people want are contrastive – people do not ask why a system chose something, but why it did not choose what they expected – and that explanation works best as a conversation rather than a single output. The framework he built on those findings, published in the Journal of Artificial Intelligence Research, is now a foundation for the field. He has led its principal international workshop since 2021, was invited to the Dagstuhl seminar on explainable AI for sequential decision making, and has applied the work with JP Morgan Chase, BAE Systems and Nightingale HQ.