From Human IN the Loop to Human ON the Loop

Wednesday, January 27, 2027 at 5:30 PM to 6:00 PM · 30 min. (Europe/London)
AI in Drug Discovery Stage
Day 1 - Wednesday 27th January
AIAI in Drug DiscoveryDrug Discovery

Information

  • Human-in-the-loop is usually framed as a person checking a model's outputs. I will argue that the loop needs to begin much earlier, during the build, with subject-matter experts acting as genuine co-developers rather than downstream end users. 

     
  • Why do good tools still fail to gain adoption? Insular development can produce systems that scientists perceive as black boxes and are consequently reluctant to use when making decisions. Addressing this requires effort from both sides: technical teams must genuinely open up the development process, while domain experts must build enough technical understanding to engage meaningfully with it. Explainability alone is not sufficient because algorithmic transparency and organisational trust are different problems. 

     
  • Trust is not permanent. Even well-validated models can drift, and moving from closely observing every step to monitoring a system from a distance can reduce the likelihood that changes are noticed or properly understood. A dashboard may show that something has changed, but determining why requires someone who is accountable for both the science and the system. Keeping a tool reliable in operation therefore requires the same scientific involvement that was essential during its development.