Team work and tech
As artificial intelligence moves from chatbots and recommendation engines into high-stakes environments such as healthcare, defence and maritime operations, one question is becoming increasingly urgent: how do we know when AI is actually helping humans work better and more safely?
That’s the focus of a new study from researchers at the University of Nottingham, which examines how people and autonomous AI systems function together in “human-autonomy teams”. Rather than treating AI as a simple tool, the research explores what happens when AI behaves more like a teammate – offering advice, making recommendations and influencing decisions.
The research was presented as a poster by lead author Ben Bowers and won Best Paper at CIEHF’s Ergonomics & Human Factors 2026 conference. It reviewed 90 studies involving AI systems in safety-critical settings, from medicine to military operations.
The authors found that many organisations still evaluate AI mainly on technical performance, overlooking crucial human factors such as trust, communication and decision-making. That gap, the authors argue, could create dangerous blind spots in real-world use.
To investigate further, the team created AI-generated simulations of a maritime emergency involving different levels of AI involvement. Human factors specialists then assessed how well the teams functioned. The results revealed a surprising tension: while AI could improve information flow, higher levels of autonomy also reduced discussion, challenge and independent thinking among human team members.
The study said: “As autonomous agents take on roles once reserved for people, making teaming more than a purely human endeavour, test and evaluation approaches must evolve accordingly. Established schools of thought, namely human teaming and human-automation interaction have scaffolded human-autonomy teaming research, but simultaneously risk casting a shadow that obscures the fields distinct challenges and possibilities. This risk is particularly salient in safety-critical domains that demand not only technical excellence but social fluency from contemporary machines with autonomous capabilities.”
The study’s central message is clear: successful AI systems should not only be accurate, they must also support healthy teamwork. As autonomous technologies become more common in hospitals, transport systems and emergency response, understanding how humans and AI cooperate may prove just as important as the technology itself.
