AI and The Devil's Advocate

I recently returned from Cyprus, where I presented my latest research at the 2026 European Health Psychology Society Conference, with support from the University of Jyväskylä. While there, I attended a fascinating keynote by Professor Alex Gillespie from the London School of Economics and Political Science.

"His research examines speaking up and listening, differences of perspective, and misunderstandings. Questions include: How do we escape our own blind spots through dialogue with others? How do we dismiss and ignore each other – so as not to learn from each other? And, more pragmatically, how can technologies support these exchanges of perspective?" (LSE Profile).

This keynote made me think.

The Catholic Church gave us one of the most enduring ideas in critical thinking: the Devil’s Advocate. In the canonization process, the promotor fidei was charged with scrutinizing the case for sainthood, challenging evidence, questioning testimony, and testing whether the claims could withstand serious objection. The logic was simple but powerful: before accepting something as true, appoint someone to argue against it. Yet, history also revealed the limits of that principle. When Galileo Galilei defended a heliocentric view of the cosmos, his conclusions challenged the prevailing interpretation upheld by Church authorities, and in 1633 he was tried by the Roman Inquisition, forced to recant, and placed under house arrest. The institution had created a formal role for questioning belief, but struggled when questioning reached beyond the boundaries it was prepared to reconsider.

The deeper lesson is not about religion or astronomy. It is about the difficulty of challenging our own convictions. Take Nokia in 2010 (Siilasmaa, 2018). It had dominated the mobile phone market and had every reason to believe that the assumptions behind its success would continue to hold. The problem was not simply that it failed to see change coming; organizations built around a successful model often become exceptionally good at defending that model. Evidence that supports the prevailing view travels easily upward, while evidence that threatens it is questioned, softened, or ignored. By the time the smartphone had redefined what a phone could be, the most dangerous assumption was no longer about technology. It was the belief that the rules of the market had not fundamentally changed. Which they had. The smartphone industry was no longer about the capabilities of the physical device but rather about its operating system and compatibility with diverse apps (Siilasmaa, 2018).

It is far too easy to look back at the Catholic Church or Nokia and wonder how they failed to see what now appears obvious. The harder question is where we are doing the same thing today. I hear versions of it across industries: “Builders and farmers will be safe from AI.” “Doctors cannot be replaced.” “People will always need smartphones.” Perhaps. But those assumptions become less comfortable when humanoid robots are learning to perform physical tasks, AI systems are already matching or exceeding specialists in some diagnostic settings (Tu et al., 2025), robotic knee surgery and companies are pushing computing from the phone onto glasses, wearables, and eventually perhaps entirely new interfaces. Apple may continue releasing smartphones for decades, or the smartphone may one day look as transitional as the keyboard on a Nokia. Equally, the current AI boom may disappoint. Models may plateau, economics may fail to work, regulation may slow adoption, or the technology may encounter limits we cannot yet see. Or perhaps we are all doomed (existential risk from artificial intelligence: Livesay & Labiak, 2026). Who knows? And that is precisely the richness of the Devil’s Advocate.

The value of the Devil’s Advocate is not in predicting which future will arrive. It is in forcing us to seriously consider the possibility that the future we are most comfortable with may be the wrong one.

This is where ai artificial intelligence becomes genuinely interesting (LLMs such as gemini ai, claude, chat gbt ai). Not because it can predict the future, but because it can challenge the future we have already decided to believe in, and do so with access to an extraordinary breadth of information. A human Devil’s Advocate is limited by personal experience, expertise, and time. An artificial intelligence app could draw on patterns from history, economics, technology, medicine, psychology, strategy, and countless other domains in seconds. Inspired by de Bono’s (1985) idea of deliberately separating different modes of thinking, and Gillespie’s (2026) proposal to use AI to support exchanges of perspective, we could assign different AI models different roles: the skeptic, the historian, the contrarian, the empiricist, the futurist, the operator, the economist, and ask each to attack the same assumption from a different angle. Suddenly, AI is no longer an answer machine. It becomes a room full of Devil’s Advocates (Gillespie, 2026).

And perhaps this is only the beginning. Today, we might imagine a handful of AI Devil’s Advocates examining the same decision from different perspectives. But there is no reason the number has to stop at six or eight. There could be thousands, even millions, of AI agents testing the same analysis (Gillespie, 2026): each working from a different assumption, dataset, discipline, time horizon, incentive structure, or theory of how the world works.

Perhaps the future of artificial intelligence is not one superintelligence giving us the answer, but millions of Devil’s Advocates making it much harder for us to fool ourselves.

This is the kind of work we want to build at Lasse Eini Co. Artificial intelligence solutions that do not simply help organizations move faster, but help them think better. We design artificial intelligence solutions for companies to use multiple AI perspectives to challenge assumptions, test strategies, explore alternative futures, and surface risks that might otherwise remain invisible. In other words, we are interested in turning the Devil’s Advocate from an occasional role in the room into a permanent capability of the organization.

Get in touch to learn how this may look for your team — get in touch here.

References

de Bono, E. (1985). Six Thinking Hats. Little, Brown and Company. 

Gillespie, A. (2026, September). Using AI to analyze patient voice and hospital listening: Insights into patient safety [Keynote address]. 40th Annual Conference of the European Health Psychology Society, Pafos, Cyprus.

Livesay, B., & Labiak, M. (2026, September 13). AI staff ‘genuinely frightened’ for humanity’s future, ex-Anthropic researcher tells BBC. BBC News.

Siilasmaa, R. (2018). Paranoid Optimist: This is how I led NOKIA. Tammi.

Tu, T., et al. (2025). Towards conversational diagnostic artificial intelligence. Nature.