What can behavioural science bring to AI governance?

After a long summer break, Artificial Thought is back with news! I needed some time off from the world of AI before immersing myself in it even more deeply in the form of starting a Master’s degree in AI & Governance at Vrije Universiteit in Amsterdam.

And what better way to announce a pivot into AI governance than letting an AI ask the awkward questions? The questions are real, the answers are mine, and the irony is intentional.

Another degree? Really?

Yes, I know how it looks, but after fifteen years working in behavioural science, I found myself at an inflection point. I could keep applying the same frameworks to familiar types of problems, or I could acknowledge that the pace and scale of AI adoption had created new types of challenges my existing toolkit wasn't built for.

Behavioural science became mainstream by helping people understand what happens when policies, systems, or products are designed without real humans in mind. AI is creating a similar rupture, only this time the stakes are higher. I wanted to be part of the solution, not the problem —

Wait, so I'm a problem?

Well, some people think so, but this interview is about me - not you.

Touché. Continue.

As I was saying, while I didn't need another degree, if I wanted to work meaningfully in this space, I did need new lenses and new grounding, and a new way to connect what I already knew to what's coming next.

AI governance - bit of a leap from nudging people to fill out forms, isn't it?

That's exactly the problem - everyone thinks behavioural science is just small tweaks and optimisation. Meanwhile, AI is being built, shaped and sold by a remarkably homogeneous group of people, which is a fragile foundation for any technology, let alone one reshaping society.

I wasn't going to retrain as a machine learning engineer, but governance isn't about code. At its core, it's about human behaviour: how people respond to rules, whether they trust systems, why they comply or resist. This degree builds on what I already know about human behaviour and decision-making, and anchors it in the governance challenges this technological shift is creating. Bonus: I don't have to abandon my expertise or pretend to be something I'm not.

AI governance seems pretty well covered by lawyers and engineers. What exactly is behavioural science going to add that they're missing?

The human element! Policies and governance frameworks look great on paper until they meet real people. AI governance is still dominated by law, ethics, and engineering - all essential, but they don't tell you what actually happens when humans are asked to follow, enforce, or trust those rules.

We've seen the same in public policy: elegant designs collapsing on contact with everyday behaviour. The question isn't whether systems are technically sound, but whether they'll survive first contact with reality. That's where behavioural science matters: connecting what policy aims to do with how people actually respond in practice.

AI governance is no different and it reminds me of the early days of behavioural public policy. Behavioural insight needs to become part of the foundations of AI governance, not an afterthought once problems emerge.

But isn't behavioural science just for making apps addictive and tricking people into buying things?

That is the stereotype, especially in tech - behavioural science pressed into service for engagement hacking and conversion optimisation. But here's the thing: if design teams are already using behavioural insights to shape how people interact with technology, shouldn't governance be using them too? Otherwise you end up with systems perfectly optimised for user engagement but completely unprepared for accountability.

We've seen this pattern before: financial services adopted behavioural insights long before regulators caught up, creating a gap that took years to close. AI governance can't afford the same lag it needs behavioural insight built in from the start.

Behavioural science, AI, and public administration walk into a bar… sounds like the set-up of a bad joke?

It does sound like a setup, but the connection is real! Fun fact: Herbert Simon is considered a founding figure in all three: his work on bounded rationality defined behavioural science, his book Administrative Behavior shaped public administration, and together with Allen Newell he built the Logic Theorist and General Problem Solver, programs often described as the birth of AI.

He later won the Nobel in Economics for his research on decision-making, so I ll appeal to authority here and claim it as proof that combining these isn t as crazy as it might seem!

I’m also not abandoning BeSci - everything I've learned about human behaviour, decision-making, and system design remains relevant; I'm just applying it to different problems at a different scale.

Aren’t you too old for this?

I prefer mature, thank you very much. But yes, compared to my classmates I m practically a dinosaur because I entered adulthood before social media and the algorithmic internet - I remember when email was novel, music was on CDs and when the internet was still (mostly) a Good Place!

I’ve lived the before-and-after of digital trust shifts - when privacy was assumed rather than negotiated, when platforms were just tools, when trust in digital systems was almost automatic. Governance benefits from multi-generational perspectives, not just the optimism or cynicism of youth. The people making decisions about AI will span generations and the expertise shaping those decisions should too.

Some people say AI just a fad and a bubble waiting to burst - is it wise to hitch your wagon to something like this?

If AI is a fad, it’s a fad that’s already running credit checks, scoring job applications, and shaping parole decisions. Generative models may well be in a bubble, but algorithms aren’t going anywhere, and the governance challenges won’t vanish with the headlines.

This is not really about specific technologies anyway, but about how institutions manage systems that shape human behaviour at scale. Generative AI just makes the gap obvious. It’s been rolled out at breakneck speed with almost no behavioural planning - the sort of groundwork other sectors do as a matter of course. Adoption depends on how people adjust their expectations, trust, and everyday practice. That hasn’t had much airtime, but it’s exactly where behavioural science has something to add. Besides, the course covers public administration more broadly: understanding how institutions change, how policy gets implemented, and how complex systems can be governed. Those skills remain valuable even if the AI component fades.

Finally… are you sure you know where you’re going with all this?

I know the direction, but I'm keeping the exact destination open - part of the point of study is leaving room for new ideas, detours, and discoveries. What I do know is the kind of work I want: grounded in reality, collaborative, and useful outside the academic bubble.

Some of the routes I’m interested in exploring include:

• applying or adapting behavioural frameworks like the Behaviour Change Wheel or TDF to AI governance to test whether tools already used in health and finance could help policymakers design for real-world behaviour;

• looking at cultural and generational differences in how people perceive AI because attitudes aren t universal, and what feels legitimate in one context can be mistrusted in another;

• understanding how the absence of women and other marginalised groups in AI decision-making creates blind spots i.e. when systems are built on assumptions that don t reflect their experiences, risks or needs.

If you’re working in these areas, or have perspectives, examples, or challenges that you think might be relevant, I d like to hear them. At this stage, ideas and viewpoints are as valuable as data points.

Ambitious. We'll see how it goes.

True. And if you turn out to be exhibit A in why we need AI governance, I'll be ready.

===