Your Government Wants You to Use AI. The Question Is How Hard It Should Push.

Malta and Singapore have built the first real-world experiments in nudging entire populations toward AI fluency. The harder question is what comes after the nudge.

In 2008, Cass Sunstein and Richard Thaler published Nudge, a book that quietly rewired how governments think about citizen behavior. Their argument was modest in language and radical in implication: most people will do the smart thing if you make the smart thing easy. Auto-enroll workers in retirement plans. Default organ donors to opt-out. Put the salad at eye level. The state, in their telling, was not a regulator but a choice architect.

Eighteen years later, governments are quietly applying the same playbook to the most consequential skill of the decade. The question is no longer whether citizens should learn to use AI. It is how far a government can -- or should -- go to make sure they do.

Malta has the cleanest example. In May 2026, the Maltese government signed deals with OpenAI and Microsoft to offer every resident free ChatGPT Plus and Copilot for a year, on one condition: complete a short government course called "AI for All." The course is free, online, and accessible through Malta's national digital identity system, which most citizens already use to file taxes and check medical records. Friction: near zero. Default state: not enrolled, but the path is short, social, and frequently advertised.

This is a textbook nudge. No one is forced. No one is fined. The state has simply made one option -- learn AI, get the tool -- dramatically more attractive than the alternative, which is doing nothing and paying $20 a month yourself.

Singapore is running the same experiment with slightly different framing. Its February 2026 budget offered six months of free premium AI tool access to citizens who complete selected training courses, with a national target of 100,000 "AI-savvy workers" by 2029. The packaging is workforce development; the underlying mechanic is identical. Lower the cost of starting. Raise the cost of not starting. Let the population sort itself.

The spectrum from nudge to mandate

What both countries are testing, in real time, is where AI policy sits on a spectrum that has well-studied endpoints. At one end is pure nudging: defaults, information, friction reduction. At the other is mandate: licensing requirements, mandatory school curricula, professional certification gates. Most successful behavior-change policy lives somewhere in the middle, and the location matters enormously.

Vaccines are instructive. Most developed countries do not legally require adults to be vaccinated. They simply make it free, default-positioned at routine medical visits, and occasionally required for school enrollment. Compliance is near-universal because the path of least resistance leads to the needle. Retirement savings work the same way: auto-enrollment in 401(k)s lifted participation rates from roughly 60% to over 90% without changing a single law about whether people must save.

The Malta and Singapore approach maps cleanly onto this tradition. Course completion is voluntary, but the reward is real, the friction is engineered downward, and the cultural signaling is strong. "Everyone is doing this" is itself a nudge -- perhaps the most powerful one.

Other governments are testing harder versions. The UAE's "AI for All" with Google and its earlier "One Million Prompters" initiative use enrollment targets that function as soft quotas; agencies and employers are expected to deliver participation, not just offer it. The UK has moved most aggressively on its civil servants, where AI training is increasingly an implicit job expectation rather than a perk. Kazakhstan's commitment to train one million people by 2030 is a target with teeth -- the kind that gets translated into employer mandates once the numbers slip.

Each is a different point on the nudge-to-mandate dial. None has yet crossed into full coercion. But the dial is moving in one direction.

What we know about what works

The behavioral economics literature is unusually clear on which design choices matter. Defaults dominate everything. Friction kills participation faster than cost. Social proof matters more than incentives. Reminders work, but only for a short window. Financial rewards are surprisingly weak compared to status rewards or convenience.

The Malta program hits most of these levers. Tying the course to the national ID system is a friction-reduction masterstroke -- citizens don't sign up for a new account, they log in to one they already use. Branding it "AI for All" frames participation as civic, not commercial. Offering a real consumer product (ChatGPT Plus) rather than a generic certificate makes the reward tangible and the social proof visible. Within weeks, your neighbor is using it. Within months, not using it feels like the deviation.

Singapore's design is slightly weaker on the friction side -- its training pathways are sector-specific and require more navigation -- but stronger on workforce integration. The course-to-tool pipeline runs through employers, which means the social pressure is denser and more immediate.

The countries that will struggle with this kind of policy are the ones with weak digital identity infrastructure, fragmented education systems, or political resistance to public-private tech partnerships. The United States, for all its AI dominance, may be the worst-positioned major economy to run a Malta-style program. There is no national digital ID, no agreed-upon course standard, no political appetite for a federal subscription giveaway, and a deep cultural allergy to anything that smells like state-directed skill acquisition.

The harder question

Sunstein himself has spent much of the last decade arguing that nudges work best when the underlying choice is genuinely in the citizen's interest. Retirement savings: yes. Organ donation: defensible. AI fluency is trickier. It is plausibly in every citizen's interest, but it is also unambiguously in the interest of OpenAI, Microsoft, and Google, who get state-subsidized customer acquisition at unprecedented scale. The line between nudging citizens toward a useful skill and nudging them into a particular vendor's funnel is thinner than the press releases suggest.

The next phase of this policy experiment will reveal whether governments can stay on the right side of that line -- and whether the populations they are nudging will notice when they don't.