The research, conducted by Hiromi M. Yokoyama of The University of Tokyo’s Kavli Institute for the Physics and Mathematics of the Universe and Shuo Wang of Tsinghua University’s School of Marxism, argues that the instruments a government selects reveal its underlying assumptions about the nature of the problem it faces. In most comparative debates about AI governance, the central tension is framed as innovation versus risk: how to encourage a transformative technology while limiting its harms. Japan’s governance problem, the authors contend, has a fundamentally different structure. The urgency there comes not from the dangers of AI being used, but from the economic and social cost of AI going unused.

That framing reflects a distinctive set of social conditions the study documents in detail. AI adoption in Japan has lagged well behind other advanced economies, a gap that persisted even as generative tools spread globally after 2022. Public trust in AI is notably low, which suppresses willingness to integrate the technology into daily workflows. At the same time, the country faces a demographic squeeze unlike almost any other wealthy nation: an ageing population is shrinking the labour force precisely as the engineers who maintain Japan’s ageing industrial IT systems approach retirement. Labour shortages run across agriculture, nursing, construction, retail and logistics. Against this backdrop, AI is cast not as a threat to be contained but as a substitute for labour that Japan’s demographic structure can no longer supply. A regulatory regime designed to slow deployment would, in this reading, compound the very shortage the technology is meant to relieve.

One of the study’s most striking findings concerns the psychology of Japanese attitudes toward AI. Drawing on cross-national surveys, the researchers found that Japanese respondents score among the highest of any population on objective tests of AI knowledge — people know the facts about how these systems work and what they can do. Yet the same respondents rank lowest in confidence about their own understanding. This combination of high objective knowledge and low subjective certainty produces a distinctive pattern of reluctance. The authors conclude that Japanese hesitancy concentrates in trust rather than acceptance: people do not doubt that AI functions, they doubt whether they, their employers and their institutions can rely on it safely. That distinction matters for policy, because trust deficits respond to different instruments than acceptance deficits. Awareness campaigns and demonstration projects can close knowledge gaps; building confidence in reliability requires sustained institutional signals.

The study also explains why a promotion-first approach could function at all within Japan’s political culture. The country has a long tradition of administrative guidance, in which government agencies steer private-sector behaviour through non-binding recommendations rather than legal compulsion. Corporations, in turn, operate within a business culture that is highly attentive to compliance and reputation, making voluntary guidelines behave like real constraints in practice. Close coordination between ministries and organised business reinforces this dynamic. Within this institutional ecology, promotional instruments — model guidelines, government-backed demonstration programmes, procurement signals — can carry genuine weight without statutes specifying penalties. The Act’s central organisational innovation, an AI Strategy Headquarters chaired by the Prime Minister, is designed to overcome the fragmentation of authority across ministries that has hampered Japanese technology policy in other fields, concentrating coordination at the centre of government.

On the question of results, the evidence is genuinely mixed, and the mixedness is the story. In the period since the Act was passed, personal use of generative AI rose sharply, and trial use within firms approached international levels. By the measures promotional policy can move — individual experimentation, corporate pilots, public familiarity — the strategy has worked. But the authors identified a second dimension that promotional instruments do not reach. The gap between large firms and small and medium enterprises in actual deployment has not narrowed, and self-assessed transformation maturity among Japanese organisations remains near the bottom of the international scale. Firms are trying AI; they are not transforming with it.

Wang’s explanation cuts to the technical heart of the adoption problem. Adoption, she argues, is not a single quantity but a layered process in which trial and organisational integration respond to entirely different instruments. Promotional measures lower the cost of trying: they subsidise experimentation, provide templates, reduce the reputational risk of a first pilot. But moving from a pilot to deep integration requires capital investment in system replacement, the technical personnel to redesign processes around new tools, and the managerial authority to reorganise how decisions are made and work is distributed. These are precisely the resources that smaller firms lack. A subsidy or a guideline cannot conjure an in-house engineering team or fund the replacement of a legacy system whose maintenance engineers are retiring. The result is a policy that successfully widens the front door of adoption while leaving the interior of organisations largely unchanged.

The Japanese government appears to have internalised this diagnosis. Its second AI Basic Plan, adopted in July 2026, shifts the objective from encouraging use toward organisational change — altering how decisions are made and how work is structured, rather than merely increasing the number of people who have tried a chatbot. This evolution suggests the promotion-versus-regulation framing that dominated early comparative debate may itself be too coarse, and that governance instruments must match the specific bottleneck in the adoption pipeline at each stage. Japan’s experience indicates that where adoption is low, the binding constraint is not risk appetite but organisational capacity, and the tools that raise trial rates do not automatically raise integration rates.

The authors also flag structural risks embedded in the promotional model itself. Voluntary compliance, they caution, offers weak protection against rare, high-impact events — the low-probability, severe-consequence failures that binding regulatory frameworks are specifically designed to anticipate. A governance architecture optimised for encouragement has little machinery for responding to a serious incident, and the absence of such machinery becomes apparent only when it is needed. Beyond this, codifying a promotional orientation in statute may create path dependencies: once a legal framework institutionalises promotion as its organising principle, shifting toward regulatory instruments when circumstances warrant them becomes politically and legally harder. The law’s very success at framing AI as an opportunity could make future course corrections more difficult.

Yet the study identifies a built-in escape valve. The Basic Plan mechanism includes annual revision, a channel for adjustment that survived even a change of premiership, allowing policy to be recalibrated without reopening the underlying statute. This iterative design may prove to be the most consequential feature of Japan’s model: it treats governance as a continuously updated process rather than a fixed settlement. As Yokoyama observes, a governance model may succeed on its own terms while the underlying problem persists in a form its instruments cannot readily reach. Japan’s case shows that the global debate over AI governance, dominated by the innovation-versus-risk balance, misses an entire class of countries whose challenge is not how to slow a technology down, but how to make it take hold — and that solving that problem demands instruments the current comparative conversation has barely begun to describe.

Subject of Research: Japan's promotion-centred AI governance policy and its social and economic roots

Article Title: Why Japan chose to promote AI rather than regulate it — and where that strategy hits its limits

Article References: Why Japan chose to promote AI rather than regulate it — and where that strategy hits its limits. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: artificial intelligence, AI governance, Japan, policy, regulation, AI adoption, Risk Sciences, generative AI, labour shortage, administrative guidance, small and medium enterprises, organisational transformation