DUBAI – At a time when public trust in government across established democracies has reached alarming lows, many worry that AI will worsen the problem by fueling disinformation and diminishing trust in facts themselves. But new AI tools could be part of the solution to democracies’ trust deficit.
According to the OECD, only 39% of citizens across member countries trust their national government, down from 45% in 2021. In the United States, Pew Research finds trust hovering near historic lows of 17%, with similar trends in France, the United Kingdom, and Australia. By contrast, the world’s most effective technocratic governments enjoy significantly higher trust levels, with scores exceeding 70% in Singapore and the United Arab Emirates. Even China’s authoritarian system dwarfs many Western democracies in this respect.
The conventional explanation for this gap – democracies invite criticism while autocracies enforce compliance – is too simplistic. The high-trust technocracies share something else. They competently deliver results while remaining responsive to public concerns. Expert-driven policymaking has been wedded to popular legitimacy.
This points to a deeper challenge facing democratic governments: the widening gap between “bounded rationality” and “abstract rationality” in policymaking. Here, bounded rationality refers to the domain of experienced insiders who produce policies shaped by political feasibility, public sentiment, and what has worked before. Abstract rationality is the realm of economists and technical experts who aim to optimize policies for efficiency. They tend to be more concerned with evidence and theoretical coherence than with real-world political constraints.
When bounded rationality dominates too much, policymaking feels cynical and poll-driven. Citizens sense that public officials are prioritizing political survival over solving problems. As technically superior policies (like a carbon tax) are abandoned for politically safer but less effective alternatives, trust erodes.
Similarly, when abstract rationality dominates too much, policymaking feels remote and tone-deaf. Governments roll out expert-designed reforms that look brilliant on spreadsheets, only to crash into political realities. Pension reforms that will save billions of dollars end up provoking weeks of costly strikes. A hospital restructuring that should improve outcomes ends up costing the health minister his job. Trust erodes because governments appear indifferent to legitimate public concerns.
The technocratic success stories have avoided this trap. Singapore’s government combines rigorous policy analysis with sophisticated insights into how policies will be received. Similarly, policymakers in Gulf Cooperation Council countries invest heavily in both technical expertise and mechanisms to gauge citizen satisfaction. In the UAE, almost all public service providers are equipped with customer satisfaction feedback booths. Rather than favoring one form of rationality over another, these countries have integrated them.
Western democracies struggle to do so, partly because governing authority is continually undermined by adversarial partisans. While economists point out that removing fuel subsidies would save money and reduce inequality, elected officials know that it would trigger a political earthquake. While treasury models underscore the necessity of pension reforms, polling data show that such policies are electoral suicide. As the two rationalities talk past each other, governments lurch between technocratic overreach and political capitulation, producing the paralysis that has come to define many democracies.
AI could help bridge this gap. Large language models (LLMs) exhibit a distinctive capacity for policy analysis. Unlike traditional decision-making models that optimize for predefined parameters, LLMs have absorbed how people actually talk about policy outcomes, reflecting moral concerns, emotional valences, underlying political narratives, and stakeholder perspectives.
For example, when analyzing a housing policy proposal, an LLM will not just evaluate economic efficiency. It can also flag language (“luxury developer”) that is likely to trigger class-based opposition, or terms (“family neighborhoods”) that may alienate younger voters. It might also find that similar policies succeeded in jurisdiction X but failed politically in jurisdiction Y, despite comparable economic conditions.
In my experimentation with AI-augmented policy analysis for government clients, I have found that these systems excel at what I call “sentiment-aware policy design.” While traditional tools might show that a congestion charge reduces traffic by 22%, AI systems can remind you that the term “congestion charge” polls substantially worse than “clean-air fee”; that implementation during election years multiplies political risk; and that exempting delivery vehicles creates coalition-building possibilities with small-business groups.
The point isn’t to replace human judgment. It is to make experienced insiders’ implicit political knowledge more explicit, systematic, and testable. With AI, the abstract-rationality crowd gets quantitative rigor, the bounded-rationality practitioners get political intelligence, and – crucially – both can see the other’s perspective clearly.
Moreover, when combined with web-search capabilities, AI tools can contribute near-real-time sentiment analysis. This matters because policies designed to address last quarter’s concerns might no longer fit the political terrain when they launch in the coming quarter. By the time a pension reform reaches parliament, murmurs of a recession may have changed voters’ priorities entirely.
AI-powered analysis can reveal how specific issues are being discussed across news, social media, parliamentary debates, stakeholder communications, and other channels. It can identify rising concerns and flag when a window of political opportunity has opened or closed. Such insights can help governments counter the perception that they are slow, deaf, and disconnected from everyday realities. AI cannot make governments omniscient, but it can make them more responsive and less blind to the political consequences of technical decisions.
The high-trust technocracies succeed partly because they have systematized the integration of technical excellence with political responsiveness. Now, AI offers democracies the means to do so as well.
To be sure, LLMs can reproduce biases, sometimes hallucinate (fabricate responses), and demonstrate a lack of deep contextual understanding. They cannot replace the minister who knows coalition partner X will never accept policy Y, or the permanent secretary who personally remembers policy Z’s catastrophic failure in 1997. But they can reveal analytic blind spots, make tacit knowledge visible and shareable, help experts understand why technically optimal policies may be political nonstarters, and enable officials to identify promising policy modifications.
Restoring trust in democratic government requires delivering competent policies that citizens recognize as both effective and responsive to their concerns. AI alone will not solve democracies’ challenges, but it can help them bridge the rationality divide that has paralyzed policymaking. It offers tools for maintaining both legitimacy and effectiveness – the combination that high-trust governments, democratic or otherwise, have mastered.
Sami Mahroum, Founder of Spark X, previously held posts at INSEAD, the OECD, and Nesta.
Copyright: Project Syndicate, 2026.
www.project-syndicate.org
