AI’s Deepening of the Democratic Deficit

Politics is messy, and AI has deepened the democratic deficit. However, this messiness is preferable to the faux-omniscience of AI-Assemblages, which manufacture an illusion of knowing everything. A mirage that crafts a feeling of pervasive awareness, unlike true omniscience, a divine attribute of complete knowledge.

Today, it is far more preferable that AI systems not merely be “trained” on the data crumbs we sprinkle everywhere like Hansel and Gretel, but that AI models be built and used in ways that revere our preferences, perspectives and priorities. After all, AI is a human. For this reason, AI systems are more likely to amplify rather than limit existing inequities, and radicalise rather than reconcile opposing opinions.

This democratic deficit is now nested within a black-box chasm in which powerful AI development and governance are concentrated in cloistered communities, which exclude broad public participation. This leads to decisions that may undermine accountability, risk exacerbating inequality and misinformation, threaten core democratic values, and not serve the common public good.

This deficit is a shortfall in democratic principles in the creation and control of AI, leading to power imbalances and risks to societal well-being. This overconcentration in a few silicon cities can create an “oligarchic AI” scenario. Standard-setting bodies and policy discussions often exclude broader societal voices. The result is AI-driven misinformation flooding the media, and “vibe coding” that erodes trust in public institutions.

Debates often get stuck on technical fixes, avoiding deeper, inherently political discussions about what values AI should embody and who benefits. AI’s widespread effects collide with its concentrated governance, creating a misalignment between its broad stakeholdership and public opinion. In essence, it is worrisome that decisions about a technology that affects everyone are being made by a few without sufficient democratic input.

Pierre Bourdieu argued that “public opinion does not exist”. He cited three assumptions about public opinion that, he argued, do not hold: first, that everyone has a view; second, that there is agreement on the question; and third, that everyone’s opinion is of equal value.

Round-tables, citizens’ juries, and other immersive methods for eliciting public opinion reveal that, contra Bourdieu, most people “have an opinion”. Yet this does not resolve the second contention, that “everyone agrees on the question”.

While John Rawls, Jürgen Habermas and Kwasi Wiredu have championed the idea that societal consensus, or something resembling it, is possible and desirable, others, such as Amartya Sen, Chantal Mouffe, Nancy Fraser and Catherine Squires, have critiqued both its plausibility and its desirability.

What is needed is a set of public institutions that can ensure the inclusion of marginalised voices. Convenors of the voices of the subalterns to create a shared understanding of the ethical issues arising from data and AI.

Of course, for legal and policymaking purposes, the traditional Westphalian notion of bounded sovereignty remains dominant. While it is legitimate for national governments to address their own AI democratic deficit in terms most germane to their populations and development traps, to focus on the nation state as the pre-eminent information agent even in a single society is increasingly naive, as multi-agent systems that transcend physical borders become ever more prominent, and cross-border problems become more prevalent.

The EU AI Act has been criticised for its “human-centric” terminology, which is problematically anthropocentric; a confused risk-based approach; and the potential for a less flexible and effective framework than the General Data Protection Regulation (GDPR).

The “human-centric” language in the Act is a perilously vague relic that can be interpreted as human-centred at the expense of other concerns, like the earth. There is a pressing need to consider the world, which has suffered from humanity’s obsession with its own centrality, and to adopt a more holistic approach with urgency.

While the Act appropriately bases its approach on protecting human dignity and fundamental rights, it seems more top-down, less flexible, and less focused on the specific protection of individual citizens’ rights compared to the General Data Protection Regulation (GDPR).

The ambiguities in definitions and the varied nature of AI technologies can lead to legal uncertainty for developers, operators, and users, potentially hindering effective compliance. The Act is an essential but imperfect step, with philosophical and practical gaps that could limit its long-term effectiveness in balancing innovation with ethical safeguards.

To make life simpler for companies across the 27-nation bloc, the European Commission has unveiled a “digital simplification package”. This revamp will tackle public fatigue with constant pop-ups.

Companies will gain greater freedom to use datasets, including certain types of personal data, to train AI models for “legitimate interests”.  The simplifications will make European companies more competitive while maintaining a high level of protection for individuals’ fundamental rights and data privacy.