Mark Zuckerberg’s AI Empowerment Vision: Individual Access Without Collective Power
103- 01/08/2026- A promising but structurally incomplete vision
In July 2026, Mark Zuckerberg* presented a vision in which artificial intelligence should not remain the exclusive privilege of governments or a small number of technology companies, but should be made broadly accessible. According to this vision, advanced AI would expand individual capabilities, stimulate creativity, increase economic opportunity, and strengthen personal freedom.
This vision is appealing, and it echoes earlier promises made of the Internet: that a technology can democratize knowledge and redistribute power simply by becoming widely available. AI already allows millions of people to learn, translate, create, and access forms of expertise once reserved for specialists.
However, a more rigorous reading must ask not whether AI can empower, but whether the conditions required for empowerment — as distinct from access — are currently in place. This article argues that they are not.
Empowerment is a political process, not a technical outcome
An individualistic conception of empowerment holds that people become empowered when they acquire new knowledge, skills, or tools. Under this view, AI assistants are, by definition, empowering instruments.
Decades of empowerment research contradict this simplification. Empowerment is not solely an individual attribute; it is a social and political process that depends on people’s ability to understand their environment, participate in decisions, influence organizations, and collectively transform the systems shaping their lives.
A person equipped with a highly capable AI assistant may become more productive and better informed while remaining entirely without influence over the institutions, markets, or social structures that govern their life. Individual capability and collective power are not the same variable, and increasing one does not mechanically increase the other. On this point, the evidence available today favors caution: no widely deployed AI system has yet demonstrated a measurable increase in citizens’ collective capacity to influence institutions or policy.
The conditions for empowerment are known — and currently unmet
Genuine empowerment requires that people understand the systems that claim to empower them. This condition is not currently satisfied.
Every AI system encodes choices made by its designers: which data are selected, which objectives are optimized, which forms of knowledge are treated as legitimate, and which risks are considered acceptable. Today, these choices remain, for the overwhelming majority of users, opaque. Model architectures, training data, and ranking or moderation criteria are typically confidential. Users interact with systems whose underlying logic they cannot inspect, contest, or meaningfully influence — the opposite of the condition empowerment requires.
Likewise, empowerment presupposes that individuals understand how their data are collected, analyzed, stored, and monetized. Data are not a neutral technical resource; they are a source of power. At present, the commercial model of the largest AI providers depends on the large-scale collection and use of user-generated data, over which users typically exercise minimal control and from which they derive negligible direct economic benefit. This is not a peripheral detail: it is a structural condition that currently works against collective empowerment rather than for it.
Democratizing access to AI therefore requires more than distributing the technology. It requires transparency about data and model design, explainability of algorithmic decisions, accountability mechanisms with real consequences, and formal avenues for user and citizen participation in AI governance. None of these four conditions is presently institutionalized at the scale of major consumer AI platforms.
From individual assistance to collective empowerment: three dimensions, three unmet prerequisites
Research on collective empowerment, particularly in the study of patient and citizen communities, identifies three complementary dimensions — each of which depends on a prerequisite that current AI deployment does not satisfy.
Community empowerment—the development of shared knowledge, mutual support, and collective identity—requires that platforms be designed to strengthen horizontal ties between users rather than optimize for individual engagement with the system itself. This is where the gap between current design and the stated ambition is most visible: most consumer AI products today are architected around a one-to-one interaction between a single user and a model, not around facilitating peer connection, mutual recognition, or the accumulation of shared knowledge across a community. This is not incidental; it follows from the same engagement-optimizing logic that underpins the accountability problem raised earlier regarding data collection, since a product built to maximize individual usage has little structural incentive to redirect that engagement toward horizontal ties between users. Zuckerberg’s vision of AI as personal empowerment fits this architecture closely, but it offers no equivalent architecture for community formation. Until platforms are deliberately redesigned to support peer connection rather than individual retention, the infrastructure required for community empowerment will remain largely absent, regardless of how capable any single assistant becomes.
Collaborative empowerment—the co-creation of solutions by citizens, professionals, organizations, and institutions—requires interoperable tools, shared data standards, and deliberative processes in which AI plays a supporting rather than a controlling role. This is a distinct and higher bar than community empowerment, since it presumes not just connection between peers but structured cooperation across different types of actors—citizens, experts, organizations, and institutions—each contributing distinct forms of knowledge to a shared outcome. Achieving this would require AI systems capable of operating across organizational boundaries, alongside common standards for how data and analysis are shared and interpreted between them, and deliberative processes designed to weigh contributions fairly rather than simply aggregate them. None of these governance arrangements exists in any generalized form today; where collaborative AI tools do exist, they typically remain confined within a single organization or platform, which limits them to facilitating cooperation within existing hierarchies rather than genuinely co-creating solutions across them.
Societal empowerment—perhaps the area where the dissonance is greatest—defined as the capacity of communities to influence institutions, laws, and social norms, can rely on AI in only two distinct ways. The first involves facilitating access to existing, legally established channels of participation (such as public consultations, petitions, and participatory budgeting) by helping citizens understand relevant texts and formulate their contributions; this approach requires no legislative changes. The second would involve creating new, binding channels of influence, which would necessitate institutional or legal reform—something AI cannot bring about on its own. Of these two options, only the first is truly within the scope of an AI product; it is precisely this role of expanding access that underpins Zuckerberg’s vision: AI as a tool to lower barriers for individuals, rather than a means to restructure institutions. However, applying the same standard of accountability used earlier regarding private data collection, expanding access to these channels fosters empowerment only if the institutions receiving citizen input are themselves subject to binding oversight; otherwise, AI merely funnels more data to collectors who remain just as unaccountable as the private companies criticized earlier. Zuckerberg’s claim that AI empowers citizens holds true, at best, only for this first, more limited path—and provided that this additional condition is met—rather than for the deeper sense of societal empowerment that entails institutional transformation. To date, this avenue remains largely unexplored on both fronts: citizens are generally unaware of these channels or deem them too complex for effective participation, and few of the relevant institutions offer citizens a verifiable way to ensure their contributions are actually reviewed and taken into account. Without these two conditions, AI-assisted citizen analysis lacks any credible path to reach—let alone influence—the actual decision-making process.
A challenge that remains unmet: building AI as infrastructure for collective intelligence
Zuckerberg is right that AI should not concentrate power in the hands of a few actors. But this stated ambition is difficult to reconcile with an infrastructure that is currently designed, owned, and governed by those same few actors. If the companies best positioned to build tools for shared governance and algorithmic transparency are also the ones whose commercial interest lies in maximizing proprietary control over data and models, the vision of democratization requires either a change in incentive structure or an external governance mechanism — neither of which the original vision specifies.
The future of AI should not be measured only by the intelligence of machines or the productivity gains they generate, but by whether the specific conditions of transparency, accountability, data control, and participatory governance described above are actually implemented — and independently verifiable. Until they are, AI expands what individuals can do alone; it does not yet expand what citizens can achieve together.
Marie-Georges Fayn