Don't Blame the Fire. Look at Who Struck the Match

Fear of artificial intelligence, concentrated power, and the case for public infrastructure

Abstract

Public hostility toward artificial intelligence is often treated as a spontaneous response to the technology and its risks. The reality is more complicated. Concerns about job losses, surveillance, disinformation, and economic concentration are well founded, but they can also be amplified and exploited by platforms, corporations, and political actors. Fear creates public support for expensive regulatory regimes that only the largest companies can afford, strengthening the very firms whose power deserves scrutiny. The central political question is who owns the models, controls the computing infrastructure, and sets the terms under which these systems may be used. A democratic response should combine effective safeguards with open models, publicly funded computing resources, independent research, and shared forms of ownership.

1. The Machine as Scapegoat

Artificial intelligence is already among the most consequential technologies of our time. Like the printing press, electricity, the internet, and industrial automation, it extends abilities that were once scarce or difficult to access. It can process large bodies of knowledge, translate languages, write software, analyse medical images, accelerate scientific research, and make specialist expertise available to more people.

Public debate increasingly describes AI as if it were an independent moral actor. "AI is taking our jobs." "AI is destroying art." "AI lies." This language is convenient because it pushes human decisions into the background. Executives choose to replace workers in pursuit of higher margins. Platforms decide to train models on material gathered without meaningful consent. Governments deploy automated surveillance. Employers introduce new tools without sharing the resulting gains with the people whose work made those gains possible.

Fire can warm a home or burn it down. Blaming the flame makes it easier to ignore the person holding the match.

2. How Fear Is Produced

The anxiety surrounding AI is grounded in real dangers. Systems that affect employment, credit, health care, policing, or access to public services require firm rules and clear lines of responsibility. Models can reproduce discrimination, disclose private information, and generate convincing falsehoods. None of this should be dismissed.

At the same time, the information economy rewards the loudest version of every concern. Platforms promote material that triggers anger, fear, and conflict because those emotions keep people watching, commenting, and sharing. A legitimate debate can quickly harden into a moral panic in which every use of AI is treated as equally dangerous and every technical distinction sounds like an excuse.

No central conspiracy is needed. Media outlets attract attention with catastrophic scenarios. Influencers build an audience around a conflict between human creativity and machines. Politicians promise protection. The largest technology companies occupy the most advantageous position of all: they can present themselves as both the creators of a powerful threat and the only institutions qualified to contain it.

Some scholars use the term epistemic capture to describe the way economic power shapes the language through which a society understands an issue. The concept is useful here. When leading AI companies are allowed to define what "safe AI" means, their internal procedures can become the template for future law. Standards built around their resources, staff, and technical architecture may then determine who is allowed to enter the market.

3. From Public Anxiety to Regulatory Capture

Regulation can protect the public, but poor regulation often protects established firms. Upfront licensing, expensive audits, extensive reporting duties, and certification systems designed around frontier laboratories are manageable expenses for billion-dollar companies. For a university department, a small business, an independent developer, or an open-source community, the same requirements can make serious work impossible.

Public fear gives this arrangement a persuasive political story. If every freely available model is described as an existential threat, restricting advanced development to a handful of "certified" companies begins to sound responsible. The likely result is a state-approved oligopoly whose members face less competition and gain more influence over the rules that govern them. Safety may improve in some areas, but the market becomes harder to challenge and public oversight weaker in practice.

There is a particular irony in seeing companies that grew during a lightly regulated period support rules that would have prevented smaller predecessors from following the same path. Once compliance becomes a fixed cost of entry, safety also functions as a competitive moat. This conflict of interest warrants closer scrutiny from lawmakers and a wider range of expert testimony.

Good rules should focus on measurable risks, specific uses, and demonstrable harm. A medical diagnostic system deserves a different standard from a locally run writing assistant. A model used to allocate welfare benefits calls for stronger scrutiny than one used for private experimentation. Regulation that ignores these differences will tend to reward scale rather than responsibility.

4. The Prospect of Private Technocracy

Allowing private companies to dominate AI would place a growing share of society's intellectual infrastructure under the authority of unelected boards. A small number of firms could influence which sources of knowledge remain accessible, which opinions their systems will express, which professions are targeted for automation, and which governments can use the most capable models.

This level of control would reach further than an ordinary commercial monopoly. AI systems are becoming intermediaries in education, research, public administration, cultural production, and everyday work. When access depends on private contracts, subscriptions, and revocable APIs, essential intellectual tools can disappear after a policy change, a price increase, or a shift in corporate strategy. The public would be left to adapt to decisions it had no role in making.

Fear also gives politicians an opportunity to appear tough while leaving corporate power largely untouched. A law may impose dramatic restrictions in public while preserving exemptions, liability protections, or technical requirements that favour incumbent firms. The same political coalition can then oppose public alternatives as inefficient or dangerous. The result is a familiar arrangement: risk remains social, control stays private, and the public pays for both.

5. Public Computing and Open Knowledge

A public role in AI does not require a government ministry to own every model or dictate what a system may say. Public ownership, political control, and scientific independence are separate questions, and a durable institutional design should treat them that way.

Governments can fund computing facilities, transparent foundation models, and lawfully assembled datasets for use by universities, hospitals, public agencies, businesses, and citizens. Model weights, source code, evaluation methods, and technical documentation should be available for inspection whenever their release does not create a specific and credible danger. Public investment can reduce dependence on a few vendors while giving researchers the ability to test claims that would otherwise remain hidden behind corporate secrecy.

Such infrastructure needs protection from both political interference and private capture. An independent governing body could operate it under parliamentary and civil oversight, with published decisions, reproducible audits, strong privacy rules, and transparent criteria for allocating computing resources. Access should be broad enough to support small research teams and public-interest projects, not reserved for institutions that already possess substantial funding. Legal responsibility should follow harmful conduct and high-risk deployment rather than the mere creation or possession of a general-purpose technology.

Open-source development belongs within this public ecosystem. Openness allows outsiders to verify security claims, study bias, adapt tools to minority languages, and build services that do not depend on the permission of a single company. Some capabilities may require controlled access, particularly where there is evidence of a concrete and severe risk. Those exceptions should be narrow, reviewable, and supported by evidence rather than vague appeals to danger.

Public computing capacity would not eliminate private AI, nor should it. Its purpose would be to give society an alternative: a shared technical foundation that cannot be withdrawn whenever it conflicts with a company's commercial interests. Treating a portion of computing and model development as public infrastructure would keep the ability to process knowledge from becoming the exclusive property of a new technological aristocracy.

Conclusion

Artificial intelligence deserves neither worship nor reflexive hostility. It needs informed public oversight and institutions capable of distributing its benefits.

The struggle surrounding AI concerns ownership, access, and accountability. A frightened public may accept a future in which a few supposedly trustworthy organisations control the technology on everyone else's behalf. A more confident democratic society can insist that systems built from collective knowledge remain accessible, open to scrutiny, and directed toward public purposes.

The fire is already burning. We should pay close attention to who holds the matches, who owns the hearth, and who is left outside in the cold.