Responsible AI or Critical AI: Two Competing Philosophies Now Define the Governance Debate

When policymakers, executives, and activists argue about governing artificial intelligence, they are usually arguing between two distinct philosophies without naming them. The first, "Responsible AI," asks companies to develop AI ethically through voluntary principles and soft law. The second, "Critical AI," demands enforceable, top-down regulation that subordinates technological progress to social and ecological justice. Understanding this distinction is the clearest way to make sense of every AI policy proposal now in circulation.

The Growing Divide in AI Governance

The divide is laid out most sharply by Paul Schütze and Benedetta Brevini, whose article in AI & Society argues that the current moment requires moving "from responsible to critical AI governance." Their central charge is that voluntary frameworks risk what they call "the appeal of placebo change" — the appearance of action without binding obligation. In their analysis, AI governance has to stop treating innovation leadership as the goal and start putting "human lives and societal justice at its center."

This is not a marginal position. The United Nations' report Governing AI for Humanity reaches a complementary conclusion from a different angle: AI's pace and reach are straining the institutions meant to hold anyone answerable for its harms. The result is a governance debate with a genuine fork in the road, not a settled consensus with cosmetic disagreements.

Why AI Governance is Urgent Now

The urgency is structural, and the UN report identifies it precisely. A societal risk is emerging, the report warns, "that ever-fewer individuals end up being held accountable for harms arising from their decisions to automate processes using AI, even as increasingly powerful systems enter the world." Accountability diffuses as automation deepens; responsibility for a harmful decision can vanish into a chain of algorithms, procurement choices, and delegated judgments.

The report draws two consequences from this. First, it "demands agile governance to ensure that accountability mechanisms keep pace with accelerating AI" — a direct acknowledgment that legal systems built for slower technologies are being outrun. Second, it observes that "if the pace of AI development and deployment challenges existing institutions, so does the breadth": AI is not confined to one sector whose regulators can specialize, but cuts across labor, health, security, and information ecosystems simultaneously. Speed and breadth together are what make the governance question immediate rather than academic.

Responsible AI: Industry-Led and Voluntary Principles

Responsible AI, as it currently operates, aims to establish and implement principles, policies, and practices that guide the "ethical, transparent, accountable, safe, and socially and environmentally responsible development, deployment, and use of AI technologies" — language Schütze and Brevini trace to frameworks such as the European Commission's 2019 guidelines. The emphasis is deliberate: foster innovation while integrating ethics into the development process itself.

The concrete ethical content of such frameworks is well established, and two declarations illustrate the shared vocabulary. UNI Global Union's "Top 10 Principles for Ethical Artificial Intelligence" calls for transparency, an "ethical black box" for AI systems, making AI "serve people and planet," a human-in-command approach, unbiased AI, shared benefits, a just transition that protects fundamental rights, global governance mechanisms, a ban on attributing responsibility to robots, and a ban on an AI arms race. The Montréal Declaration for Responsible AI, in the principles covered by the available record, lists well-being, respect for autonomy, protection of privacy and intimacy, and solidarity.

What unites these documents is that they define what good AI should look like. What they cannot do, critics note, is compel anyone to build it that way — and that gap is the crux of the debate.

Critical AI: Prioritizing Public Interest Through Regulation

The critical AI position accepts much of the ethical vocabulary but rejects voluntariness as the enforcement mechanism. Schütze and Brevini argue that AI governance "needs to take a stance against current big tech power dynamics" and that this "has to happen top–down on the policy level." While corporate AI governance frameworks have value, they write, "this is not enough."

More pointedly, the critical approach is willing to say no to entire classes of technology. It calls for "strong regulation of companies and the enforcement of principles oriented toward the common good and sustainability," and — citing the Bits & Bäume movement's demands — for actively prohibiting the development of AI technologies that may not align with genuine sustainability goals. The authors give a concrete example: the use of large language model products such as ChatGPT, Claude, or Gemini may contradict sustainability goals and societal well-being, so how their use should be regulated more strictly becomes, in their words, a crucial question. That willingness to subordinate innovation is precisely what separates critical governance from its responsible counterpart.

Comparing Responsible AI and Critical AI Governance

The table below places the two philosophies side by side across four dimensions, drawing on the characterizations in the AI & Society analysis. The comparison reveals the trade-off every policymaker now faces.

Responsible AI versus Critical AI governance, as characterized in recent academic analysis
Dimension	Responsible AI Governance	Critical AI Governance
Core Goal	Guide the ethical, transparent, accountable, safe, and socially and environmentally responsible development, deployment, and use of AI.	Prioritize social and ecological justice and the public interest, challenging dominant tech ideologies.
Primary Mechanism	Voluntary principles, policies, and practices, largely through soft law.	Strong, top-down regulation and hard law with enforcement.
Stance on Innovation	Foster innovation while integrating ethical considerations into development.	Subordinate technological progress to social justice and democratic control — including prohibiting technologies that conflict with sustainability goals.
View of Corporate Power	Positions companies as key actors defining and implementing ethical guidelines.	Seeks to contain big tech power dynamics and ensure AI serves broader public interests.

Read across the rows, the pattern is consistent: Responsible AI asks how to develop AI well, while critical AI asks whether some AI should be developed or deployed at all. Neither is a strawman. The responsible model reflects genuine corporate ethics programs and a fear that heavy-handed rules will freeze beneficial applications; the critical model reflects the observation, in Schütze and Brevini's phrase, that soft law offers only "placebo change" — and that self-regulation rarely restrains the entity doing the regulating.

Navigating the Governance Crossroads

The practical question for any reader evaluating a new AI policy proposal is not "is this ethical?" but "which philosophy does this embody, and does its mechanism match its ambition?" A corporate transparency pledge, however sincere, belongs to the responsible tradition; a statutory ban on a class of applications, backed by enforcement, belongs to the critical one. Proposals that announce critical-sounding goals while relying on voluntary mechanisms sit in the contested middle — exactly where Schütze and Brevini and the UN report both locate the current accountability gap.

These are not symmetric abstractions. The UN report's warning about ever-fewer accountable individuals gives the choice a deadline: each deployment that outpaces institutional capacity widens the diffusion of responsibility. Whatever one concludes about innovation costs, the two analyses agree that governance which cannot assign responsibility for harm is not governing.

Navigating the Future of AI Governance

Policymakers and stakeholders must critically assess whether voluntary "Responsible AI" frameworks are sufficient or if more stringent "Critical AI" regulations are necessary to address societal risks and ensure justice. The observable signal will be the nature of new AI policy proposals and enacted legislation, specifically whether they lean towards voluntary guidelines or mandatory, enforceable regulations, and their explicit prioritization of innovation versus social justice. Watch that signal, and the direction of the world's largest experiment in governing a general-purpose technology will be legible well before its consequences arrive.