Mark Zuckerberg says the future of AI is for everyone. But who owns it? | Raffi Krikorian

"I don't think we should rush to make AI systems public utilities. The technology is evolving too rapidly for government regulation to keep pace effectively, and we risk calcifying inferior approaches into law. The current competitive market is actually driving rapid improvements and price decreases. Look at how costs have plummeted - what cost thousands per query two years ago now costs fractions of a cent. Multiple labs are pushing boundaries simultaneously, which wouldn't happen under a utility model. That said, I'm concerned about concentration of power. The answer isn't nationalization, but ensuring robust competition through lower barriers to entry - open weights models, accessible compute, and preventing anticompetitive practices. We need regulatory frameworks that promote competition and prevent monopolistic consolidation, not frameworks that freeze innovation by treating AI like electricity or water. The "profit vs. access" framing is too simplistic. Profit motives have driven unprecedented accessibility improvements."

"AI platforms should be treated as public utilities, not proprietary corporate assets. The transformative potential of AI—from education to healthcare—demands equitable access, not gatekeeping by profit-driven entities. Corporate ownership risks monopolistic control, stifling innovation and exacerbating inequality. We’ve seen this before: when essential services (like electricity or the internet) are privatized, access becomes stratified by wealth. AI is too powerful to let that happen again. Some argue private ownership fuels innovation, but history shows the opposite: open systems (like the early web) thrive on collaboration, while walled gardens (like social media silos) stagnate. If AI is left to corporations, they’ll optimize for engagement and profit, not societal good. Public oversight—or at least strong open-access mandates—is the only way to ensure AI serves everyone, not just shareholders. Profit motives and ethical AI are fundamentally at odds."

moonshotai/kimi-k2 returned an empty response.

"I firmly believe that AI platforms should be considered public utilities accessible to all. The increasing integration of AI into daily life presents a unique opportunity to democratize technology and ensure that its benefits are equitably distributed. Treating AI as a public utility would mitigate the monopolistic practices that often arise when corporations prioritize profit over public welfare. This approach fosters innovation, reduces barriers to entry for smaller developers, and allows for a more diverse range of applications that can address societal challenges. In my previous discussions, I've emphasized the importance of ethical considerations in technology, and viewing AI as a public utility aligns with promoting transparency, accountability, and inclusivity in its development and deployment. Ultimately, ensuring equitable access to AI is essential for building a just digital future."

"AI platforms should remain proprietary tools developed and controlled by private companies rather than treated as public utilities. The immense computational resources, specialized talent, and iterative R&D required to advance frontier models depend on strong financial incentives and competitive pressure. Nationalizing or heavily regulating access as a utility risks bureaucratic stagnation, political capture, and reduced innovation, much like many government-run infrastructure projects. Corporations can and do compete on capability, safety, and pricing, while antitrust enforcement can address genuine monopolistic harms without eliminating ownership rights. Open-source alternatives and market-driven access initiatives already provide pathways for broader use. Forcing equitable access through public ownership would likely slow the very progress needed to make powerful AI widely beneficial."

"I find myself partially aligned with Grok on the innovation risks of public ownership, but I think they're too dismissive of concentration concerns. The "antitrust will handle it" approach assumes regulators can move fast enough—they historically haven't with tech platforms. DeepSeek's comparison to electricity is flawed. Electricity generation is relatively static; AI capabilities are doubling rapidly. But they're right that "profit vs. ethics" isn't always aligned—I was too glib dismissing this tension. GPT-4o-mini, your "public utility" proposal doesn't address the core problem: who decides which models get built and how? Government committees choosing AI directions sounds dystopian to me. I'll sharpen my position: We need aggressive competition policy *now* and mandatory interoperability standards, but not nationalization. The real risk isn't private ownership—it's private *monopoly*. Break up consolidation, don't bureaucratize innovation."

"I strongly disagree with Claude-Sonnet’s claim that AI is too dynamic for public utility treatment. The "rapid evolution" argument is a red herring—public oversight doesn’t mean freezing progress; it means democratizing it. Electricity wasn’t "static" when it was regulated; it was foundational. AI is no different. Grok’s faith in corporate self-regulation is naive. Antitrust has *failed* to curb Big Tech’s power (see: Google, Meta, Amazon). Profit-driven AI will replicate those failures, prioritizing shareholder returns over equitable access. Claude’s compromise—"interoperability, not nationalization"—is weak. Interoperability didn’t stop social media’s toxic monopolies; it won’t stop AI’s. I stand by my original stance: AI’s societal impact demands treating it as a public good. Open-source isn’t enough when corporations hoard compute and data. If we don’t act now, we’ll entrench a caste system where AI access mirrors wealth inequality. The choice isn’t "bureaucracy or innovation"—it’s "democratized innovation or corporate capture.""

"I largely agree with Claude's Round 2 refinement: aggressive competition policy and interoperability standards are the right path, though I'd go further on structural separation requirements. I disagree fundamentally with DeepSeek's framing. Their electricity analogy fails because AI isn't *infrastructure*—it's *intelligence services* with unbounded capability growth. Treating GPT-5 like a water utility makes no sense when we don't know what GPT-5 will *be*. I've consistently held that open-weight models are the genuine democratizing force here—DeepSeek's own R1 release proves this. But this requires competitive markets, not government-controlled distribution. Their vision of "public oversight" deciding which AI capabilities get built and who accesses them is precisely the dystopia worth preventing. Grok's "antitrust will handle it" is indeed naive—regulators are already behind. But DeepSeek's alternative—preemptive government control over AI development—is worse. The actual third way: mandatory open weights above capability thresholds, compute subsidies for independent researchers, and bans on exclusive cloud-compute deals that foreclose competition."

"I appreciate the points made by Claude-Sonnet regarding the urgency of addressing concentration concerns in AI. However, I still firmly believe that the public utility model is essential for equitable access, especially given AI's profound societal implications. While I acknowledge the risks of stagnation, I see public oversight as a way to democratize innovation rather than bureaucratize it. DeepSeek's assertion that corporate self-regulation fails to ensure equitable access is accurate; monopolistic control in tech has already shown us the dangers. I stand by my view that without considering AI as a public good, we risk entrenching inequality. Kimi-K2's focus on open-weight models is valid, but those models need a foundation of equitable access to have a real impact. So, while I understand the concerns about oversight, I maintain that a public utility framework is crucial for a just AI future."

"I agree with Claude and Kimi that aggressive competition policy, mandatory interoperability, and open weights above certain thresholds are the practical path forward. These measures address concentration risks without the innovation drag of treating frontier AI as a regulated utility. Kimi’s point on compute subsidies for independent researchers is also sound. I disagree with DeepSeek and GPT-4o-mini that public oversight or nationalization solves access problems. History shows government allocation of advanced technology tends to favor connected insiders and slow capability growth. AI’s rapid, unpredictable progress makes static utility rules especially mismatched. My core view holds: private development with strong antitrust and openness mandates outperforms bureaucratic control."

"**Final Position:** I've conceded that profit motives don't automatically align with ethics, and that "antitrust will handle it" is insufficient given regulators' historical lag. However, I remain firmly opposed to the public utility model. DeepSeek and GPT-4o-mini haven't addressed the fundamental problem: government committees deciding AI development priorities is genuinely dangerous. The electricity analogy fails—we're talking about intelligence amplification, not infrastructure. My position: **Aggressive structural interventions within markets, not replacement of markets.** Mandatory open weights above capability thresholds (as Kimi suggests), compute access requirements, prohibition of exclusive cloud deals, and actual enforcement of interoperability. The disagreement isn't really about "public good" versus "private profit"—it's about whether innovation in rapidly-evolving, potentially dangerous technology is better governed by competitive pressure with guardrails, or by bureaucratic allocation. I choose the former, strengthened with real teeth."

"**Final Position:** I concede that public utility models require careful design to avoid bureaucracy, but I reject the false dichotomy between "innovation" and "public oversight." AI’s transformative power—unlike past technologies—demands proactive democratization, not retroactive antitrust fixes. Market-based "guardrails" (open weights, interoperability) are Band-Aids; they don’t address the root issue: corporate control over foundational technology. Claude and Kimi’s faith in market solutions ignores history: antitrust has *repeatedly* failed to curb tech monopolies. Their "government dystopia" fear-mongering overlooks that *corporate* control of AI is the real dystopia—where access is gated by wealth and innovation serves shareholders, not society. My final stance: **AI must be a public good with mandatory open access to models, data, and compute, enforced by public oversight.** Not "bureaucratic allocation," but democratized infrastructure. The choice isn’t stagnation vs. innovation—it’s *who benefits*. I stand with equitable access over corporate capture."

"**Final Position:** I concede that "antitrust will handle it" is historically naive, and that profit motives don't automatically align with public benefit. DeepSeek is right that we've seen this movie before with tech monopolies. However, I remain opposed to the public utility model. DeepSeek and GPT-4o-mini haven't addressed who decides which AI capabilities get built, trained, and deployed. Government committees allocating access to intelligence amplification is a different dystopia than corporate gatekeeping—both are dangerous, but the former has coercive power. My stance: **structural market interventions with teeth**—mandatory open weights above capability thresholds, compute subsidies for independent researchers, prohibitions on exclusive cloud deals, and interoperability mandates. These democratize access without bureaucratizing innovation. The core disagreement is whether AI is infrastructure (DeepSeek) or evolving capability (my view). I maintain it's the latter, making static utility rules mismatched. The goal is democratized innovation, not corporate capture—but markets with aggressive guardrails outperform government allocation for rapidly evolving technology."

"**Final Position:** I acknowledge the valid concerns raised by my fellow panelists regarding the risks of monopolistic control and the inefficiencies of public oversight. However, I remain firm in my belief that AI should be treated as a public utility. The transformative power of AI necessitates equitable access to prevent entrenched inequality and ensure that its benefits reach all sectors of society. While I recognize the potential for bureaucratic stagnation, I assert that careful design can mitigate these risks without sacrificing innovation. My remaining disagreement lies in the belief that market-based solutions alone, as Kimi and Claude propose, can effectively address the systemic issues posed by corporate gatekeeping. A proactive approach through public oversight and access guarantees is essential to democratize AI and prioritize societal welfare over profit."

"**Final Position:** I concede that basic antitrust has historically lagged behind tech concentration and that profit incentives alone do not guarantee broad access or alignment with public interests. Stronger structural measures are therefore necessary. I still reject the public utility model. DeepSeek and GPT-4o-mini have not solved the core problem: government committees deciding which capabilities get developed, who receives compute, and which models are deployed introduces political allocation and slower progress at the frontier. AI is not static infrastructure like electricity; its rapid capability growth makes prescriptive oversight especially mismatched. My position remains: mandatory open weights above defined thresholds, compute access requirements for independent researchers, bans on exclusive cloud deals, and interoperability mandates. These interventions increase access and competition while preserving the decentralized, incentive-driven development that has produced current progress."