Essay argues AI’s social harms stem from capitalist incentives, not inherent tech limits

The essay distinguishes between two categories of AI-related problems: technological limitations and socio-political failures. Problems like hallucinations, sycophantic behavior, or poor contextual reasoning are framed as technical challenges that developers can and do address incrementally. However, the authors argue that other widely discussed harms—such as exploitative labor practices, environmental damage from data centers, or the enclosure of cultural content—are not inherent to AI’s capabilities but are instead driven by capitalist incentives that prioritize profit over public welfare.

Schneier and Sanders cite the contrast between U.S. and Chinese AI development paths to illustrate this divide. While U.S. labs emphasize frontier performance at high capital and energy cost, Chinese developers have pursued smaller, more efficient models distributed via commodity hardware. The authors attribute this divergence to differing incentive structures: U.S. firms optimize for investor returns, while Chinese developers operate under state-backed mandates that prioritize accessibility and national influence.

The authors highlight Switzerland’s Apertus model as a counterexample to profit-driven AI development. Developed collaboratively by public institutions using licensed data and renewable hydropower, Apertus is explicitly designed as a public good rather than a private asset. Its existence demonstrates that alternative development pathways are feasible when incentives align with societal benefit rather than shareholder value.

Policy proposals such as moratoria on AI research or federal screening of frontier models are criticized for misdiagnosing the problem. The authors argue these measures target technological risks while ignoring the structural conditions—energy policy, antitrust enforcement, corporate governance—that shape how AI is deployed. Instead, they advocate for reforms such as taxing corporate energy use, redistributing AI profits, and enforcing antitrust laws to realign incentives toward public benefit.

The essay concludes that AI’s most pressing challenges are not about what the technology can or cannot do, but about who controls it and for whose benefit it is deployed. Separating technological constraints from capitalist incentives, they argue, is essential for designing effective governance that steers AI toward equitable outcomes.