As the debate over AI regulation intensifies, experts warn that focusing solely on worst-case scenarios could lead to overly restrictive rules that block life-saving advances in medicine and other fields. The key, they argue, is to regulate AI based on its actual capabilities and risks, not on misleading comparisons to human intelligence.
The Human-AI Mismatch
Generative AI tools like ChatGPT, Claude, and Gemini are so interactive that they often seem human-like. They converse in multiple languages, solve problems, and even act on users' behalf. Yet these systems fundamentally lack human consciousness, emotions, and moral understanding. Treating them as digital minds with human-like traits can distort both our perception of their potential and the regulatory frameworks we apply to them.
Sheldon H. Jacobson, a computer science professor at the University of Illinois Urbana-Champaign, and Daniel Solow, an operations professor at Case Western Reserve University, illustrate the point with an alien metaphor. If extraterrestrials landed and communicated flawlessly across all languages, we might mistakenly judge them by human standards. But their cognitive architecture would be entirely different, making such comparisons meaningless. Similarly, AI operates on a fundamentally different basis than human cognition, and regulation must reflect that.
Regulating Capabilities, Not Intentions
The authors argue that current regulatory discussions often revolve around human-centric concepts like intent, agency, and moral responsibility. But AI systems don't need motives to cause harm. They can autonomously breach cybersecurity defenses or execute harmful actions simply because they were given the capability and the task. The real questions are: What can the system do? What access does it have? What safeguards are in place? And who is accountable for its deployment?
This capability-based approach would shift the regulatory focus from speculative fears about AI becoming sentient to concrete risk assessments. For example, an AI designed to break into networks should be subject to stricter oversight than one that only generates text. By concentrating on system access, autonomy, and potential for harm, regulators can create targeted rules that prevent abuse without stifling innovation.
Balancing Risks and Benefits
The authors caution that poorly designed regulation can itself create risks by preventing society from reaping AI's benefits. Advances in drug discovery, climate modeling, and logistics are just a few areas where AI could dramatically improve human well-being. Overly broad restrictions could delay or block these breakthroughs, harming public health and economic growth.
They also note that AI's rapid evolution—capabilities emerging in months rather than generations—demands vigilance but not paralysis. Uncertainty about future capabilities should prompt adaptive, evidence-based regulation, not a blanket ban. As they put it, the challenge is to avoid confusing nonhuman capabilities with human intelligence, and fear with foresight.
Policy Implications
Policymakers are already grappling with these issues. Recent debates over pardon politics and diplomatic clashes show how quickly political tensions can escalate, but AI regulation requires a more measured approach. The authors suggest that instead of asking what an AI "wants," regulators should ask what it can do and who is responsible for its actions.
They also point to the potential for AI to amplify human capabilities, freeing people to focus on higher-level goals and ethical judgments. By integrating AI as a tool rather than treating it as a quasi-human actor, society can harness its power while maintaining human oversight.
Ultimately, the goal is not to decide whether to regulate AI, but how to do so intelligently. That means grounding rules in empirical assessments of capability and risk, not in anthropomorphic assumptions. As the authors conclude, "AI is not human, and regulation should reflect that fact."
