A new research paper submitted to arXiv on June 9, 2026, is sounding the alarm on a phenomenon that's quietly accelerating across society: humans increasingly consulting AI systems before, alongside, or instead of human experts, peers, and their own independent judgment. Titled "The Social Consequences of AI Delegation," the paper argues that while significant debate has focused on whether LLMs can substitute for human research participants, the more urgent question is whether people are using these systems as surrogates for their own deliberation—and under what conditions this delegation occurs.

The Core Argument: AI as Consequential Social Actor

The researchers, including Yamir Moreno, contend that large language models should be treated as consequential social actors in a functional sense—systems whose outputs actively shape human decisions, social norms, and collective dynamics. This framing moves beyond the academic debate about whether LLMs can mimic human responses. Instead, it positions AI systems as agents reshaping how people think, choose, and interact with expertise itself.

Spreading Across High-Stakes Domains

The paper identifies five critical domains where delegation is already visible: health, law, finance, education, and personal guidance. Whether it's someone asking an LLM about medical symptoms before seeing a doctor, seeking legal interpretations without consulting an attorney, or getting investment advice from a chatbot instead of a financial advisor, the pattern is consistent—people are treating AI outputs as legitimate substitutes for expert input.

Why This Demands Urgent Study

The researchers acknowledge that evidence for actual delegation remains uneven and context-dependent. But they argue this very uncertainty makes the phenomenon an urgent social-scientific object of study. If we wait for definitive proof of harm, the behavioral shifts may already be baked into how people approach decisions. The paper calls for a dedicated research programme to understand when delegation is benign, when it's risky, and what institutional or individual interventions might help.

Key Takeaways

  • LLMs are increasingly consulted across health, law, finance, education, and personal guidance before human experts
  • Researchers want frameworks treating AI systems as consequential social actors shaping collective dynamics
  • The phenomenon remains uneven but demands urgent study before patterns become entrenched

The Bottom Line

This isn't just an academic exercise—it's a warning that we've crossed a threshold where AI outputs are influencing real-world decisions at scale, often with zero friction and minimal scrutiny. If we don't start treating this delegation as a social science emergency now, we're essentially letting the training data of human judgment get overwritten by probability distributions.