As AI automates junior research tasks, the industry faces a critical question: are we gaining efficiency today at the cost of the expertise we need tomorrow?
The Cognitive Commons
AI may be making research more efficient. But could it also make the research profession weaker?
A new paper by Nolan Lovett, published in Human Resource Development Review, gives a name to something many of us in insights have been sensing for a while. He calls it the Cognitive Commons.
The argument is that professions can lose more than jobs to AI. They can lose the process through which expertise is passed from one generation to the next—even while every firm’s decision to automate makes perfect sense.
Lovett borrows from Garrett Hardin’s Tragedy of the Commons: each herder gains the full benefit of adding another animal to shared pasture, while the cost of overgrazing is spread across everyone. The same can happen with professional expertise.
Every agency that hires fewer junior analysts because AI can code verbatims and draft toplines captures the immediate savings. Ten years later, however, the industry could have fewer experienced researchers capable of judging whether the AI got it right.
Two Types of Expertise
Lovett discusses two different types of expertise:
Internalized Mastery
Internalized Mastery develops through sustained practice. It is what you acquire while trying to understand why the NPS scores contradict the open ends, or while moderating your first hostile focus group and learning to read the room rather than simply follow the guide.
Distributed Mastery
Distributed Mastery is the ability to work effectively with AI systems. You know what to delegate, what to question and where the output needs to be checked.
Distributed Mastery depends on Internalized Mastery. Lovett calls this the Validation Tether. You can catch an AI output that is fluent but wrong only when you already know enough to recognize the error.
Anyone can spot a malformed table. Recognizing that a synthetic sample overrepresents early adopters—or that a driver analysis is statistically plausible but commercially misleading—requires experience.
Surface-level checking is easy. Substantive validation is scarce, and junior-level work has traditionally helped build this capability.
Much of our entry-level work—including verbatim coding, questionnaire scripting and first-pass reporting—resembles the kind of codified work that AI can absorb.
Why Market Research May Be Particularly Vulnerable
Applying Lovett’s five factors to market research suggests that the industry may be particularly exposed.
Task substitutability: How easily AI can perform the tasks through which beginners traditionally learn. Verbatim coding, desk research and first-draft reporting are highly substitutable.
Regulatory intensity: Whether licences, accreditation or formal supervision limit who may perform the work. Market research has few such barriers.
Safety criticality: Whether mistakes have immediate and serious consequences. Medicine and structural engineering face strong pressure to preserve human oversight. Research errors are usually less visible and emerge more slowly.
Professional-association strength: Whether professional bodies can enforce training requirements and standards. ESOMAR, MRS and the Insights Association provide guidance, but they cannot regulate entry or practice like a medical board.
Work modularization: How easily a role can be divided into separate tasks that AI can handle. Junior research work is particularly modular—from scripting surveys and coding responses to producing initial charts.
What Research Agencies Can Do Now
There is no easy solution to the free-rider problem. An agency that invests in junior development will compete against agencies that do not and may appear less efficient on paper. However, some practical steps are still worth taking.
1. Protect Independent Practice
Give junior researchers selected analyses to complete without AI before comparing their work with an AI-generated version. This preserves the repetition through which judgment develops.
2. Use AI as a Feedback Mechanism
Ask junior researchers to investigate every disagreement between their analysis and the AI output. Later, give them AI-generated work containing subtle methodological or interpretive errors and ask them to diagnose what went wrong.
3. Test Substantive Validation
Assess whether researchers can identify biased samples, weak causal claims and conclusions unsupported by the raw data. Prompt-writing ability is a poor substitute for research competence.
4. Keep Junior Researchers Involved in the Whole Project
When their work is divided into isolated checking tasks, analysts may never learn how the business question, research design and final interpretation connect. Let them follow projects from the first briefing through to the client conversation.
5. Consider New Ways to Assess Independent Research Judgment
ESOMAR, MRS or the Insights Association could potentially assess whether researchers can design, interpret and challenge research without relying on an AI system’s conclusions.
This would give clients something concrete to request and give agencies a reason to continue investing in foundational skills.
None of this removes the commercial incentive to automate junior work. Every time an agency removes a task, however, it should ask what capability that task used to build—and where the next generation will acquire it instead.
The industry should avoid becoming “Buy Now, Pay Later”—capturing the efficiency gains of automation today while deferring the cost of weakened professional expertise to the future.
About the Paper
Nolan Lovett, “The Tragedy of the Cognitive Commons: How AI Could Disrupt the Regeneration of Professional Expertise,” Human Resource Development Review, 2026.
