(Whose defaults?) Is artificial intelligence reorienting archaeological methods?

URL SCAN: (Whose defaults?) Is artificial intelligence reorienting archaeological methods?
FIRST LINE: Computer Science > Computers and Society

The Dissection

This paper audits AI-mediated method selection under the cover of a question about disciplinary diversity. Its strongest finding is that LLMs generate a narrower, historically overrepresented method menu than the literature contains, especially without expert guidance. The models act as statistical priors, steering inquiry toward methods that are common, legible, and already encoded.

The paper stops at the epistemic surface. It measures recommendation convergence but does not follow the chain into labor substitution, funding, institutional incentives, or ownership of AI capital.

The Core Fallacy

It treats methodological diversity as the central casualty. Under DT, that is secondary damage. The primary rupture is the severing of human labor from economically necessary production. An archaeologist can retain access to 241 method clusters while losing the need to execute any of them.

The paper also risks treating recommendation as reorientation. Its evidence supports convergence pressure, not causal proof that archaeological practice has already changed. The authors acknowledge this limitation, which makes the analysis cautious but incomplete.

Hidden Assumptions

• Abstracts accurately represent methods actually used.
• The 25/241-category ontology does not conceal important distinctions.
• A post-2023 shift meaningfully reflects generative AI rather than unrelated changes.
• Two open-weight models and 28 problems represent future research workflows.
• Narrower recommendations produce narrower research.
• Expert prompting or governance can preserve human control at scale.
• The local LLM used to classify the literature is a neutral measurement instrument rather than another source of methodological compression.
• Preserving methodological pluralism preserves productive human participation. It does not. It may preserve only the appearance of agency after execution is automated.

Social Function

Classification: partial truth with a transition-management function and a mild ideological-anesthetic effect.

The paper identifies a real mechanism: LLMs reproduce high-frequency defaults and compress the search space before research begins. But by asking how archaeology can retain methodological diversity, it converts a power-and-ownership rupture into a governance problem. The comfortable remedies are better prompts, benchmarks, and safeguards. The decisive question remains untouched: who owns the systems, and how many researchers remain economically necessary afterward?

The Verdict

This is a useful smoke detector pointed at the wrong room. It documents convergence pressure in AI-generated method recommendations, but it does not establish that LLMs caused archaeological methods to converge, nor does it confront the deeper DT outcome. Methodological diversity can increase while human productive participation collapses. The discipline may retain a rich vocabulary of methods as researchers become curators, verifiers, and permission layers around automated production. The paper sees the narrowing menu. It does not yet see the kitchen being repossessed.