The scholarly AI gradient is a typology of scholarly knowledge work introduced by Jason Miklian (2026) in the working paper "What is an Academic Article For?". It maps seven distinct activities that constitute the research process onto a gradient of current AI capability: information retrieval and synthesis, conceptual development, interpretive judgment, normative reasoning, writing as thinking, collaborative reasoning, and autonomous research agency. At one end, AI capability is already substantial and the scholar's labor shifts from finding to judging. At the other, the scholar's cognitive engagement constitutes the contribution itself, and no capability gain changes that relationship.
Information retrieval and synthesis sits at the high-capability end: AI can process thousands of articles and generate structured summaries, while the scholar's essential contribution becomes selection, source evaluation, and forensic verification of citations. Conceptual development and collaborative reasoning occupy the emerging middle, where AI proposes and the scholar evaluates. Autonomous research agency, in which agentic systems execute multi-step research workflows, shifts the human role from direction to auditing.
Interpretive judgment, normative reasoning, and writing as thinking anchor the other end. Interpretation requires tacit, situated knowledge that grounds claims in specific traditions and fieldwork. Normative reasoning requires holding commitments as a moral agent. And composing text is itself a cognitive process through which ideas are discovered and refined; the writer is changed by writing in ways that prompting an AI does not replicate.
Treating all scholarly work as a single activity called research produces policies that are too restrictive for synthesis work or too permissive for interpretive work. The gradient supplies the missing vocabulary. For positivist research, the knowledge claim is separable from the process that produced it, so AI involvement can be evaluated on the substance of the finding. For interpretivist scholarship, how the knowledge was produced is constitutive of what it means, so process belongs inside the evaluation.
The framework grounds three governance principles: accountable knowledge evaluated on substance, with the scholar able to defend every claim under sustained scrutiny; deliberate institutional protection of intellectual formation for junior scholars; and equitable, adaptive governance with built-in revision mechanisms.
Two empirical tests anchor the framework. A corpus analysis of 2,750 open-access articles in 15 social science journals found AI-associated language rose 151 percent from 2022 to 2025, with the strongest increases in synthesis-heavy journals and the weakest in journals built on original theoretical argument. A governance audit of 70 top-ranked universities across 16 countries found that none differentiates AI policy by discipline or epistemological tradition, none addresses AI use in peer review, and none includes a revision or sunset clause.
Miklian, Jason. "What is an Academic Article For? The Production of Scholarly Work and its Meaning in an Agentic AI World." Working Paper, 2026.
Miklian, Jason. "What is an Academic Article For? The Production of Scholarly Work and its Meaning in an Agentic AI World." Working Paper, 2026.
The scholarly AI gradient is a typology introduced by Jason Miklian (2026) that breaks the research process into seven activities and orders them by current AI capability, from information retrieval and synthesis, where AI is already strong, to interpretive judgment, normative reasoning, and writing as thinking, where the scholar's own cognitive engagement constitutes the scholarly contribution.
The framework proposes evaluating scholarship on substance rather than process: a scholar who uses AI extensively and can defend every claim under sustained scrutiny has met the standard, while one who cannot reconstruct the reasoning has not, regardless of AI involvement. Disclosure should describe what AI contributed, such as synthesis, collaborative reasoning, or drafting, rather than a binary yes or no.
Yes, measurably. Miklian (2026) found the density of AI-associated vocabulary in 15 social science journals rose 151 percent between 2022 and 2025, from 8.2 to 20.5 markers per 10,000 words, with the sharpest acceleration after ChatGPT's mass adoption. The increase was strongest in synthesis-heavy journals and weakest in journals centered on original theoretical argument.
Mostly not. An audit of 70 top-ranked universities across 16 countries found no institution that differentiates AI policy by discipline or epistemological tradition, none that addresses AI use in peer review, and none with a sunset or revision clause, even though AI capabilities have shifted dramatically every 12 to 18 months since 2022.