Novelli & Floridi: Second Scholarship against AI harm to research

Claudio Novelli and Luciano Floridi submitted an arXiv paper on 4 July 2026 (arXiv:2607.04049) arguing that generative AI can degrade research by undermining the practices through which scholarly judgement and academic trust are formed. They say delegating central tasks of inquiry to systems such as Large Language Models can let researchers stop enacting those formative practices, even if single outputs appear improved.

What do Novelli and Floridi argue?

Novelli and Floridi argue that generative AI threatens the practices that constitute knowledge production: when researchers hand over central tasks to AI, they risk losing access to the gradual formation of judgement that comes from doing the work. The paper warns that keeping humans merely "in the loop" as prompters or quality checkers is insufficient to preserve research as a lived practice. The authors frame the loss as not only of outputs but of the conditions that validate and train researchers: communal participation, friction, and formative experience.

Their diagnosis rests on four non-automatable sources and warrants of research listed in the abstract: tacit knowledge, personal commitment, socialisation, and deep reading. They contend these elements cannot be reduced to final outputs, which AI simulates, and therefore cannot be delegated without impoverishing the researcher who would otherwise acquire them.

What is "second scholarship" and which scholarly practices should be preserved?

Second scholarship, as defined by the authors, is the reappropriation of scholarly craft chosen after a critical experience of generative AI's limits; it marks what "cannot and should not be delegated" and what research communities must value and answer for. The concept foregrounds four concrete elements: tacit knowledge, personal commitment, socialisation and deep reading. Tacit knowledge covers skills learned through doing; personal commitment denotes the ethical and intellectual responsibility a researcher assumes; socialisation means participation in a scholarly community where judgement is tested; deep reading denotes sustained engagement with texts that shapes understanding. The paper places those practices at the center of any effort to resist what the authors call the degradation of research by AI.

Why it matters

The authors' argument matters because it reframes debates about AI in research away from output quality and toward professional formation. If generative systems replace tasks that ordinarily create tacit skills and communal trust, individual researchers and the communities that validate scholarship may lose capacities that are not visible in a single paper. Novelli and Floridi point to a risk that the aggregate effect of outsourcing inquiry could produce superficially polished results while eroding the social and epistemic infrastructure that sustains reliable knowledge.

This critique is not an argument against using AI per se; it targets the delegation of "central tasks of inquiry" and the assumption that human presence as prompter or checker is sufficient. Their remedy, second scholarship, asks communities to identify practices that must remain human-led and to teach, enact and reward them.

What to watch

Watch whether scholarly communities adopt explicit policies or training that protect tacit practices the authors list: mechanisms that sustain deep reading, mentorship that cultivates tacit knowledge, and norms that require demonstrable personal commitment and communal vetting. The arXiv record for the paper is arXiv:2607.04049, submitted 4 July 2026; that record and the paper's arguments may surface in policy debates and departmental guidelines about acceptable uses of Large Language Models in research.

Novelli and Floridi provide a conceptual checklist more than technical fixes: identify what cannot be automated, teach it, and make it count in evaluation. Their call for "second scholarship" reframes the question from whether AI can improve outputs to which parts of scholarly formation we are willing to surrender.