Educators and especially librarians will find this paper by McCrary (2026) very interesting. The core claim is that AI features baked into platforms students treat as authoritative redistribute research agency without students noticing or institutions reviewing.
Predictive search, LLM article summaries, generative concept maps, and retrieval-augmented answers all reshape what students see and how they interpret it. McCrary borrows the phrase “ghosts in the machine” from Meza and Chantal (2023) to describe the resulting student experience.
How AI in Library Databases Redistributes Student Agency
McCrary builds his argument on two theoretical ideas. Hanson’s composite agency theory (2004-2014) says most research actions are joint products of human intention and nonhuman systems acting together. Hernández-Orallo and Vold’s (2019) concept of AI as cognitive extender pushes further: AI doesn’t just assist; it becomes part of how a student remembers, reasons, and decides. Together, the two ideas explain how an undergraduate using a database isn’t acting alone, even when the interface looks like a familiar search bar.
The redistribution happens across four research stages. Predictive algorithms shape the question itself by suggesting related concepts before any results appear. ML systems decide source visibility through ranking. LLM summaries pick what gets emphasized and what gets dropped.
At verification, retrieval-augmented generation often produces answers without clear source attribution. McCrary writes that “the student becomes a spectral presence in a system where the provenance of knowledge is unclear and the capacity for independent critical thinking is redistributed” (p. 3).
The connection to Gerlich (2025) on cognitive offloading and to the cognitive debt argument from Kosmyna et al. (2025) is direct. AI handles the work, students stop building the skill, the skill atrophies, and we graduate researchers who can’t research without an algorithm in the loop.
The Marie Case: Ghost or Architect
The case study is the strongest part of the paper. McCrary walks the reader through “Marie,” an undergraduate nursing student running a literature review on adolescent anxiety, in two scenarios. In the first, she accepts the algorithm’s source ranking, takes AI summaries at face value, and lets a GenAI concept map define her thinking. She becomes a ghost in the machine.
In the second, she filters search results manually, reads AI summaries as scaffolds before opening the full text, treats AI concept maps as hypotheses to test, and verifies AI claims against original sources. McCrary’s observation lands cleanly: “Both scenarios demonstrate that AI makes research faster, but the second proves that it need not make the process less hers” (p. 4).
This is the move teachers can use immediately. The same AI tools either build student skill or drain it. The pedagogical question is whether students are taught how to be the architect, not the ghost. AI literacy can’t stop at telling students to avoid the technology. The real work is teaching them how to direct it.
What McCrary’s Argument Misses
I have two concerns. First, this is a conceptual paper, and McCrary admits as much. The “Marie” scenarios are constructed, not observed. We need empirical work, the kind he proposes for future research, before we treat the ghost-versus-architect binary as a stable description of student behavior. My guess is that real students drift between both modes within a single session, sometimes within the same article.
Second, the paper centers librarians, which makes sense given the journal, but it understates how much AI use now happens outside library platforms entirely. A nursing student writing a literature review in 2026 might use ClinicalKey AI for one task, ChatGPT for another, and Perplexity for a third, all in one afternoon. Library instruction can shape part of that, not all of it. AI literacy has to be a campus-wide project, which is what I argued in my earlier post on scaffolded AI literacy in academic librarianship.
The paper’s vendor critique is sharp and worth keeping. AI integrations from major platforms are default features, not opt-in additions. Marketing copy like “trusted content, powered by responsible AI” frames these tools as institutionally endorsed, even when no formal review has happened.
McCrary calls this out plainly, and I’d take the argument further. Most institutions can’t actually evaluate these AI integrations because they don’t have the in-house expertise. The asymmetry between what vendors ship and what librarians can scrutinize is widening fast.
McCrary’s central recommendation is that we shouldn’t try to remove AI from libraries, since vendor integration makes that unrealistic, but should design the human-AI composite network so the student’s agency stays at the center. I agree. The challenge is execution. Library systems built by vendors don’t treat student agency as a design priority, and pretending otherwise is wishful thinking.
Vendors will keep adding AI features. The pedagogical question stays the same. Are we teaching students to direct the system, or letting the system direct them?
References
- Gerlich, M. (2025). AI tools in society: Impacts on cognitive offloading and the future of critical thinking. Societies, 15(1), Article 6. https://doi.org/10.3390/soc15010006 //
- Hanson, F. A. (2004). The new Superorganic. Current Anthropology, 45(4), 467–482. https://doi.org/10.1086/422080.
- Hernandez-Orallo, J., & Vold, K. (2019). AI extenders: The ethical and societal implications of humans cognitively extended by AI. AIES ’19: AAAI/ACM conference on AI, ethics, and Society (pp. 507–513). ACM.
- Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X.-H., Beresnitzky, A. V., Braunstein, I., & Maes, P. (2025). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing tasks. MIT Media Lab. https://www.media.mit.edu/publications/your-brain-on-chatgpt/
- McCrary, Q. D. (2026). Are we ghosts in the machine? AI, agency, and the future of libraries. The Journal of Academic Librarianship, 52, 103181. https://doi.org/10.1016/j.acalib.2025.103181
- Meza, B. E., & Chantal. (2023). “Have we finally become ghosts in the machine?” The Philosopher. https://www.thephilosophe
r1923.org/post/have-we-finally-become-ghosts-in-the-machine