1 of 14

Automating the Expert: AI, Deskilling, and the Crisis of Professional Identity in Health Sciences Librarianship

Presenter: Megan Kennedy

CAPAL Conference 2026

Hello, my name is Megan Kennedy and I am a health sciences librarian at the Geoffrey and Robyn Sperber Health Sciences Library at the University of Alberta. I am here this afternoon to talk about the impact of AI tools in evidence synthesis - a core part of the work of health sciences librarians. Thus far, the literature has mostly focused on the quality of AI generated outputs when it comes to evidence synthesis - essentially asking, are they sufficient, do they meet the expected scientific rigor and are they comparable to their human generated equivalent. But what is missing from the conversation is an exploration of the human disruption caused by these AI tools. My aim was to explore the crisis of professional identity in health librarianship to understand how our work might need to adapt in the future.

2 of 14

Defining Evidence Synthesis

Image from Evidence Synthesis, by Wilfrid Laurier University Library, 2025. (https://library.wlu.ca/research/evidence-synthesis).

3 of 14

The Efficiency Imperative & The Information Tsunami

The driving force behind the push to integrate AI into evidence synthesis is an ongoing information tsunami. In modern healthcare, there is an urgent, non-negotiable demand for rapid, high-quality synthesized evidence to develop policy and directly inform clinical practice guidelines that impact patient care.

The problem is that traditional, gold-standard methods of evidence synthesis require immense amounts of time and human power and in our current institutional landscapes, both are usually in incredibly short supply.

This resource scarcity is exactly why researchers and institutions are turning to AI tools. The promise of automation isn't just about a technological trend; it is a desperate search for efficiency. However, as these tools are introduced to automate complex review processes, they bring a hidden sociotechnical disruption that directly challenges the traditional role, visibility, and professional identity of the health sciences librarian.

4 of 14

"Expert Searchers"

The traditional role of health librarians supporting evidence synthesis

Traditionally, health sciences librarians are recruited onto research teams for a very specific, highly visible skill: our role as 'expert searchers.' In the complex world of evidence synthesis, designing, translating, and executing a comprehensive search strategy is how we explicitly demonstrate our specialized value to a project.

Because this is a tangible, visible output, it has historically formed the bedrock of our professional identity and served as our primary currency for co-authorship on published reviews.

But in the age of AI, we have to face a uncomfortable question: What happens to our professional standing when the mechanical execution of that search is handed over to an algorithm? If our identity is tied solely to doing the search, automation risks making our traditional value proposition obsolete, forcing a critical re-negotiation of what it actually means to be an expert in this field.

5 of 14

Invisible Labour

What was once a visible contribution to evidence synthesis becomes more and more invisible in the wake of AI-automation

Image generated with "A cinematic, futuristic digital art piece depicting a human holding a mirror to the sky" by OpenAI's ChatGPT 5.5, 2026, https://chatgpt.com.

When we talk about artificial intelligence in evidence synthesis, the immediate conversation is almost always about speed, efficiency, and powerful outputs. But what we need to look closer at is the profound shift this causes in the visibility of library work.

The concept of invisible labour is already well-documented in health sciences librarianship literature. Historically, our labor was made obvious to research teams through the highly visible act of designing, translating, and executing complex database searches. It was an explicit, technical output that teams could see, value, and measure.

However, AI-automation threatens to drive this work further underground. As discovery platforms increasingly automate search-strategy generation and syntax translation, our role shifts. We move from being the creators of the search to the overseers or auditors of machine-generated outputs. We are behind the scenes validating terminology, ensuring boolean logic isn't broken, and correcting 'black-box' errors. This oversight is absolutely vital for maintaining scientific rigor, reproducibility, and transparency - yet it is structurally far less visible to the research team.

And here is where the tension directly impacts our professional progression: authorship. Under current International Committee of Medical Journal Editors (ICMJE) criteria, if a librarian creates and executes the search architecture, our claim to co-authorship is quite cut-and-dry. But if our role is perceived by research teams as mere 'technical assistance' or simple oversight of an AI tool, that justification becomes incredibly murky.

For academic health librarians, this is not just an existential identity crisis, it is a material career threat. In an institutional environment where peer-reviewed co-authorship is a heavy currency for tenure, promotion, and professional legitimacy, the invisibility driven by AI automation has real-world consequences on our career development. We cannot afford to let our intellectual interventions be naturalized into the background of the technology.

To keep our authorship authority, we must argue that critical oversight and output validation ARE creative, intellectual contributions necessary for research integrity, not just "technical assistance."

6 of 14

Emotional Labour

Sociotechnical tension between AI-mediated efficiency, scientific rigor, and the actualized cognitive and emotional weight behind the scenes

Image generated with "A realistic, high-contrast dramatic photograph capturing the invisible emotional labor of a modern health sciences librarian" by OpenAI's ChatGPT 5.5, 2026, https://chatgpt.com.

We must also confront the affective, emotional dimensions of this technological transition. A massive sociotechnical tension is currently playing out in real-time behind the scenes

Drawing on emotional labour theory, our professional identity has always involved active emotional regulation. In evidence synthesis support, a primary, yet unwritten, part of our job is guiding anxious or overwhelmed researchers through highly demanding methodological terrain. We manage their expectations around scope, feasibility, and time. We are expected to project confidence, patience, and absolute authority while absorbing and regulating our own frustrations when methodological rigor is undervalued.

Now, introduce AI into this relational ecosystem. This creates an intense role conflict. On one hand, AI-tools promise instantaneous, effortless efficiency to researchers under massive institutional pressure to publish quickly. On the other hand, we are bound by our professional identity as methodological stewards to champion reproducibility, transparency, and strict adherence to reporting guidelines.

The emotional labor here intensifies significantly. We now have to absorb the frustration and impatience of research teams who wonder why human validation, translation, and critical appraisal take weeks when an AI tool produced a search query in five seconds. We are left carrying the emotional and cognitive weight of defending scientific integrity against the seductive allure of automated speed. This ongoing negotiation, balancing empathy for busy researchers with a uncompromising defense of research standards, is a profound driver of stress and burnout in our field. It reminds us that navigating AI in evidence synthesis is never just a technical challenge—it is a deeply human and emotional struggle.

7 of 14

Knowledge Extraction

Capitalizing on documented expertise, the published work of librarians supporting evidence synthesis trains AI-tools to perform searches efficiently

Image generated with "A cinematic, futuristic digital art piece depicting a human mentor training an AI entity in a glowing, holographic archives room." by OpenAI's ChatGPT 5.5, 2026, https://chatgpt.com.

This brings us to a highly provocative dimension of our changing landscape, and that is what I call data labour, or knowledge extraction.

We need to ask a fundamental question: How did these evidence synthesis and generative AI tools become so incredibly efficient at understanding systematic review architecture and search logic in the first place? The answer is simple: they learned it from us. These algorithmic models rely entirely on training data consisting of previously published reviews and their search strategies - a large number of which were meticulously built, validated, and documented by health sciences librarians.

This means our profession has engaged in a massive amount of uncredited data labor. Every time we publish a gold-standard search strategy to ensure reproducibility, that strategy is scraped and ingested into a corpus that trains an LLM to mimic our expert search behavior.

The deep tension here, as highlighted by Abbott’s theory of professional jurisdiction, is that our own past labor is being codified, aggregated, and redistributed through automated systems to perform tasks historically claimed by our profession. This knowledge extraction threatens to weaken our traditional authority over the search process by making the execution look easy and automated to the end-user.

But understanding this dynamic is also our leverage. It proves that the machine's outputs are only as good as the librarian logic it was trained on. It gives us a brand new mandate to step up and demand strict oversight, validation, and professional accountability over how our intellectual property is used.

8 of 14

Safeguarding Epistemic Trust

Nothing is neutral and when bias mitigation is the aim of the game, critical interrogation of AI-tools is essential

Image generated with "A realistic, high-contrast conceptual photograph about research integrity and epistemic trust in the age of automation." by OpenAI's ChatGPT 5.5, 2026, https://chatgpt.com.

Next let's talk about our role in safeguarding epistemic trust and research integrity. Epistemic trust is the trust we place in knowledge and the methods used to produce it. In evidence synthesis, that trust has traditionally been built on total transparency and visibility across the research process, things we, as librarians, enforce through meticulous search documentation and adherence to reporting standards.

But AI tools completely disrupt this transparency by operating as 'black boxes'. Many proprietary discovery platforms intentionally obscure how query processing happens, leaving users with only an input and an output, but no way to audit the steps in between. Because these systems generate sleek, immediate, and highly polished interfaces, researchers frequently fall into the trap of overestimating their reliability, completeness, and neutrality.

But as critical librarianship reminds us, no technology is neutral. Algorithmic systems are fundamentally shaped by the data used to train them, which means they inherit, embed, and actively amplify the systemic biases and inequities already present in the scholarly record. When a black-box tool filters information through an un-auditable algorithm, it risks burying crucial evidence or misinterpreting studies.

If bias mitigation is the absolute aim of the game in evidence synthesis, then the uncritical adoption of AI tools represents a direct threat to research integrity.

And this is precisely where our professional identity must adapt and assert itself. We must transition from 'information fetchers' to critical mediators. Our value is no longer just about executing the search; it is about cross-examining the machine.

9 of 14

What can we do?

So, given these heavy disruptions—the invisibility, the emotional weight, the data extraction, and the threat to epistemic trust—what is our move? How do we respond?

Faced with these overlapping pressures—the loss of visibility, the emotional strain, the extraction of our data, and the erosion of epistemic trust—we arrive at a critical crossroads. The question isn't just 'How do we change our day-to-day workflows?' It is an existential question: What can we do as a profession to secure our future?

To answer this, we have to look outside of librarianship and turn toward the sociology of professions. Sociologist Andrew Abbott's foundational theory of The System of Professions teaches us that professions are not static. Instead, they live in a shifting ecosystem where they must constantly defend and negotiate their 'jurisdiction'—their claimed right to do a specific type of work.

Abbott argues that when a profession’s jurisdiction is disrupted—whether by institutional changes, economic shifts, or, in our case, automation and technological encroachment—the profession faces a fork in the road. If we do nothing, our jurisdiction degrades, our work becomes naturalized or deskilled, and we lose our professional autonomy.

So, our response to AI in evidence synthesis cannot simply be a matter of administrative adjustment. It must be a deliberate, strategic act of professional re-jurisdiction. We are at the exact moment where we must redraw our borders. As the mechanical execution of the search shifts to algorithmic systems, we must explicitly claim a new, higher-level jurisdiction over the governance, critical evaluation, and ethical oversight of these tools.

Let's look at exactly how we can map out this structural shift across our professional landscapes

10 of 14

Avoid the "Paradox of Expertise"

Disengage as "expert searchers"

BUT

Remain critically engaged as "information experts" in evidence synthesis

The first step is that we must actively avoid what organizational sociologists call the 'Paradox of Expertise'. This paradox occurs when established professionals see an automated tool encroaching on their territory, and their immediate reaction is to completely disengage from the technology, reject it outright, and try to double-down on traditional manual methods.

History has shown us that this is a professional dead-end. We saw it during the internet boom of the late 90s and early 2000s, where some in librarianship resisted digital search engines, which only served to limit their relevance as users moved online anyway. If we simply refuse to engage with AI tools for evidence synthesis, we won't stop researchers from using them. Instead, we will just isolate ourselves from the research lifecycle entirely, allowing poorly constructed, biased, AI-generated reviews to be published without any professional quality control.

Therefore, our survival depends on a strategic identity shift. We may need to relinquish the hyper-specific, narrow title of "expert searchers". But in doing so, we reclaim a much larger, more powerful identity as 'information experts' and critical mediators.

We are some the only professionals inside the research ecosystem with the deep methodological knowledge required to evaluate these algorithms, expose their limitations, and interrogate their inputs. By remaining critically engaged with AI evolution, we ensure that we are not replaced by the machine, but rather positioned as the indispensable human authority supervising it.

11 of 14

Credit Where Credit is Due

Modernize evaluation frameworks and expand authorship criteria so that the work of librarians in evidence synthesis is recognizable in a meaningful way

SUCH AS

Utilising the CRediT author statement taxonomy to explicitly document contributions like methodological oversight, validation of AI-generated search strategies, and ethical AI use throughout the process of evidence synthesis

As I mentioned earlier, our traditional avenue for professional recognition in evidence synthesis has been co-authorship. But as the mechanical execution of building a search becomes increasingly mediated by automated platforms, we run a severe risk of being pushed into the footnotes or dropped into the acknowledgements section. To prevent this, we must actively advocate for modernized evaluation frameworks within our universities and explicitly expand what constitutes a 'meaningful scholarly contribution'.

A highly concrete, actionable tool we can adopt right now to solve this is the CRediT author statement taxonomy, or the Contributor Roles Taxonomy. The CRediT taxonomy breaks down a research project into fourteen specific, recognizable roles. Under this framework, our work in an AI-mediated environment is explicitly visible and quantifiable. We can claim formal authorship by documenting our intellectual labor under roles like Methodology oversight, Validation of AI-generated search strategies, and ensuring Data Curation and ethical compliance throughout the entire lifecycle of the review.

By pushing our institutions and journal editors to adopt these modernized standards, we take the murkiness out of the authorship conversation. We provide a standardized, undeniable framework that proves auditing a machine-generated synthesis is a high-level cognitive intervention. This ensures that our labor is recognized in a meaningful way that directly supports our career development, tenure, and promotion processes.

12 of 14

Knowledge is Your Greatest Tool

Education about AI literacy, AI-tools for evidence synthesis, and algorithmic evaluation will be crucial skills for health sciences librarians going forward

SO THAT

Librarians understand the limits of AI technology and can educate researchers to maintain scientific rigor and ethical integrity in evidence synthesis

To effectively occupy this new territory as critical mediators, our professional competencies must evolve. This transition requires a deep commitment to AI literacy and algorithmic evaluation. This isn’t about learning to code; it is about understanding the boundaries, limitations, and data mechanisms of these discovery platforms.

When we master how these tools operate, we gain the expertise required to protect our institutions from a dangerous loss of critical oversight. This technical literacy allows us to look a research team in the eye and explain precisely why a seemingly perfect, instantaneous AI search output is actually methodologically flawed, terminology-deficient, or structurally biased.

Ultimately, this upskilling is what converts our professional anxiety into professional authority. We stop being passive users of technology and become the educators who guide researchers through the ethical minefields of automation. It ensures that when a health system looks to integrate AI into evidence synthesis, the library is sought out as the essential governance hub ensuring the work remains valid, reproducible, and trustworthy.

13 of 14

Final Thoughts

We're not deskilling, we are reskilling

To conclude, it is clear that AI does not mean the obsolescence of the health sciences librarian; rather, it marks a profound and disruptive transformation of our role. While the mechanics of searching are being automated, our labor is actually being freed for higher-dimensional, cognitive value creation.

We are uniquely positioned to step into the role of supervisors of human-machine collaboration. However, to secure this future, we cannot remain passive. We must actively demand changes to our institutional systems of acknowledgment, push for modernized authorship criteria like the CRediT taxonomy, and commit to algorithmic literacy.

The future of evidence synthesis still hinges entirely on human expertise. AI can automate the retrieval of data, but it cannot automate the critical judgment, ethical stewardship, and epistemic trust that librarians bring to the scientific record.

Thank you