I recently came across a blog post from Center on Privacy & Technology at Georgetown Law that squarely confronted this question. They realized that as a center focused on privacy and law in the digital era, writing constantly about AI, their own language was undermining their critical mission. What’s interesting: their response wasn’t “learn some tools.” It was four principles for disciplinary self-awareness:

1. Be specific about what the technology actually does
2. Acknowledge when corporate opacity prevents full understanding
3. Name the corporations responsible
4. Attribute agency to humans, never to the technology itself

That last principle strikes me as the crucial one, especially for library users. Yes, anthropomorphic language is imprecise, but it also shapes expectations, calibrates trust, and obscures accountability. If an AI tool “made a mistake,” who’s responsible? When users routinely hear phrases like “the model knows,” “it doesn’t want to answer,” or “it prefers coherent completions,” this framing does some heavy lifting. It has real consequences for how people evaluate and emotionally relate to these technologies.

The Georgetown intervention might point to a practice that academic libraries could adopt. As they pivoted: We write about privacy and law constantly, and we realized our own language was undermining our critical mission. We could try something similar: We’re librarians, we teach information literacy about AI. What if we applied that same critical apparatus to the language we use to describe AI? Could uncritical discourse upstream affect everything else downstream?

I’ve come to suspect that “AI literacy” as a phrase may be part of the problem. It naturalizes AI as a coherent category requiring specialized knowledge. This framing could inadvertently send professionals looking for new competencies when they already possess the critical skills needed for this work. In the Georgetown example, there’s no mention of becoming an expert in anything. Just noticing that words matter.

That’s the good news. We don’t need a new literacy. We have the ones we already practice. Consider the ACRL Framework for Information Literacy. “Authority Is Constructed and Contextual” describes exactly what happens when corporate communications position AI as “knowing” or “understanding.” Authority is being linguistically constructed for algorithmic systems. “Information Has Value” illuminates why: anthropomorphic framing has economic value, building trust and driving adoption. “Information Creation as a Process” reminds us that press releases announcing AI capabilities are themselves information products with authors, interests, and purposes.

For the past year, I’ve been experimenting with what this might look like in practice.  The project, which I call Discourse Depot, uses structured prompts and data schemas to systematically analyze how AI is described in corporate communications, news articles, research articles, and  policy documents. (Yes, I’m using an LLM to deconstruct language about LLMs). The goal is to make visible the metaphorical moves, surface linguistic choices and audit explanations that construct AI as an autonomous agent or a “mind” rather than what it actually is: an impressive (but mechanistic) technology and a corporate product with human designers, investors, and beneficiaries. 

From Discourse Depot: Example reframing from an audit of Google’s AI and the Future of Learning

These formulations aren’t perfect, but they emphasize human agency and institutional accountability, countering the framing of AI as autonomous. But the intellectual work was in the translation: taking the critical habits information literacy had already given me and applying them to a new domain.

The bad news: the technology behind generative AI tools will always deal in probabilities not facts. I’ve had to make peace with this. And as a community, we’ll need to reckon with things like whether placing these systems between users and sources is worth the risk. This will require attention. But attention is what we do. Critical evaluation is what we teach.

What excites me isn’t that librarians can find a way to lead on AI literacy since I recognize that they already possess the tools and critical energy for that leadership. We’ve been teaching source evaluation, authority construction, and information creation processes for decades. The only thing new is the object of analysis.