The Quaternary Problem

orange wingtip shoes with polka dotted socks

I started using generative AI partly because I was tired of being the person whose thoughts arrived at the meeting wearing the wrong shoes. Because, in my world, thoughts wear shoes, if only the beginning of the situation. I would have a question, a reaction, or a connection that made sense to me but seemed a little out of step with what was expected. AI offered a place to rehearse. I could ask it to soften an edge, rearrange an explanation, or untangle my reasoning. My ultimate hope was that the predictive capability of language models could help my thoughts gain entry to conversations. 

Now I sometimes wonder whether that assistance has introduced another distance. Are people responding differently to my writing because they suspect AI helped? I do not know. That is my interpretation, and it deserves the same scrutiny I would give any other unsupported claim. But there is an irony here: I sought help being understood, and now I wonder whether the help has made me harder to recognize. Has my desire to fit in at work thwarted confidence in my thinking? If so, should I somehow show my fancy footwork?

Before generative AI, we had a familiar vocabulary for the relationship between information and evidence. Primary sources supplied original material: letters, interviews, observations, research findings. Secondary sources interpreted that material. Tertiary sources organized and summarized existing knowledge in encyclopedias, handbooks, and reference works.

These categories did not assure truth. An eyewitness could misremember. A historian could overlook a letter. A textbook could carry an old assumption through several editions. Still, the categories helped us trace a claim. Who observed this? Who interpreted it? What supports the conclusion?

Evidence gave us something to examine beyond the confidence of the prose. Citations offered a route back. Methods explained how findings were produced and could potentially be reproduced. The route could be incomplete or misleading, but we could ask to see it.

Generative AI complicates that map.

A language model produces text through prediction. It may work from documents we provide or sources it retrieves. It can also generate a convincing explanation without a reliable trail connecting its claims to evidence. The National Institute of Standards and Technology (NIST) identifies confidently presented false or erroneous content as a risk of generative AI. Sentences generated by AI do not necessarily look different when the footing beneath them gives way.

I am calling this the quaternary problem: the difficulty of establishing a text’s relationship to evidence when predictive generation can produce the appearance of informed synthesis.

This is a working concept, not a newly discovered shelf in the library. AI does not consistently sit one level beyond tertiary sources. It can summarize an original letter, explain a research review, or produce an answer without consulting either. “Quaternary” gives me a way to name the disruption in how we recognize what we are reading.

The disruption reaches authorship, too. What belongs to the person who brings the question, rejects three answers, supplies the context, and revises the fourth? What responsibility comes with putting our name above the result? Writing style alone cannot settle those questions.

There are possibilities here worth keeping. Students can begin with a provisional explanation and investigate where it holds, where it fails, and whose perspective is missing. Writers can try different ways of expressing a difficult thought. An unexpected connection can become a question worth pursuing. Text can help us think before it earns our confidence.

But we need to distinguish an idea worth exploring from a claim that has not arrived with the same cost as human ideas. We need to be able to explain what we accepted, what we checked, and what remains uncertain.

Perhaps the quaternary problem invites us to become more explicit about evidence and more curious about one another. We can ask how a writer arrived at a piece of writing. And we can leave a little room at the table for the thought that shows up wearing unusual shoes.