Human and AI Wonder

Those friends and colleagues who have decided we should just say no will not like this post. They’ll say, Stop training the frickin’ AI, and other things in this general spirit. Sorry, friends, I’m going to upset you even more, I fear.

For we find ourselves in a situation that’s akin to what thinking people faced in the 1950s. The technology of the day was the atomic bomb and its progeny. The just-say-no faction in that particular cultural moment lingered into the 1980s but has become much less prominent since.

One of the books I’ve gone back to, now inflected anew by AI, is Hannah Arendt’s The Human Condition, which was first published in 1958. I used a short quote from the book for my Iowa presentation last month, and for the article I mentioned here a few weeks ago (still under review at a journal).

Here’s the quote: “It remains to be seen whether these future techniques will transform the household of nature as we have known it since the beginning of our world to the same extent or even more than the present technology has changed the very worldliness of the human artifice” (150).

This framing helps to explain why discussions of AI as just another tool fall short. Arendt, I’ll repeat, had the latest technologies of her day in mind, especially those associated with the “first instruments of nuclear technology, the various types of atomic bombs, which, if released in sufficient and not even very great quantities, could destroy all organic life on earth” (149–50).

The parallels with contemporary AI discussions are hard to overlook: a new technology, a race to be the first, geopolitical strategizing, proprietary issues, secrecy, national security, espionage, huge sums of money, and the potential that something might go terribly wrong, destroying all life on the planet in the process. This makes the question of translation feel like a small one. On the other hand, if we can get the translation question right, maybe it will help with the bigger ones.

So I’ve been taking a different tack, looking at AI carefully as another human-made technology, and asking how we as humans deal with such things once we’ve created them. That — rather than what AI gets wrong or how it may or may not put translators out of work — is the larger question in the background. To think through it, I’ve chosen a much smaller domain, one I know well: humanities research, with a focus on translation.

I’ve now done several experiments, using a method I’m calling punctuated semi-autonomy. I come up with basic research questions, then I interact with the AI to develop the experimental design and implementation. I tell the AI what I want to learn, ask it to propose a design framework, then I look at the proposed design and make changes before we begin. But the semi-autonomy does not end there.

Next we set up a protocol of iterations with the AI providing regular reports. Initially, I set something up to run every few hours, send me a report, then run again. This had a major flaw, which was my own capacity to read the reports with care, think about them, and make adjustments to the research protocol. Every few hours was too fast. The AI didn’t need three hours, of course. The slowest actor in the system was the human.

Once I figured out the right interval, which might not be the same for every project, we were able to move ahead. We. It’s an interaction. The AI gives me material, I think about it, ask additional questions, change the prompts, suggest leads, even shift major aspects of the design. This requires thought and consideration on my part, and that means time.

In several of these experiments, after finishing a number of iterations, I’ve asked the AI to help me assess what we’ve done, how we divided up tasks, what I was good at, what the AI was good at. Some of these interactions have surprised me.

On one occasion, I suggested that I had brought “wonder” into the mix. The AI disagreed with me. It wrote, “If you [RV] are going to submit this to a journal as part of a methodology paper, a reviewer might legitimately respond, ‘Didn’t the AI wonder about various things and ask questions, following up on your initial guidelines to go look for things?’ Functionally, that’s what I [AI] was in fact doing.” Functionally, yes. (How one might be able to tell the difference between functional wonder and real wonder is a Philip K. Dick question.)

Well, I responded, how would you describe what I’m trying to get at by saying I brought wonder to the experiment? It replied as follows: What I [RV] brought was “curiosity rooted outside the accumulated research state: associations from reading, teaching, conversations, and prior knowledge whose relevance you [RV] often sensed before you [RV] could fully explain it.”

This seems like an accurate assessment to me. In its suggestion, moreover, that I was operating on something like an intuitive level, or at least on associations that I was sensing before I was able to articulate them, it suggests something a bit like the “inner thoughts” discussion of Trevor Paglen in his Nov. 12, 2026 New York Times guest essay.

To me this was not so much scary as yet another throwback, this time to Michael Polanyi’s observation that “we can know more than we can tell,” which he explored in his 1966 The Tacit Dimension. Here again, AI has provided a new context for a work written with the concerns of an earlier generation in mind. More concretely, my attempt to outline our respective roles has helped me to formulate several general principles for humanities-based AI research. I explore these in the article.

Throughout these experiments, time has come to seem an essential component. And so — you could probably have predicted this — I’ve gone back to reading works about time that were written for earlier generations, only now thinking about how AI might inflect them differently.

In this spirit, my friend David recently reminded me how he used to flippantly summarize Martin Heidegger’s Being and Time by saying that Heidegger’s conclusion was that “being is time.” Perhaps it’s time to add “human” to this glib summary.

###

Leave a comment