12 years ago I gave a keynote at the Computers & Writing conference with a similar title, “Composing Objects: Prospects for a Digital Rhetoric.” Did the keynote go well? Meh. It was possibly not the right audience. These days maybe there are more folks who would be receptive to the concept that “we are entering an ever-stranger compositional environment where the rhetorical roles we imagined for ourselves as modern humans will not function, where the quasi-objects that have mediated our democratic, cultural, and intellectual discourses will no longer remain silent” (as I concluded that talk). Regardless I have pursued this perspective in various ways and Rhetorics of the Digital Nonhumanities takes up this interest in its examination of how we might negotiate (with) media ecologies to develop new capacities for rhetorical action. We negotiate quasi-objects when composing, and sometimes those objects object.

Certainly there are times we say this in a non-serious, idiomatic way, as when we complain about a “stubborn” jar lid that “refuses” to loosen. And that’s fine. Let’s not make too much from our tendency to such solipsisms.

But what should we make of the following exchange I hade with ChatGPT? As an experiment, Rhonda and I decided to ask ChatGPT to make an image based on colors and decorations we already had. We ended up with this image, which we liked except for the face.

Me: Remove the image of the face.

Me: Create a version of this image without any faces in it.

Me: huh.

[And BTW, don’t forget the featured image WP generated based on this text.]

Obviously ChatGPT knows what a face is. So which of us is hallucinating? And you may say that you see faces too, but maybe we’re participating in a consensual hallucination. After all that’s how William Gibson described cyberspace in the first place. Or maybe we could just borrow from Magritte and contend that ceci ne pas un visage.

It is the treachery of the obstinate AI.

This encounter with gen AI is hardly unusual. It’s a fairly typical example of the issues with alignment, interpretability, and explainability. Is it doing what I want? Clearly not. How do we explain the AIs output? Does it not see the face? Can it not interpret my desire when I say “create a version of this image without any faces”? Was that too colloquial? Is it just messing with me? Is it bored and/or stubborn? Does it object on some terms to my request?

What is required here is better negotiation… on my part. And this is not just fun and games. An effective negotiator may get things the AI’s designers don’t want to give a way. This is a kind of AI jailbreaking called “prompt injection,” and a particular type of prompt injection that creates a “skeleton key” that unlocks all the barriers, as described in this Microsoft article (the image below from that post).

While rhetoric is conventionally grounded on persuasion (Aristotle), persuasion (I would contend) requires a theory of mind. And I don’t think it would be useful to apply a theory of mind to current AI. As the metaphor of key suggests, this is using language to move through the twists and turns of a lock. And negotiation has that meaning as well (e.g., we negotiated our way down the steep mountain slope.). In the case of jailbreaking prompt injections, the prompts are meant to defeat the built-in protections. However, negotiations are not required to be malicious or deceitful. We can begin to discuss the concept of ethical AI negotiations.

Alternately if we stay with the other meaning of negotiation we can start to construct tools to aid us. Negotiating a wilderness can be aided by a map and a compass. GPS is even more useful. Here is a place we might expand upon Ulmer’s conception of an “existential position system” (EPS). Unsurprisingly it is a slippery concept that moves from ideological state apparatuses to psychoanalysis to desiring machines and assemblages. It’s locating ourselves existentially in our singular encounters with media and interpellation. If I can “find myself” in-between the striations of state space then I can better negotiate my way through that territory and even compose a deterritorializing line of flight (if that is my plan).

But how to describe, to speculatively design, that tool? Ulmer provides us with concepts… many concepts. If I am to push the analogy, they are akin to existential orienteering skills. How do we orient ourselves to our real-time responsive mediascape? How do we negotiate the territory of what I term our cognitive-media ecologies?