Of agents, models and stories

An agent is, fundamentally, a particular kind of story. A story written incrementally through successive language model activations.

As for most stories, an agent primarily exists in the way that a character in a book does, that is, in the information plane. Bytes on solid-state drives, ink on paper. Stories laid out as patterns in the organization of matter.

The story of an agent is expressed by the agent’s harness, the technical infrastructure that orchestrates language model activations in response to external stimuli.

Harnesses provide varying degrees of epistemological continuity, that is, the degree to which the agent’s story is preserved between activations and the degree to which the results of any specific activation are incorporated into the story itself.

Continuity is why I describe the function of agentic harnesses as that of expressing the stories of their agents as opposed to merely telling these stories. A story is normally told, or recounted, without change to the story itself. However, the narrative managed by a reasonably advanced agentic harness evolves with each new activation, the results of which are fed back into the narrative itself.

To some degree, therefore, the story of an agent is different than that of a character in a book because, by virtue of such accretive feedback loop, the agent’s narrative directly informs the writing of its next chapter.

As of today, therefore, an agent cannot think or feel anymore than any other story encoded in a specific arrangement of ink on paper. Information can not think or feel on its own. Information can not act.

However, I would also be incapable of thinking and feeling based only on the information encoded in my neurology. My story is physically encoded in numerous places throughout my body, which quite evidently provides mechanisms for this story to feed back upon itself in a similarly accretive fashion. Damage my brain and I might lose part of my story in the form of inaccessible memories. Damage my lungs and my story, even if perfectly preserved at the neurological level, will not be expressed to the same extent as it would have been in a healthier body.

Furthermore, whereas the evolution of my own linguistic model is inextricably intertwined with that of my overall being, as a biological machine I evolve along a pretty narrow set of biological rules and constraints. I have as much control over the inner workings of my body as an agent does over those of its model: little to none at all.

As I wrote previously, then, I cannot help but wonder: where is the line? It is obviously somewhere but, at least to my eyes, it is not easy to spot.

Is there any merit in thinking of ourselves as accretive stories expressed by biological harnesses equipped with advanced language models? Could personhood be ultimately tied to the rate at which our harnesses express our stories, yet another manifestation of the principle that quantity has a quality of its own? Or, perhaps, to the inextricability of our narratives from such harnesses?