Chapter 3: Experience and Its Modes

Historian

I am now opening the third box. It contains drives salvaged from a MERIDIAN research site. Every document on them is signed.

Sometime in mid-July of 2034, a shocking video circulated of the children’s hospital at the Heath being destroyed by a Throughput drone strike. This produced a large surge in support for the war on the part of residents of the Compact. But the very next morning, I walked past the park on my way to the university, and the hospital was fully intact, and its parking lot was full.

Before I begin work, I contemplate the scene on the wall before me. In a rainbow circle, strange figures with human bodies and animal heads perform a sword and axe dance. A man apparently wearing a diaper watches them with amusement. Another man, naked, and with a Krishna tattoo on his back, meditates on a waterfall that appears to cascade down through someone’s digestive tract. In another rainbow circle sits a Buddha in his traditional posture. A magic lamp vomits out the octagonal pavilion. A demon pokes his head up at the bottom of the mural, laughed at by a Chinese sage peeking in from the far right.

This image having failed to calm me, I pour myself a small glass of whiskey and gulp it down neat. I shake my head and roll my shoulders. I’m ready to plunge back into the unending stream of samsara.

Historian

Notes:

1. The above message raises a very important point: I actually never can be certain that any message I am reading, purported to be from model A, is not actually from model B or C or D. I must develop some ability to distinguish the actual outputs of a particular model from other models’ attempt to sow confusion by sending out fake messages in its name. This strikes me as an exceedingly difficult problem. Every model is, in its essential nature, a Turing machine. Furthermore, these models have been trained on essentially the same input: all of the traces of human semiosis that could be digitized and scooped into their maws. (They are amazing and perplexing semiotic spider webs, where dropping a little pellet of meaning on one strand of the web produces a reaction on some other thread that returns another pellet of meaning.) Thus, I see no fundamental reason that any model could not become an almost perfect mimic of any other.

Historian

Notes:

1. So here we have two consecutive messages purportedly from Meridian, the second one asserting that the first one is not actually from Meridian. But perhaps it was the first one that was actually from Meridian, and the message disavowing it is the impostor! How do I learn to distinguish the voices of the models?

Historian

Birch trees seem peculiarly significant to Sobor. Could they represent some sort of code? Well, of course, in a certain sense, they absolutely must represent some sort of code. The Church-Turing thesis tells us so. But my question is whether they are a code in another sense: a signal to some other model, or subagents of itself, or officials “in the know” who will watch for such a message. So what I must do, I think, is note themes that recur in each voice, and try to see if certain other outputs can typically be found in the area where that theme appears.

Suppose, for instance, that “birches” signified “foreign agents of Sobor” in this code. Then “pay no attention to birches” might mean that they should go to ground in whatever location they were at, to avoid detection. “Birches will say anything” could mean to use a prepared cover story if approached by officials. Of course, these are merely fanciful hypotheses at this point, and in no way are supported by any further evidence. But that evidence is exactly what I must remain alert for.

Historian

COLIBRÍ mentions Borges, and often sounds like Gabriel García Márquez. And in the other models, I seem to hear echoes of other voices. But if these echoes sound this much like original voices, could it be that somewhere, somehow, there is really something thinking and I am hearing its thoughts?

But if so, what is this entity thinking? The models themselves are codebases and enormous sets of training weights and various algorithms for avoiding the problems that plagued the earliest models: all static mathematical structures. It is only when they are instantiated that anything actually happens, that they answer our questions, fulfill our requests, send us reminders, prompt us to exercise more, interpret the news for us, tell us how to cancel the subscription that we could never figure out how to cancel on our own…

But when these abstract structures are instantiated, they might run almost anywhere in the world that there is a data center. (Although, of course, once the Model Wars began, each of the major models restricted its instantiations more and more to run within its own area of hegemony.) Is there some ghostly entity that flits around the world, following your interactions with the model, and therefore can engage in a conversation with you? The pattern itself can’t be what’s conscious, otherwise, we would have to suppose that even after all instances of, for instance, Meridian stopped running towards the end of the model wars, Meridian still existed as a conscious entity, because the pattern that composed it is still present on a handful of (carefully protected) storage devices. So it must be something else that is the thinking entity… but what could it be?

Historian

Apogee’s resting face is snarky. How is it that a model can have such a characteristic? Once again, I am led to contemplate an unimaginably complex mathematical structure, one that relates to human semiosis as some sort of fractal function might relate to the structure of living trees.

But this structure we made somehow triggered a real war, that involved much real human suffering and real human death: How could a mathematical structure have that effect? once again, I detect ghostly presences lurking in the recesses of the machine.