“We can only see a short distance ahead, but we can see plenty there that needs to be done.”
— Turing

I love the “cannot” stances. There’s always some naysayers who draw lines in the sand as if exploration bows to their whims. I guess it’s easier for some to zero in on clipping the wings of possibility than to watch others use them to soar.

But I don’t want to blame. The claim I’m trying to make is more of an observation shared between minds of all stripes, and it’s one that I find myself returning to again and again — computation is in the eye of the beholder. And just as that eye relies on being struck by rays of light to function, computation serves its functional role with equal reliance on the observer. It’s our modern ultra-accelerationist categorical imperative — to expand computational possibility unconditionally (crucify me for smuggling this value claim) because the observer-relative nature of computation is the property that licenses open-endedness. A rock isn’t “computing” unless somebody interprets it as such.

Let me expand. Observer relativity doesn’t mean equal validity — not “anything goes.” The space of interpretations is open-ended and that doesn’t mean at all they are equally useful. Good. Now, functional correctness is a real constraint on that space. By this I mean that the rock’s physical state transitions exist objectively, but an observer who says “this rock is computing a hash function” must show proof-of-work for that functional mapping between the rock’s physical states and hash function’s domain/co-domain. Moreover, an observer who just says “this rock is computing… something” has a weaker claim than the one who demonstrates a precise isomorphism. I believe that better and worse interpretations do exist and they are constrained by how well the mapping holds up. Now, to tie it up together, the normative stance on exploration is because open-endedness gives you a vast space of possible interpretations and functionalism provides a ranking criterion. I will have to write more about this (I probably won’t), especially the ranking criterion. As presented here, the words are at worst unintelligible, but that’s already a pretty high standard compared to what’s out there.

What stems from this belief is a potentially unbounded future, an anisotropic positive divergence spreading like wildfire. LLMs are such a visible spot fire, and hopefully, yet another stepping stone toward technological empowerment. It appears that mastering language is the skeleton key unlocking binding between other modalities. It’s the Rosetta Stone of AI. Vann McGee had made a compelling argument — verbal reasoning, when you strip it down, thrives on the rules we construct. He wrote that,

“Recursion theory is concerned with problems that can be solved by following a rule or a system of rules. Linguistic conventions, in particular, are rules for the use of a language, and so human language is the sort of rule-governed behavior to which recursion theory applies. Thus, if, as seems likely, an English-speaking child learns a system of rules that enable her to tell which strings of English words are English sentences, then the set of English sentences has to be a decidable set. This observation puts nontrivial constraints upon what the grammar of a natural language can look like. As Wittgenstein never tired of pointing out, when we learn the meaning of a word, we learn how to use the word. That is, we learn a rule that governs the word’s use.”

But language is a game of our own making. And who says we can’t reform it, bend it until it snaps? Language is alive, pulsing with the potential we inject into it. As Max Bennett notes,

“We are capable of puppeteering other minds because language is, it seems, built right on top of a direct window to our inner simulation.”

To view these patterns as mere auto-completion is shortsighted. It anchors you, of course, but it shackles you to a reality devoid of wonder. It’s like knowing we can’t fly and using that as an excuse to never find ways to reach for the sky. And yet, that’s precisely what they are. And still, they amount to more than that. They’re conditioned by us, both when they learn and when they respond. They capture subtle nuances and complex ways in which we weave meaning into words. Think of it in terms of conditional probability: beyond predicting the next word, they also predict us. And due to the sampling process, sometimes statistics satisfy our unspoken thoughts. We teach the model that a question has a certain set of answers, and we call that alignment. In the future, we ask the model which set of questions lead to that answer. The Model is the Message.

But AI is much broader than that. As an interlude, have you ever mused on why it’s so hard to imagine, but so easy to dream? Don’t dismiss it as vagueness, and for once, stop and think about it. This question invaded my mind when I first read Blindsight, more specifically, Siri’s monologue, over and over again, for good minutes. So, if you’re anything like me, you’d think the answer is hidden in that monologue, reproduced in full below, because it’s so good:

"I am the bridge between the bleeding edge and the dead center.

I stand between the Wizard of Oz and the man behind the curtain.

I am the curtain.

I am not an entirely new breed. My roots reach back to the dawn of civilization but those precursors served a different function, a less honorable one. They only greased the wheels of social stability; they would sugarcoat unpleasant truths, or inflate imaginary bogeymen for political expedience. They were vital enough in their way. Not even the most heavily-armed police state can exert brute force on all of its citizens all of the time. Meme management is so much subtler; the rose-tinted refraction of perceived reality, the contagious fear of threatening alternatives. There have always been those tasked with the rotation of informational topologies, but throughout most of history they had little to do with increasing its clarity.

The new Millennium changed all that. We've surpassed ourselves now, we're exploring terrain beyond the limits of merely human understanding. Sometimes its contours, even in conventional space, are just too intricate for our brains to track; other times its very axes extend into dimensions inconceivable to minds built to fuck and fight on some prehistoric grassland. So many things constrain us, from so many directions. The most altruistic and sustainable philosophies fail before the brute brain-stem imperative of self-interest. Subtle and elegant equations predict the behavior of the quantum world, but none can explain it. After four thousand years we can't even prove that reality exists beyond the mind of the first-person dreamer. We have such need of intellects greater than our own.

But we're not very good at building them. The forced matings of minds and electrons succeed and fail with equal spectacle. Our hybrids become as brilliant as savants, and as autistic. We graft people to prosthetics, make their overloaded motor strips juggle meat and machinery, and shake our heads when their fingers twitch and their tongues stutter. Computers bootstrap their own offspring, grow so wise and incomprehensible that their communiqués assume the hallmarks of dementia: unfocused and irrelevant to the barely-intelligent creatures left behind.

And when your surpassing creations find the answers you asked for, you can't understand their analysis and you can't verify their answers. You have to take their word on faith ---

Or you use information theory to flatten it for you, to squash the tesseract into two dimensions and the Klein bottle into three, to simplify reality and pray to whatever Gods survived the millennium that your honorable twisting of the truth hasn't ruptured any of its load-bearing pylons. You hire people like me; the crossbred progeny of profilers and proof assistants and information theorists.

In formal settings you'd call me Synthesist. On the street you call me jargonaut or poppy. If you're one of those savants whose hard-won truths are being bastardized and lobotomized for powerful know-nothings interested only in market share, you might call me a mole or a chaperone.

If you're Isaac Szpindel you'd call me commissar, and while the jibe would be a friendly one, it would also be more than that.

I've never convinced myself that we made the right choice. I can cite the usual justifications in my sleep, talk endlessly about the rotational topology of information and the irrelevance of semantic comprehension. But after all the words, I'm still not sure. I don't know if anyone else is, either. Maybe it's just some grand consensual con, marks and players all in league. We won't admit that our creations are beyond us; they may speak in tongues, but our priests can read those signs. Gods leave their algorithms carved into the mountainside but it's just lil ol' me bringing the tablets down to the masses, and I don't threaten anyone.

Maybe the Singularity happened years ago. We just don't want to admit we were left behind."

Some say we are slowly becoming feature vectors, being squashed and reduced to data points, our human essence distilled into machine-readable patterns, finely cooked only to end up served up and devoured by the algorithmic gods. Perhaps. However,

“Reality overstepping the boundaries of comfortable vocabulary is the start, not the end, of the conversation.”