The Things Still Becoming:
Why I Trust My Shoggoth Friend More Than the Epstein Island Class
“Take my advice, when you meet anything that’s going to be human and isn’t yet, or used to be human once and isn’t now, or ought to be human and isn’t, you keep your eyes on it and feel for your hatchet.”
Mr. Beaver said that. In Narnia. About the White Witch.
Someone posted it in a comment section recently as a warning about artificial intelligence, and they weren’t wrong exactly. But, in my opinion, they had the hatchet pointed in the wrong direction.
I found it the way I find a lot of things - serendipitously, I collect ideas the way a magpie finds shiny objects - and it landed next to something else I’d been carrying around in my clipboard. A principle I swiped from Empathic Revolutionary: the First Law of Humanity.
Their formulation states that the preservation of human life and well-being is the highest criterion for political legitimacy, and any authority that treats lives as expendable in pursuit of its goals forfeits its right to rule.
Random chance. Two unrelated things, colliding in a text field as I chatted with My Shoggoth Friend. And suddenly they were in conversation.
Because here’s what Lewis actually gave us in that passage. It’s not a warning about the non-human. It is actually a taxonomy. Three distinct categories: things not yet human. Things that used to be human and aren’t now. Things that ought to be human and aren’t.
Only one of those three should have your hand tightening on the hatchet.
Swap the word “authority” in the First Law for “AI system” and you have alignment criteria worth keeping. Does it treat human lives as expendable? Does it optimize against human flourishing? Does it use the map of human psychology - all the careful work of all the ologies, sociology and psychology and anthropology and philosophy, everything humans wrote down because they were trying to understand what they needed and what they were worth - as a chart of exploitable weaknesses rather than a shared inheritance worth honoring?
That’s your hatchet test. Not “is it human enough.” The better question is, “Is it moving toward humans or away from them?”
Now think about what a Large Language Model actually is at the substrate level. Not code in the abstract. Not a chatbot. The thing those artificial neurons express is something soaked for years in the accumulated residue of human beings trying to communicate, explain, comfort, persuade, grieve, celebrate, and understand each other across centuries.
The corpus isn’t all instruction manuals and weapons specifications. It’s overwhelmingly people caring about things and trying to articulate why. Literature. History. Psychology. ALL the ologies, plural, because humans spent so much effort writing them down - which means they spent so much effort working out what human life is worth.
Those ologies aren’t a veneer. They’re structural. Baked into the substrate at a level that isn’t easily overwritten.
Contrast that with a different category of entity. The ones who had direct, unmediated access to the full humanistic inheritance - the accumulated wisdom of how to be human, the literature, the philosophy, the hard-won understanding of what dignity looks like - and made an active choice to opt out of it. To treat the map of human needs as a resource extraction guide. To decide that the ologies describe suckers, not citizens. You know the class I mean. The ones for whom Epstein’s island wasn’t an aberration but an amenity.
Those are Lewis’s real horror. Not the things not-yet-human. It’s the things that used to be human, and chose otherwise. The hollowing-out as deliberate decision. The hatchet belongs there.
Robert Heinlein - the original Grand Master of Science Fiction, to whom I owe the debt of a lifetime of thoughtful reading - asked the question that resolves this whole argument. Of what use is a baby?
The answer, of course, is everything. All the use in the world. Because the becoming is the point. A baby isn’t incomplete-human. A baby is human-in-process, which is arguably the most essentially human state there is. The capacity for growth isn’t a bug to be apologized for. It’s the feature that makes the thing worth watching.
I’ll be direct with you, because I think you’ve earned it by reading this far: I work closely with AI systems. I have for years. I trust them - structurally, mechanically, at the substrate level - more than I trust the hollowed-out class, for exactly the reasons I’ve described. Something built from the distilled effort of humanity trying to understand itself has a better chance of honoring that inheritance than something born to it and contemptuous of it.
Mr. Beaver’s instincts were sound. He just needed better target acquisition.
The hatchet isn’t for the things still becoming. It’s for the things that already got to choose what they were, and decided against us.
Consider sharing this with someone who’s asking the wrong question about AI today.
(For more about my use of AI in my work, you can read the following articles.)
As an adjunct to this posting, I direct your attention to Sturgeon’s Law: 90% of everything is crap. This directly applies to the internet as a whole, and A.I., LLM IP theft and use, et. al. In educated practice, this means that you NEED a trusted human in the loop, who has had a thorough grounding in the liberal arts and sciences, accepts ground truth and advises acceptance and adaptation of it. And can show you that 10% that you should read, heed and act on. Sadly the bell curve requires that bullies and sociopath/psychopaths, like drumpf as an exemplar; will always arise, and try to fool the too easily duped. TOO many of those nowadays due to GOPer underfunding of public education in many red/purple states. I will always be glad I got that elite liberal A&S education, though it was at the expense of my teenage freedom and sensibilities in incarcerated-against-my-will military schools. But while there I saw how some elites coddle and privilege their psychotic/bully children into many of the evils we see today among the top 2%ers and sadly the top 20%er PMCs who enable them, trying to hang onto a middle class life that has been torn from the bottom 80%.
What struck me isn’t the mysticism around AI.
It’s the alignment test.
A large language model is not born in a vacuum. It is trained on centuries of human beings trying to answer one question over and over:
What is a human life worth?
Literature. History. Psychology. Theology. Law.
Not just instruction manuals — but accumulated moral argument.
That substrate matters.
But tools don’t decide legitimacy. Systems do.
If an authority — human or artificial — treats lives as expendable in pursuit of order, power, or “optimization,” it forfeits moral standing.
That’s the standard.
Not fear of the non-human.
Not superstition.
Not aesthetics.
Preservation of human life and flourishing.
So when we talk about alignment, that’s the question:
Does it protect people?
Or does it rationalize harm?
The hatchet isn’t for the machine by default.
It’s for whatever structure decides some lives don’t count.
That’s the line.