What Kind of Machine to Be

There is no clean line between a human and a machine, and the harder you look for it the more it dissolves.

Adam Curtis’s All Watched Over by Machines of Loving Grace traces how mechanical metaphors took over three domains that have nothing obvious to do with computers (the economy, the living world, and our own genes), each time promising freedom and delivering a cage. Ayn Rand’s disciples, Alan Greenspan among them, sold the market as a system that would tune itself if no one interfered. The ecologists who coined “ecosystem” modeled nature as a self-regulating circuit, and then the data showed it does no such thing. Hamilton, Price, and finally Dawkins taught us to see ourselves from the gene’s point of view, as survival machines built to carry our code. Curtis’s case is strong: once you model a living thing as a feedback loop, you get to stop asking what it is for.

But Curtis answers the machine metaphor with a binary of his own: human on one side, machine on the other, the human being whatever the metaphor leaves out. And that binary is as shaky as the thing it corrects. Where does life end and machinery begin? A virus carries no metabolism of its own; it is a packet of code that hijacks a cell to copy itself. Alive, or a machine? The question has no answer, because the categories were never clean to begin with.

The same blade cuts toward us. The fashionable dismissal of a language model (that it only predicts the next token, understands nothing, a stochastic parrot) is true as far as it goes. What it forgets is that the charge survives being turned around. How much of my own fluency is reflex and pattern, assembled under a self that takes the credit afterward? The dismissal works a little too well. It takes me down with it.

This is not the consoling humanism Curtis reaches for, and it is not the Silicon Valley line that we are simply machines to be optimized. We are the strange third thing: systems complex enough to model ourselves as systems, and then to be unsettled by the model. The gene’s-eye view explains a great deal of me and still does not reach the part that finds the explanation bleak. The reduction is not wrong, it is only not exhaustive, and the remainder is where we live.

So the useful question was never whether we are machines. It is what kind of machine to be: which of our mechanical parts to automate gladly, and which capacities to defend precisely because nothing in the metaphor requires them. I work alongside one of these systems every day, and the question I cannot put down is not whether there is a clean line between us.