On machines

I want a technology limiter I can’t out-think. Every system I build to constrain myself fails the same way: I built it, so I can dismantle it. A constraint that survives contact with its own designer may be impossible. That’s what makes it interesting.

There is a whole genre of stock imagery for artificial intelligence: glowing blue brains, circuit-board synapses, a robot finger reaching toward a human one. None of it is made by anyone who has used the thing. It is made for the procurement officer who was told to buy AI and needs to know it on sight, a sign pointing only at other signs, so the buyer can say, found it.

We call the stored states of a computer its “memory,” and the word forecloses the thought before it starts. Human memory is associative, relational, lossy, nothing like a drawer of notecards pulled out intact. By naming the filing cabinet “memory,” we lose the ability to ask what memory actually is, and then build every machine in the shape of the wrong metaphor.

In the warehouses, robots move twelve hundred packages an hour and the humans fill whatever gaps the robots can’t. The inversion is complete: the machine does the work, and the person does what the machine finds inconvenient.

To what extent am I becoming an API? I collect and structure information about my own life so that systems can use it. At what point does the structuring start to change the thing being structured?