Manic technology

The opening sec­tion of this post from Dean W. Ball ticks off a list of projects he’s accom­plished in the past month using AI coding com­pan­ions. The volume is intended to be striking, and it is, but/and by the fourth or fifth bullet (of thirteen) it’s also … A LOT.

I have sev­eral friends who seem to have entered the same space. Of their AI coding com­pan­ions, each has said some ver­sion of: “It never gets tired of talking to me!” You can see the ways in which this is great — refreshing — and you can also see the ways in which it might cause problems.

I’m starting to think lan­guage models are a fun­da­men­tally manic technology, in part because they operate exclu­sively through logorrhea, the “yeah, yeah, YEAH!” of the all-nighter.

If my assess­ment is true, it’s good news for the busi­ness of AI: cap­i­talism loves mania; it loves caffeine, all the amphetamines; it loves urgent possibility; it loves solving the problem of “too much X” with “even more Y”. I don’t intend that in any par­tic­u­larly snarky way; just a plain his­tor­ical observation.

Mania is another word for “bubble”, of course, just as we use “depres­sion” for eco­nomic as well as psy­cho­log­ical contraction. I’m par­tial to the over­ar­ching theory of psy­cho­log­ical mania and depres­sion as a change in the brain’s eager­ness to con­duct signals: the manic brain lights up too easily, while the depressed brain is too reluctant. That’s basi­cally how econ­o­mists see (inflationary) manias and (deflationary) depres­sions, too: dis­rup­tion of, or diver­gence from, a real economy’s “ideal” output.

The “ideal” set­ting for a brain (or an economy?) isn’t nec­es­sarily straight down the middle. A dip into the realm of mania can be useful, some­times revelatory. I don’t know if many cre­ative projects would ever get started if our brains didn’t some­times relax the stan­dards by which they light up.

Yet for a human mind and a human heart, one really good project is more nour­ishing than ten cruddy ones; that was true a hun­dred years ago, and it’s true today. The AI coding com­pan­ions will never ever say: “Hey … whatever hap­pened to that other thing you were working on?”

I sup­pose you still need friends for that, people who know you, who know when you’re talking too fast, and the gleam in your eye has taken on a hard edge.