Defense against AI-induced cognitive rot

AI is going to abrade how people acquire knowledge, skill and experience. I share that worry, and I’m especially curious how top-tier law firms will raise a new crop of partners who can work at the highest levels of the profession. By “highest levels” I mean the people whose work, for good or ill, shapes the law in high-stakes cases. For every David who slays a Goliath, or Judith who slays Holofernes, there are a multitude of lawyers who are not even almost famous. In high-stakes litigation, mergers and acquisitions, and lobbyist-driven legislation, our legal landscape is shaped by their work.

The firms’ problem is how to manufacture the next generation. Traditionally the answer was leverage. Clients paid for baby associates to do a ton of grunt work. After you’ve read versions of the same documents over and over, you begin to get it, and you move on to drafting new versions of the same documents over and over. Through it all you get a steady diet of negative feedback. After about seven years of that, if you’re also good at rainmaking, you make partner. Otherwise you become “senior counsel,” with an internal market called “eat what you kill,” or you go elsewhere, often in-house.

Clients are tumbling to the fact that much of what they’re paying hundreds of dollars an hour for is now available for AI tokens. As in-house counsel I once got a bill for a cookie-cutter transaction for work on the principal document, which differed from the last one by fewer than 200 words. As the total bill was reasonable, I didn’t gripe about the excess hours on that line item. A colleague of mine did a page turn on every redlined document of every pattern transaction, from the cover, past the signature page, to the last page of the last exhibit. It’s how he was trained. It was not a useful trait.

So if rubbing their noses in the, uh, stuff until the beatings stop is no longer an option, what replaces it? I don’t know.

I do have a view about those who already have the expertise, lawyers especially: the questions matter more than the answers. Far more.

Answers in the real world are contingent. In almost everything we do from day to day, there’s no Newtonian clockwork that tells us what will or should happen next. Experts in any domain know this, and you can hear it when they think aloud: “What if we change this?” Large language models are very good at taking the percentage shot, and they’ve become very good at playing to our cognitive biases. We’re flattered, confirmed, and treated to a Glengarry Glen Ross standard of confidence. And we’re implicitly asked to accept that the average opinion of average opinions is the right answer.

Who is immune to that? Children. “Why, Mommy?” followed by “But why?” We seem to have an inborn instinct for driving toward first causes. Aristotle’s Metaphysics reaches for an unmoved mover, and his Nicomachean Ethics asks what is it that we pursue for its own sake rather than as a means to something else.

You can overdo it. Regress far enough and you’re asking what came before the Big Bang, which helps no one. That’s the force of Justice Holmes’s line: “The life of the law has not been logic: it has been experience.”

In programming, I’ve found that focusing on the what, not the how, is the best path. It’s school algebra writ large: I have $𝑥$ and I want $𝑦$. What functions do I apply to $𝑥$ to get $𝑦$? That usually means breaking $𝑥$ and $𝑦$ into their parts and composing a function $𝑦=𝑓\left(𝑥\right)$ to fit in baby steps at a time. It has saved me countless hours.

My prescription for keeping AI from stealing your ability to think is simple. Stay curious. Ask why. Don’t take the first plausible answer. The cost is some cognitive discomfort. The rewards are great.