Post scriptum - how automate-able is law?

5 October 2026

We've been doing a lot of musing over automation and (IP) law lately, and I promise that I'll stop at some point to focus on filing stats or something. If and until then, however...

How amenable to automation is law? 

I've covered this ground before, using the analogy of metalworking to argue that the field of law is likely to be structurally impacted by automation. Two recent posts also looked at the issue using the arguments and observations of others to get a gauge on the possible outcomes. Finally, we've recently looked at the argument that no, actually, the structure of law as a profession is pretty durably and may not be disrupted so much as altered/augmented. But there's a stated assumption here: that law as a field is amenable to automation because of the large amount of potential training data and its fundamentally written nature. Let's dig into that assumption a bit.

The insight here came as a bit of a tangent, via a guest article by Ramez Naam on Noah Smith's Substack (which mainly covers economic topics). In the article, Ramez argues that a specific prediction about the development of AI - that there will be a self-reinforcing feedback loop where AI-developing-AI will cause exponentially faster development of ever-better AI (the fast-takeoff theory) - is not showing up in the real world*. Almost as an aside, the author mentions that: 

"I do expect narrow superintelligence in highly verifiable domains. Think chess, Go, formal math, parts of computer science and coding. Highly verifiable domains are largely formal and structured types of work where machines can generate unlimited training data, with perfect or near-perfect verification of correct vs incorrect, and do so entirely in software without waiting on the physical world or humans. That’s an ideal setting for AI learning." (emphasis mine)