Building memory maps and doing register allocation by hand is excellent training for becoming a computer programmer, but nobody's done it since compilers become decent.

Well, people who work on video codecs, low-level crypto code, and compilers themselves do manual register allocation when they write platform-specific assembler stubs, but yeah, it's rare.

You don't learn anything through menial document searching and reading other than menial document searching and reading.

Junior lawyers will still have plenty to do.

If it's so menial and you "don't learn anything", then why not just pay people off of the street to do it?

I think you're heavily discounting the fact that searching through case law is not just "type in words into LexisNexis", but it's searching, parsing and consolidating case law that is relevant for a Partner. You actually read the case law, spend time understanding it, and build a report around it.

If you implemented the same compression algorithm over and over again, each time from scratch, the only skill you'd learn would be how to do it more efficiently. You wouldn't learn anything special, except maybe some nuances of the algorithm you didn't already know.

Yeah it's not a skill that everyone can do, but once trained to do it there's no point only doing that same task over and over because you won't advance your skills in other aspects of the field.

That's a flawed analogy. It's more like implementing compression of different data under different conditions every time, which would train you to get much better with compression and programming in general.

Every case has its own unique characteristics. It's not just doing the same thing over and over again.

You can literally say this for every profession in existence.

Comparing a simple compression algorithm to extensive law research is extremely disingenuous and shows a lack of awareness of what it takes to actually BE competent as a lawyer.

Just because it's not as difficult/involved as being a lawyer, while also not training you how to be a good lawyer, doesn't mean that it doesn't still require an education. An education to be able to understand the documentation and law you are working with.

But at the same time not all of them require law degrees either, many para-legals and helpers aren't full lawyers.

True. But this will reduce a lot of demand for human legal services, since the few humans left will be more effective per hour. Unfortunately the pipeline to produce senior lawyers is packed as if AI isn’t coming, and for that cohort life is going to suck a bit.

On the other hand, maybe the cost to review an employment agreement will enter into the sub-$100 range. Or to draft an NDA for a startup, etc.

The path from passing the bar to being a successful / well-paid attorney has been extremely difficult for a decade now. AI will simply continue this trend.