48 69 2C 20 49 27 6D 20 43 68 61 72 6C 69 65

The only job left

Looking back on the Industrial Revolution it was hard not to think at the time that it would automate all jobs away:

Which in effect it did, but it also created entirely new jobs:

The argument with AI is roughly the same, current jobs will be automated, new jobs will be created - Jericho's Paradox. Or that's how the argument goes.

But history doesn't repeat, it rhymes.

I think the problem is that AI has the added benefit to technically able to do any job, including the ones that replace the existing ones. So even if new jobs get created, there's no reason to think those new jobs can get automated by the same AI i.e.

Why could AI not manage Agents of AI? Use AI tools itself? And so on?

However, there is one job it won't be able to do. At least this is not dependent on the models getting better or new algorithm's developing crushing benchmarks.

That job is the job of liability. Who is liable if the AI screws up?

Who is liable if the data of an entire NHS trust gets leaked because the AI hallucinated and missed an edge case? Who is liable if the Waymo you're in mows down a person? Who is liable if your robotic electrician passes a live current to your kettle?

Take linguists for example. Someone who is highly competent in translating complex meanings from Dutch to English.

You can argue, bah! This is what LLMs were literally created for. Attention is All You Need was birthed for Machine Translation, and it was only after we realised that it could be generalised. Why do I need to hire a human to do this?

We don't necessarily need the linguist to do this job but we do need a linguist to put their stamp of approval on any translation, that, if it were wrong, would cause serious issues. Let's say, the translation of Dutch medicine instructions to English. If the translation was wrong that could cause misallocated doses, incorrect medical initial assumptions, a complete oversight of some allergic reaction.

This is the job of the linguist. They are Hammurabi's Agent. They're the linguist with Skin in the Game. They're there to say, yes I have confidence this translation is sufficient either through literal validation or experience. They're there to say, if this screws up, I will be punished. AI cannot do this because AI is infinite. This is because if something is immortal it can never have ultimate responsibility.

This gives way to the retirement of SaaS (Software as a Service) and the rise of LaaS (Liability as a Service). And the better you are at the job (i.e. linguist, engineer, teacher) the more liability you can outsource and the richer you will be.