Post

@grimalkina The mindless tasks can be automated by deterministic processes (which can include ML!).

GenAI on the other hand implies a significant validation burden for any nontrivial task, which can also be exhausting. I'm not so sure the equivalency holds.

@grimalkina yaaaas I think the mistake is of thinking as "automation" as replacement. When all these success are more than just replacing the human, it is about *rethinking the way work is done* and ofc, the place and type of work we consider acceptable for humans.

@grimalkina @jilleduffy I think the problem for me is that what LLM programming tools in particular seem to create is work that is actually harder.

Like, people talk about it as a solution for ‘boilerplate’ code, and it’s definitely true that the task of e.g. repeated copy-and-paste with minor changes each time can introduce errors; but my perception is that the LLM replaces the copy-and-paste and just leaves behind the ‘vigilance’ part, of making sure that pages full of nearly-identical code don’t have subtle errors.

So it’s not the sort of knowledge work that lets you exercise your intellect or whatever, but just a different sort of mind-numbing repetitive task.

@grimalkina Aye, class related blindspots aside, we need people to reflect on the detail and nature of the different types of work they do to make good decisions about what to automate (in their context). To flip this on its head slightly, one of the prompts I've been using is asking folk to consider where there is value in the process rather than just in the output.. Where there isn't, automate away. It's not a simple task and likely needs revisited as a practice.

@grimalkina Hm, this comparison evokes the opposite reaction for me. AIUI, a big reason those factory jobs are mentally and physically draining is precisely because they've had lots of slow bits optimized out of them by Taylor and Gilbreth and the like. The pre-factory days of craft were not as draining. (I'm not an expert; I have no studies to cite, so apply salt as needed.)

The kind of complex work that is good for us isn't necessarily _difficult_, as much as it is varying. It's the repetitive part that destroys us. My theory is this is why medieval hand copied manuscripts have the art in it.

But we can thank Henry Ford for that optimization, and whoever else for having turned it to knowledge work, too.
@grimalkina

@grimalkina Perhaps it's the definition of "hard work"? Because repetitive mindless tasks are the hardest for me.

@grimalkina @jilleduffy

Yeah, I’ve heard far more often what is roughly the opposite argument — “I wanted a robot to do the dishes for me so I could make art, not a robot to make the art for me” — and that feels like it resonates more

@grimalkina Thank you. I think probably I should have made a distinction between automating factory work versus knowledge work, although you still have a point that there is some amount of automation that is of course useful. Templates are useful, for example.

The "hype" to me around AI says "figure out how to automate as much as you can with AI or you will be left behind," and I just don't think that takes into perspective how most knowledge workers actually work