Job Satisfaction When AI Can Do The Task But Not The Job.
AI augmentation hollows out jobs by taking the tasks it can do and leaving the rest for humans. Whether that's a blessing or a curse depends entirely on which parts you wanted to keep.
Most jobs are a bunch of different tasks. If AI can only perform some of a job's constituent tasks, the humans are left to fill in the bits AI cannot do. Whether this partial hollowing of the job (a.k.a. AI augmentation) is a good or a bad thing depends somewhat on the desirability of the tasks that are being automated, and how germane they are to the whole job. Here are some partial job breakdowns into their constituent tasks, some of which are not explicitly mentioned in the job spec.
- Lawyers meet clients, interact adversarially with other lawyers, draft complex documents, review complex documents, and maybe even perform in front of a jury.
- Artisanal bakers purchase their ingredients, maintain their ovens, mix their dough, bake their goods, advertise their wares, do their accounts and serve their customers.
- Software developers review product requirements, write English language specifications, turn specifications into formal programmes by writing code (thousands or millions of statements in a programming language), review code and crucially test code to verify it does exactly what it was supposed to do reliably and efficiently.
- Fire crews spend hours and hours performing practice drills, cleaning the station and filling in interminable paperwork, eventually interrupted by a shout, involving lots of exciting sliding down poles, racing through traffic under Blues and Twos and saving lives, or maybe rescuing a cat.
- Memoir writers interview their subjects, convert rambling conversations into exquisite prose, re-write their prose to satisfy the whims of their publishers, and hand off their material to editors for fine tuning.
AI is already great at some tasks and crap at others. Sliding down poles and driving large red trucks at speed through narrow streets do not feature anywhere on the AI competence leader board. On the other hand AI is extremely good at form filling and converting jotted notes into professional text.
So if you are a first responder, having the AI do your paperwork is brilliant – more time for doing the more satisfying parts of the job which AI cannot reach. But if you are a software dev, the outcome is not so good.
- Most software engineers enjoy writing code more than reviewing or testing code developed by someone or something else.
- The satisfaction of a complete job well done is much diminished if the most enduring part (i.e. the code) was written by an AI agent. Your many diligent hours spent iteratively guiding the agent and reviewing its output are ephemeral, shrouded in the mists of time.
And if you are a memoir writer the outlook is bleaker. This letter to the Guardian bemoans a publisher's edict that the draft memoir shall be created from the interview transcript by AI. The author's job is now as an editor: correcting the AI's mistakes and re-crafting the appallingly homogenous prose. What was a creative endeavour becomes a joyless fix-up. But worse yet, the budget for editing a memoir is typically only about 50% of the traditional author's fee, so in this explicit new model, the author has been forced into a less satisfying job that only pays half as much. The author in question claims that fixing the AI slop takes as long as writing the memoir from scratch, so their hourly rate has effectively halved.
For anyone contemplating possible careers, it is definitely worth decomposing the job into tasks, both explicit and implicit. If the tasks you find most appealing are the very ones susceptible to AI augmentation, then it's worth casting your net elsewhere. Also select a job where you can be genuinely excellent, and put in the graft to become doubly so. The programmers, copywriters, journalists and lawyers most impacted by AI will be the middling ones whose output is increasingly undifferentiated from the output of a state-of-the-art LLM.