Some things are easier to automate than others, and while the automated process is generally cheaper to perform than the not automated equivalent, it usually requires some up front investment. This leads to a disparity in what tasks get automated.
One classic example used in future of work discussions is: how much more than you currently earn would someone have to pay you to shovel poop all day? Given that number is likely reflective of general sentiment, why are people who shovel poop paid so little?
There is rule of thumb that automatibility is related to how instruction-driven a task is. There is some truth to that - one of the reasons poop shovelling is poorly rewarded is that basically anyone can do it, and explaining what to do is simple. However, it certainly isn’t easily automatable.
Rodney Brooks users a counter example, to support his idea that the threat to jobs is often digitisation, not automation itself. His example is toll talking at a bridge. Automating that is hard - you’d need a robot that can accept and return cash, evaluate what it had been given, and so on. In reality, toll taking is being replaced with small transceivers - they’re not automating the task at all, but providing an easily machine readable identifier, that can then be linked to a payment system.
Again though, digitisation requires some up front investment. The consequence of these trends is that the type of work that doesn’t get automated is either:
-
High value, specialised, changeable work
-
Low value, low volume, difficult to digitise work
The first is akin to the type of work done in a workshop versus that of a factory. Anything that needs enough variety, or bringing together a really wide range of approaches will lean towards human work. That said, it will likely also lean towards automation-aided: a good example is the modern tooling around writing software, a very carried task.
The other case is work that is hard to digitise, and that is too low value to automate. One common surprise for people contracting a manufacturer in China is how much assembly is actually manual - it can be far cheaper to hire a set of people to stitch a complex pattern together than tool a machine for it, even if that machine can operate at a much higher rate.
There is a kind of priority inversion that can occur with this, where the human time ends up invested in the least valuable, and often least pleasant, tasks because those are the least amenable to automation.
The structural employment issue that arises is if the decision maker is not the person doing that work. If the owner of capital invests in automation it looks very different than an individual themselves doing so. Even then, it’s quite possible to do this to yourself within your own workflow, and end up spending more time on the worst part of your job. Optimising for current results can lead in some suboptimal outcomes.