The original note, as written at the time
Look at it this way. We might have found a way to measure the commensurate input from various human jobs, especially creative ones. All you have to do is compare the proportional amount of energy and resources that an AI tool has to utilise to come up with similar output.
Ongoing development
We might have found a way to measure the commensurate input of various human jobs, especially creative ones: compare the proportional amount of energy and resources an AI tool has to use to produce similar output.
I think this is the most economically important idea in the collection, and it is also the most dangerous, and both for the same reason — it is measurable.
The intuition is good. If a machine needs a great deal of computation to produce a thing, the thing is hard, and difficulty is the closest available proxy for value. A brief that a model can produce in a second is a brief nobody should be paid a week for. A design that takes the machine thousands of attempts is a design that contains a real problem.
But the measure is the machine’s effort, not the human’s, and the two diverge exactly where it matters. A photograph that took a person twenty years of practice to see can be reproduced by a model cheaply, because the practice is not in the image. The computation measures the difficulty of the output, and most of the value of creative work is in the judgement about what to output.
There is a second problem: an energy price is a price the platform sets. Whoever owns the models decides what counts as expensive, which means the benchmark of human worth would be set by the party with an interest in setting it low.
Used honestly, this is not a valuation tool. It is a floor. Anything cheap for a machine to make is cheap, and a profession that consists of it should know that. Anything expensive is a place where human judgement still has weight.
The danger is not the measurement. It is the assumption that price follows from it.