Swapping Human Labor for AI: A Critical Look
September 30, 2023Came across a medium article titled “4 Questions to ask before swapping human labor for AI”. I am not sure if I agree with the base discussion of the article, that one should consider a direct swap of human labor for AI, but it is a good starting point for a discussion.
So the 4 questions were:
- How much time does the task require without AI assistance?
- How highly paid are the people who perform this task?
- How capable is the AI of completing the task correctly?
- How easy is it for humans to determine whether the AI output is accurate?
So let’s take a look at each of these questions.
1. How much time does the task require without AI assistance?
If time is the only measure this is probably a good question to ask. Amdahls Law reminds us that speeding up a small task doesn’t mean that the overall process will be sped up. However, this is probably the question that should be asked. I think a better question is if AI assistance is valuable to the task. Perhaps the AI won’t make the task complete faster but improves the quality (value, yield, …) of the task output. This is a better question to ask.
2. How highly paid are the people who perform this task?
If the task is something that is very highly paid, then superficially that sounds like it is a good candidate for automation. However, this is not the really an important question to ask. The operating cost of automation is dramatically less than the cost of human labor. The real question is how much effort it takes to setup the task. With thaat cost rapidly dropping I would say that any task that can replace a human is a good candidate for automation.
I think this question becomes is the task understood well enough to automate. If the task is not understood well enough to automate, then it is not a good candidate for automation. However, if the task is understood well enough to automate, then it is pretty much always a good candidate for automation.
- Reference blog post on cost of a human vs. automation/AI/robot.
- anything more?
3. How capable is the AI of completing the task correctly?
This is an incredidbly good question to ask. If the task is something where the AI/automation failure rate is unacceptable then it should not be considered. With the rapid improvement of AI/automation this is a question that should be asked regularly though. What was not possible yesterday may be possible today. So perhaps a better question is when will the AI be capable of completing the task correctly or at an acceptable failure rate.
4. How easy is it for humans to determine whether the AI output is accurate?
The problem I have with this question is that it is based on the assumption that the AI is not capable of determining if the output is accurate and that a human is required. This implies that this is an augmentation stuation and not really an automation one.
… more to come …