Short of a massive government intervention, which is entirely possible given who is in charge of the US government, I am struggling to understand how imitative AI stays alive, financially, as a business. Even putting aside, the lawsuits around its use of copyrighted material to train data and its enormous environmental costs (there is zero chance they will be required to pay those externalities under the Musk/Trump Administration), these tools simply do not live up to their hype. Two recent articles really highlight that problem.
First, the job of prompt engineer, a hot topic even last year, has essentially disappeared as a distinct role. At one point, being able to craft a prompt that maximized the utility of an AI response and minimized the bullshit it spewed was expected to be a hot job in the new era of imitative AI. It largely does not exist today. Part of that decline is imitative AI firms saw a weakness and tried to improve input processing, and part of it is that firms that use imitative AI realized that they could train their existing employees to be better at prompting rather than hire a specialist. Why spend money and introduce a bottleneck when you can just add that to the existing responsibilities of people in your firm who use the tools directly?
The importance of this, obviously, is that it diminishes one of the arguments against moving forward with mass imitative AI adoption. Granted, I don’t hear this as much, which is telling in its own right, but the argument used to be that imitative AI might destroy some jobs but would provide new and likely better jobs. This argument is based on the flawed idea that new technology has always done this (it hasn’t, not in the way and to the extent that its boosters claim) and therefor imitative AI would follow that path. But it hasn’t really, and it doesn’t look like it will. It is not even creating that many new data science jobs, as imitative AI is being attempted by only a handful of companies and being done large scale by really only one — OpenAI. Imitative AI may eat jobs, but it certainly doesn’t appear to be producing new ones.
And it might not even eat that many jobs, at least long term. A recent study shows that imitative AI tools cause more work than they remove. There are a couple of caveats, the primary one being that the study only ran through 2024. I am sure that hypers will claim that the newest of the new are all the new hotness and about to revolutionize work (i.e. fire everyone and let a company be run by a founder and a chatbot), and I will get to that in a moment. This is not especially surprising, as imitative AI systems have a bullshit problem. Anything it produces needs to be checked and double checked by an expert. Otherwise, you will find out the hard way that they are confidently wrong in damaging yet hard to find ways. Any timed saved is largely lost by the work you or others need to put in to clean up its messes.
And that brings me to the second objection to this study: another study showed that worker productivity was improved by 15%. 15% is not nothing in productivity terms, but it’s not earth shattering, either. It is not, probably, enough to justify massive job losses that are likely required to make imitative AI firms profitable. The kicker is the fine details of that study. The study apparently only focused on jobs that were highly suited to working with AI. In other words, the best-case scenario is a middling productivity improvement, far below what is needed to financially save these firms. Absent an infusion of government money, always a possibility given who is running the government now, there just doesn’t seem to be a path to making real money with imitative AI.
And that is fine. History is littered with failed technologies that at one point seemed as if they would be viable products. 3D televisions, NFTs, the metaverse, etc. The problem is how they die. They can die like 3D televisions — manufactures realized people didn’t want them and stopped making them. Or they can die like “pivot to video” in which Facebook lied about video numbers and destroyed uncounted number of news jobs and news sites. If we aren’t careful, we will let the imitative AI hypers do the same.
Want more oddities like this? You can subscribe to my free newsletter
