Economists Luis Garicano, Jin Li, and Yanhui Wu predicted in Messy Work: Jobs Beyond the Touch of Artificial Intelligence that “if every team could produce better AI-based analysis to support their arguments, the need for conflict resolution and authority-based decision-making would increase dramatically.” They noted that many workplace decisions are not made on merit alone; They also involve deciding “who gets what they want.” Who is ready for big action, or who lacks the experience to take on onerous responsibilities? What idea always sounds good but never works? What is the CEO thinking but never saying it? This information is not explicit but presupposed—it is known, but not written down—and therefore cannot be used by AI systems. Furthermore, the proliferation of AI-generated work will make it more difficult for decision-makers to gather the implicit information they need. If every cover letter is well-written and every memo is detailed and well-structured, how will your boss know who to trust? If everyone uses artificial intelligence to generate ideas, how do you know who is truly creative?
ChatGPT first appeared in 2022; Crowder, 2023. Almost immediately, predictions of a coming jobs disaster began. There’s no doubt that people find AI useful: studies and surveys show that more and more office workers are now using AI on a daily basis. Certain fields — coding, recruiting, science research, law — do seem to be shifting. Overall, however, the impact of artificial intelligence is difficult to measure. Many employees appear to be using it semi-secretly on their own devices, perhaps saving themselves time or improving their work in ways that aren’t reflected on the bottom line. Recent college graduates are finding it increasingly difficult to find work, and customer service jobs may be disappearing, but software engineer job openings will decrease significantly in 2025 but increase in 2026. Does this mean artificial intelligence is creating software jobs? Or is the industry simply rebounding after post-pandemic layoffs? No one knows.
“The early evidence is not the final word on the future of work in the artificial intelligence world,” a team of Stanford University researchers warned in July. Part of the difficulty is that, with the addition of artificial intelligence, we realize that we don’t necessarily know how things work. Why are our jobs set up this way? How much can they change? Which of the things we do are overtly human, and which are ripe for automation? What makes working with someone valuable beyond the work they do? As more and more people use artificial intelligence, the rigid idea of an AI-driven employment apocalypse is being replaced by a growing number of challenging problems that managers and workers are only beginning to grapple with.
Economists have a term – production function – to describe how things are made. Imagine you are hosting a dinner party for ten people. If you decide to make steak fries, you must cook the steak and fries within a few minutes of your guests sitting down to eat. If you only have four burners on your stove, you’ll need to fry the steaks in batches; if extra guests arrive, you’ll have to cook an extra steak. Alternatively, you can make a big pot of stew. In this case, you can get almost everything done a day or two ahead of time and have the pots on the stove when your guests arrive. There may be enough available if extra guests show up. Steak fries and stew serve completely different functions. If you plot them as a graph, effort on one axis and results on the other, you get completely different curves.
Among other things, Chaotic Work touches on the precise ways in which artificial intelligence is changing the productive functions of work. The authors argue that artificial intelligence creates a “new progressive shape” for what we do, and it’s not just up and to the right. They describe a study in which artists were given artificial intelligence tools to help them quickly deliver finished products—illustrations of scenes from novels. In half an hour the artists achieved “a level of quality that would have taken two hours by hand” – by then, however, progress had slowed. Because artists used artificial intelligence to “achieve beautiful images before giving much thought to the composition,” they worked to improve it; “even as artists kept tweaking cues and tinkering with details, further gains were barely noticeable.” Eventually, the artists fell into two categories: those who paused their work after about an hour, and those who continued working fruitlessly.