In Praise of Boring AI
Automation has always been about killing tedious work. AI can do the same.
We spend a lot of time discussing the aspects of AI that are, for better or worse, exciting. The idea that super-intelligent AIs may one day murder or save us all — certainly not dull! The ways AI might displace our jobs or transform education — interesting! But, today, I want to cover the boring aspects of AI.
As context, one of the first major experimental papers on the impact of ChatGPT on work just came out in Science (based on the free working paper here) and the results are pretty impressive: in realistic business writing tasks, ChatGPT decreased the time required for work by 40%, even as outside evaluators rated the quality of work written with the help of AI to be 18% better than the ones done by humans alone.1 After using it, people were more worried about their jobs… but also significantly happier - why?
Because a lot of work is boring, and AI did the boring stuff fast.
This isn’t new. Automation has always been about eliminating work that is repetitive, and often dangerous or boring. Think of the factory worker soldering cans for 8 hours a day, the miner digging with a pick, or the job that most typifies boring and dangerous: the man who sat on a one-legged stool in Alfred Nobel’s Scottish dynamite factory, watching the thermometer to make sure it the TNT didn’t explode ("The surroundings are rather trying to sensitive nerves - your life depends, at every moment, upon a thermometer and a man on a one legged stool.”) We developed automatic can soldering equiment, mining machines, and automatic temperature controls than transformed all of these jobs. As opposed to these previous waves of automation, AI isn’t purpose built to replace any task in particular, instead it does a lot of things, some better than others.
And this is well-suited to making many of our lives better in some narrow, but important, ways, because it allows us to automate tasks. Scholars studying work often conceive of jobs as a bundle of tasks. Take my role as a business school professor. My job isn't just a single, indivisible entity. Instead, it comprises a variety of tasks: teaching, researching, writing, filling out annual reports, maintaining my computer, writing letters of recommendation, and more. The job title 'professor' is just a label; the daily experience consists of this mix of tasks. What tasks are involved in my job is somewhat arbitrary - professors at other schools may have different tasks, but these are mine.
And some of those tasks are boring. Many of them are boring and important (writing letter of recommendation is an honor, and an important one, but not a lot of fun) and I have written about the risks of automating those tasks before. But a lot of boring work is just tedious, and is not worthy of deep focus. In an ideal world, we wouldn’t do this stuff at all. In our less ideal world, AI can do it for us.
Some ways I automated one-off boring tasks
In fact, one of the ways I most use AI is to get good-enough stuff done quickly, which lets me focus on more important things. For example, I needed to create a quick financial model in Excel for a fake startup that I was using as a minor teaching example.
Prior to AI, this would have taken time, and effort, for something that was going to only be used for a few minutes, and just once. But now, I can ask Code Interpreter: I am an entrepreneurship professor teaching an MBA course. Create an example revenue projection as a downloadable CSV file for a startup that delivers homemade food from amateur chefs. make it a five year projection. make any assumptions you need. A couple more suggestions, and I was done, with a working CSV file based on math I could quickly check.
Or consider another time-consuming, relatively low-value task: putting together images for PowerPoint presentations. While I commission artists for many of my projects, I don’t do that for one-off presentations. In the past, I tended to just look online for creative commons photos that I can use. This is time consuming, and often results in boring pictures.
Now, I take a different approach:
I go to Midjourney and use the /describe feature (literally, type “/describe” and it lets you upload a picture), which generates a bunch of prompts that would result in similar pictures. As you will note, the prompts are incomprehensible (“caffinol developing”?) but that doesn’t matter, I can just use them as-is.
I add a couple words to the prompts. Here, I take a picture of women at a laptop, generate prompts using /describe, and add to those prompts a phrase “at a cafe”, “dramatic scene”, or “underwater.” The result is a more interesting picture than the original. Doing this takes a minute or two (and is much more fun than searching Pexels for the 20th time).
If that doesn’t work, I can even hand-draw a terrible Microsoft Paint draft of what I am thinking of, upload it to Bing, and ask it to turn it into a better image.
And, of course, the AI is excellent when you want to fill out a form, or rewrite a document for a different audience, and basically any other one-off task that would otherwise cause you to sigh and wish you were doing something else. I am especially excited by how good it is getting at filling out complex paperwork. For example, here is Bing walking me through the process of launching a restaurant in Philadelphia, including reading requirements PDFs and writing responses. I wouldn’t trust its work without review, yet, but it helps save some time and pain.
A machine for killing boredom
While not all work has to be thrilling, a huge amount of it is boring for no reason, and that seems to be a big problem. Not only is boredom a top cause for people leaving companies, but we do crazy stuff when bored. One small study of undergraduates found that 66% of men and a quarter of of women choose to painfully shock themselves rather than sit quietly with nothing to do for 15 minutes. And in a set of preregistered studies of 7,000 people, boredom was linked to sadism. For example, 18% of bored people killed worms when given a chance (only 2% of non-bored people did), and bored parents and soldiers both act more sadistically. Boredom is dangerous, in its own way.
So it is odd that we let so much of work become boring. In surveys, people report being bored about 10 hours a week at work. In an ideal world, managers would spend time trying to end the useless and repetitive work that leads to boredom, and to adjust work to focus on the more engaging tasks. But, despite years of management advice, most official rituals, forms, and requirements persist long past their usefulness. If humans couldn’t end this tedious work, the machines can.
Thus, if we want to think about the first work we truly give to AIs, maybe we should start the way every other automation wave has started: with the tedious, (mentally) dangerous, and repetitive. Companies and organizations could start with thinking about how to make boring processes “AI friendly,” allowing machines (with human supervision) to fill our required forms. Rewarding workers for slaying boring tasks with AI could also help streamline operations while making everyone happier. And, if this sheds light on tasks that could be safely automated with no decrease in value, so much the better. Maybe that is work that can be eliminated.
As I have written before, AI is going to have many effects, good and bad, and a lot will depend on how we decide to use it. One way to start us on a positive path is to begin by thinking about ways that AI can help us flourish by automating what holds us back. For many people, a quarter of their working life is tedious. Changing that is a good first step, and establishes a precedent of using AI as a way to free ourselves from drudgery, allowing us to focus on what matters.
And this is the old ChatGPT-3.5, much less capable than GPT-4