The press, consultants, and IT market research firms would have us believe that AI is the most exciting technology available today. And it is—but in part because there aren’t a lot of exciting alternatives. Blockchain isn’t turning out to be the revolutionary technology it promised to be, and it will take a long time to come to fruition. Similarly, the Internet of Things is taking forever, in part because there are way too many standards, and none of them is sufficiently influential. It’s not surprising, for example, that the vendor C3 changed its name from “C3 IoT” to “C3 AI.”
There is a lot happening in the world of AI, and I don’t see us retreating into another AI winter again. However, as I have argued elsewhere on this site, it is improving at a linear rate rather than an exponential one. And more importantly, the types and applications of AI that are providing the most value are the most boring ones. Perhaps the sooner we begin thinking of AI as a “boring but important” technology, the better off we’ll be.
The Most Boring Types of AI
What’s so boring about AI? To begin with, the types of AI that are the most useful and pervasive are also the most boring. Take machine learning as an example. It is probably the most common form of AI, but the one that has been around the longest. Some argue that it is just statistics, and not really AI. I would argue that it’s basically the same as predictive analytics. But nonetheless, supervised machine learning is incredibly useful and valuable stuff. There is a lot to say for being able to predict the values of outcome variables from a model created from other data. I’m not the only person who believes that machine learning is boring; at least one prominent vendor of that technology has said so.
Perhaps even more boring than machine learning is robotic process automation. RPA, which is really just a collection of previous technologies like rule engines, graphic user interfaces, and screen scraping, is perhaps the AI technology most likely to be called “not really AI.” But I put it in the category because it does things that humans could only do previously, and it is increasingly being combined with tools like machine learning. It automates structured, repetitive—i.e., boring—work tasks.
But the widely-acknowledged fact that RPA is boring doesn’t mean it’s not useful. It’s growing faster than any other enterprise technology, according to Gartner. One RPA vendor, UiPath, has grown about 5000 percent since 2016. Companies find it easy and fast to implement, and it doesn’t cost much relative to other types of AI. As a result it often provides rapid returns on investment. The technology may be boring, but the returns are not.
Boring AI Use Cases
There are certainly some sexy AI use cases, although many of the most-discussed ones (curing cancer, powering autonomous vehicles) remain just out of reach even after many years of work. But it’s the boring applications of AI that are really taking off. I could write a book about these, but frankly it would be too boring to read. But let me just list a few of the “boring but incredibly useful” applications of AI:
AIOps—The use of AI to support IT operations is both boring and pervasive. It’s a fast-growing market that’s expected to reach over $11 billion by 2023. And it’s important as more and more companies become digitally-driven. However, it’s unlikely that you’re going to get your CEO excited about it, and its benefits fall into the category of incremental efficiency. By all means do it, but don’t expect to be lauded for your efforts.
Data Integration—If you’ve been hibernating for the past twenty years or so, you may not realize that companies have difficulty integrating data across different databases. We’ve had various labor-intensive methods for doing this, but none have worked out terribly well. What is working well is the use of probabilistic matching-based machine learning to integrate data elements across databases. Several companies I’ve spoken with, including GE and GlaxoSmithKline, have had great results from it. But like IT operations, it’s a form of inside baseball that is unlikely to excite the rest of the world.
Contract matching—I have a friend who works in AI and who is convinced that the ability of the technology to extract legal terms from contracts, and match them to what was actually received from the contract, is going to make for a very successful startup. It is certainly true that firms can, with relative ease, figure out whether their suppliers actually shipped all of the items they were supposed to according to a contract. From what I have heard, they often don’t—so there is a lot of money to be saved in tracking down this sort of mismatch. It doesn’t make for great news stories, but I suppose it isn’t so boring if you care about making money.
Okay, I had better stop describing these boring use cases if I want you to keep reading. But the common element to them is that AI helps a lot in these domains, and a lot of money and time can be saved by using the technology. The only problem with these AI applications is that they don’t make for interesting stories. At best the accounts of them might appear in the back pages of some boring trade journal.
But if you can persuade your organization that AI “moon shots” are not the best way to pursue the technology, and that “boring AI” can make or save a company a lot of money, I think you’ll be more successful with it. You won’t raise expectations too high, and you won’t see AI descend into the “trough of disillusionment.” You may not be able to hire the best-known AI rock stars, but you may be able to hang onto the AI experts that you do hire. Your company may not appear in the press for AI innovations, but it will be more likely to appear in the list of companies that make the most money for their shareholders or owners. Isn’t that somewhat more important?