Too Much of Today's AI Is a Novelty Without a Clear Plan to Make Money
The 2018 artificial intelligence landscape looks an awful lot like the Sharper Image catalog. It's chock full of products that were built merely because we can build them, and because they're marketable.
Much like these products, too much of the AI on the market today is disposable novelty technology. Nobody conducted market research to determine the total addressable market for bacon toasters. They didn't have focus groups with likely customers. They built a novelty that was good for a chuckle and had just enough utility to convince a few people to part ways with a small amount of cash and give it a spin. If that doesn't sound like a lot of the AI for sale these days, I don't know what does.
In a half-hearted defense of the industry, AI experts like to remind us that "it's still early." Others explain that the first wave of enterprise AI is "doomed to fail," and somehow that's both preordained and acceptable. Given its well-understood power and potential impact on society, shouldn't AI be held to a higher standard?
As a result, smart customers are asking: Why is there so much hedging and so little accountability in AI?
Researchers run amok
I love visiting research labs as much as the next nerd, but we need to be careful about researcher-led AI implementations in business scenarios. With a huge talent shortage in AI, many companies are poaching PhD's from universities across the globe. Facebook boasts an AI research team of over 100 researchers on staff, a luxury few other tech companies can claim, yet the Facebook Messenger AI group was shut down soon after achieving a 70 percent failure rate.
Some might argue the platform failed despite the massive investment of capital and academic talent, but we need to be honest with ourselves: It failed because of it.
Money and talent matter. They matter a lot. But, the failure rates we're experiencing in this industry look a lot more like scientific research than IT implementations. Nature recently reported, "More than 70 percent of researchers have tried and failed to reproduce another scientist's experiments, and more than half have failed to reproduce their own experiments."
The AI industry has imported droves of academic research scientists, and the result is a ton of experimentation with customers' businesses. Don't get me wrong -- I value research, experimentation and even failure as a technology entrepreneur. But any entrepreneur would agree that it's unacceptable to ask customers to shoulder all the risks.
Meanwhile, researchers, by necessity, are focused on the technology and its inner workings. They're not trained for, nor are they typically very good at, ensuring optimal business outcomes. Consider for a second that AI failures are not a result of a shortage of PhD's in artificial intelligence; they're a result of the absence of business analysts and customer success specialists on their teams.
If you were in this far over your head in terms of business savvy, you'd be hedging too.
The collateral damage of navel gazing
Researchers are one crucial part of the AI ecosystem. But thousands of developers and technologists have flooded into the space as it gained steam over the past decade. If you've ever spent time on sites like Stack Exchange or Hacker News, you'll find devoted communities of talented technologists debating the merits of new technologies, arguing over the finer points of programming languages and tools, platforms and standards. This is how the technology industry advances itself, one step at a time.
Since AI is still in a relatively nascent stages, discussion and debate around all these topics is at a fever pitch. As an industry, we're still working to establish best practices and standards, and the process requires that our technical leaders look inward at the technology itself.
The good news is that we've done this for decades -- this is how we ironed out digital transformation and the transition to cloud computing, then mobile, and now we're doing it for AI.
The bad news is that most in the industry have spent little time and energy understanding their customers and their business needs. Silicon Valley has a long history of building glamorous new technologies that fail on the first try because they don't have product/market fit. Building the best technology is not the same thing as building the best technology for my business.
This exact phenomenon is what we're seeing in AI right now, at least with developers that haven't obsessed over their customers.
Customers, customers, customers
The next great breakthrough in AI isn't going to come from a lab at Stanford. It's not going to happen in code with a customer. It's going to happen in HR departments, where recruiting teams will enact strategies to hire businesspeople that have the capacity to bridge the gap between business technologies and business outcomes.
We need to obsess about the business of the AI buyer, and we need to obsess about their customers, too. AI is not a one-off technology -- it affects the entire value chain from start to finish. These technologies need to fit the business, not the other way around.
Ever since the days of "digital transformation," we've taken to lecturing businesses about how their IT should work. That's not going to work anymore. AI reaches too far into a business and touches too many processes for any sane executive to let technology companies tell them how to run their business.
We need businesspeople that are good at listening to their customers, and their customers' customers -- because that's where AI has a real impact.
Thanks to AI's capacity to transform businesses, the customer is, once again, always right.