AI is computer science disguised as hard work

Roger Magoulas recently sat down with Rob Thomas and Tim O’Reilly to discuss Thomas’s AI framework called the AI Ladder, which, according to his recent paper, is a framework that describes “the increasing levels of analytic sophistication that lead to, and buttress, a thriving AI environment.” Thomas notes both in his paper and in a recent keynote discussion he had with O’Reilly that “there is no AI without IA [information architecture].” In this interview, Thomas and O’Reilly delve deeper into their conversation, outlining the “rungs” of the ladder and some of the big-picture opportunities and consequences of AI in real-time business environments.

Here are some highlights:

Thomas notes that the framework embraces an iterative approach: “I would call it a builder’s market, where people have got to get their hands on the tools and go try things. It’s not about building a nine month strategic plan, and then building a huge team. It’s about picking a problem—making better predictions or automating something, or trying to optimize a business process—and go give it a try.” (01:20)

O’Reilly expands on the iterative concept, noting that working iteratively helps get beyond the hype and the “all-or-nothing” perspective AI hype fosters. “AI has had so much hype attached to it,” he explains, “that everybody comes away with a sort of binary: it’s either the complete self-driving car that didn’t really work or is not doing as well as they thought, so even the big dudes can’t do it.  So therefore, it’s over-hyped, or it’s nothing. And part of what the AI Ladder gets across is, yes, there are very futuristic projects that tend to represent AI in people’s minds, but there’s actually a series of steps used to get there. So companies think, ‘Well, what’s that one big win that we could have like the one that Google’s working on?’ And that’s not the right way to think about this. You want to get up there, but you have to start at the bottom of the ladder, and you have to do a bunch of work to get ready, and then you do a bunch of small projects and you gradually build your competency, rather than simply saying, ‘I’ve got to get some of that AI magic, so I’ll go to a vendor who promises to do something that sounds magical to me.'” (01:54)

The work, Thomas says, starts with the “lingua franca of the AI world,” which Thomas and O’Reilly list as such languages as Python, TensorFlow, and PyTorch. “This is computer science,” says Thomas, “and it’s computer science disguised as hard work. You better roll up your sleeves. … I think it’s hard for a lot of people to get their heads around the idea that whatever we’re doing today, we’re probably going to be doing something different in six to 12 months. So, it will take constant learning to do this well.” (03:22)

Communication is going to be key, Thomas notes, which is going to require a way to unlock normal human communication—in written form, spoken form, structured and unstructured text, etc.—to get to the real insights. “That’s why I say NLP is ultimately going to become this nervous system,” Thomas says, “where if you can do that very well; it’s going to make a big difference. And there are industry benchmarks on this. The newest one’s called SuperGLUE. … So we’re getting almost to a human level in NLP, and these benchmarks will continue to move the bar, which is good because it challenges us to be better. (11:12)

Thomas says his company encourages clients to take steps toward AI adoption because major factors are coming together to make this an opportune time to get on board early. “This is finally becoming a board-level topic for companies I interact with,” he said. “Just look at the economics: $16 trillion of GDP is expected to be accrued from AI between now and 2030. It’s hard to overlook numbers that big. Let’s say that’s off by 50%; it’s still a big number. So, there’s an economic piece. Adoption today—meaning companies that have seriously done something with AI—depending on who you believe, is somewhere between 4-8%. You take those two things—the biggest economic opportunity any of us will ever see in our lives, and very low adoption—that’s a pretty good opportunity to step into the moment and do something as a company.” (14:45)

It’s important for companies to innovate in these areas, too, O’Reilly notes, because the problems we’re going to face in the coming decades are going to require it. “The thing I get most excited about is that we are growing our data universe, and we have to grow our “understanding universe” as well. You think about things like handheld DNA sensors. We had a demonstration of this device at our Science Foo Camp. They were using it to look at a virus that was affecting cassava roots in Africa–they literally were doing handheld gene sequencing in the field. Think about how compute power is going out to an edge like that, and you start adding up all of those edges–we had a presentation this morning, for example, about how a new crop disease or plague of insects, or whatever, in some part of the world could have an effect on commodity prices worldwide. That’s the kind of stuff we’re going to be building systems for so we’re increasingly able to respond in real time. When I look at the arc of history, the problems we’re going to be hitting in the 21st century are so large that we’ll need all the help we can get.” (15:36)