Tim O'Reilly is a kind of bard of technology, a lyrical poet of computing's past, present and future. Ask him a question and whole paragraphs of reflection bubble up.
Is the present state of artificial intelligence, for example, bigger than the open-source software revolution, an epochal development chronicled in detail from the front lines by O'Reilly's publishing company?
"That's an interesting question," says O'Reilly, before re-framing it, declaring that there is something bigger than AI itself.
"I think in the long run, this transformation to machine autonomy, and to, basically, machines that are in a new kind of hybrid existence with humans — they talk about AI as separate from us, but all interesting machines are hybrids of human and machine — we have this machine that has been amplifying things we can do, and I think of the human-machine symbiosis as a trend that is probably bigger than the internet, and bigger than open source, and of which AI is one manifestation."
It is vintage O'Reilly, a sweeping, swerving rumination that is always intensely focused on the currents actually going on in the world of technology, rather than some rarified theory. More than many futurists or pundits, for decades O'Reilly has observed and embraced whatever is emerging from a given programming language or software "stack." He sees and doesn't pass judgment, but takes it all in and points out the enduring and significant.
The prompt for such reflections can in many cases be something very mundane. In this case, it is a chat with ZDNet on the sidelines of "Strata," the tech conference that O'Reilly's eponymous publishing company hosted this week in New York.
O'Reilly has been on stage earlier in the day to discuss a paper, published in conjunction with O'Reilly, authored by Robert Thomas, head of data and artificial intelligence efforts at IBM. The paper, "The AI Ladder," is a brief guide for corporate executives about how to start to think about employing AI in their companies.
Like a lot of IBM publications, it is a bit on the banal side, filled with recommendations that are of an extremely general nature, things like, Make sure you have good data if you're going to do AI.
Nevertheless, O'Reilly is able to excavate such material and come up with gems, weaving in observations from other writers and thinkers. He sees connections to the writings of MIT economist David Autor, who describes "superstar" companies that are gobbling up a disproportionate share of global wealth, companies such as Google and Facebook and Amazon.
"If you look at all those companies that are superstars, they are infused with AI," says O'Reilly. "It's because they are in some sense digitally native." O'Reilly takes a reporter's iPad Pro, takes the Apple Pencil, and draws a sketch of the "ladder of AI." At the top of the latter, the kind of ultimate stage, is the "infusion" of AI into a company's operations, a term Thomas uses. "What I like about this ideas is, at the top of any company that is a superstar is a certain kind of machine intelligence that's operating at an autonomous level," he says. "That's true of Amazon and Wal-Mart, and, increasingly, of a lot of companies."
What about those who fear autonomous machines, whose say taking the human out of to loop is dangerous? O'Reilly doesn't buy it.
"I don't think most people would mind talking to a machine," he asserts. "Take the example of customer service: Do you really care when you call Fidelity or a bank, if you can have a chatbot, and it gives you the right answer, you're pretty happy, versus a human rummaging around for several minutes looking for the answer.
Even if AI hasn't mastered common sense, there are many ways in which it can take over the drudgery that humans have had to do. "We're talking about companies where humans will be the managers, overseeing autonomous processes of computers," he says. "There is a whole class of problems that are still on paper.
"Think about a place like Ernst & Young; they've got three billion pages a year of contracts, and they are now digitizing that, and using language understanding. I'm pretty sure you can get a computer to tell apart three billion pages of contracts better than a team of human paralegals."
Of course, most companies are not digitally native. They won't find it as easy as Google, say, to incorporate "autonomous processes." But to O'Reilly, this is all about the "democratization" of AI, moving beyond just the stuff that's always in the headlines.
"There is a democratization of AI, with things such as cloud ML, and TensorFlow and PyTorch." He credits IBM for being open-minded, willing to work with all manner of tools, and with being willing to take companies on a journey. "To get better at this stuff, you have to start at the beginning, at the bottom rung of the ladder," he muses. "Companies will say, I'm going to be ready to go buy this product," meaning, AI. "Well, no, first what are you trying to do, and what data do you have" are the kinds of preparatory questions that companies must ask, he argues.
All well and good, but what about all the failure cases of AI? What about the adversarial examples that make image recognition systems fail, and the "deep fakes" online, where people can use machine learning for mischief? AI today is both fragile, O'Reilly admits, and also potentially dangerous to society.
To all that, the answer is more AI, he thinks, not less. "Let me put it this way, the problems we face as a society are so large, we are going to need all the help we can get," says O'Reilly.
"The way through is forward."