BABBAGE: UP IN SMOKE
Host - Kenneth Cukier: What is the role of Artificial Intelligence in business? To discuss this, I’m joined in the studio by Zia Chishti, Founder and CEO of Afiniti; A company that uses artificial intelligence to monitor the patterns of human behavior. Prior to this, Zia was the founder of the Align Technology and began career in the mergers and acquisitions section of Wall Street.
Zia, welcome to Babbage!
Lots of people have great expectations about AI in business but how do you see AI moving into the commercial sphere?
Zia Chishti: Well first of all, I think the whole segment is overhyped. If you think about it, what is now conveniently called Artificial Intelligence is really just rebranded algorithms which have existed for 20 years and some of the flowery descriptions like Deep Learning are simply re-characterizations of what was called a multi-layer neural network, back when I was in college.
Kenn: But the difference is that they now work.
Zia: And they worked back then too. What’s changed really is the computational capacity so they work faster. But the underlying principles are identical, they really haven’t shifted much, no real tremendous algorithm breakthrough. First we need to distinguish what is this AI category. A lot of this is just hyped, naive venture capitalist thrusting money on anybody who takes that label as part of their pitch. So when you abstract away that whole piece, what impact are you really having from AI? At the moment is quite dilute. Despite what you may think in the ambient media, there is nothing that has substantively changed in the way of how enterprises are running their businesses that would in my mind get to this level of transformation that seems to be common place in thought.
Kenn: Okay, this is really interesting because although it’s true that the fundamental aspects of AI and algorithms are largely the same as they were 20 years ago, the implementation is incredibly different. But when I look out at the world and see things like self-driving cars, although certainly not on the near horizon or the far horizon, or delivery drones that rely on AI for crash avoidance, or even the fact that every single bank and post office uses OCR (Optical Character Recognition) based on AI to identify hand written material and extract that information, it looks to me as if AI is everywhere even if you’re calling it algorithms.
Zia: Let's look at those three things that you describe. The easiest one is OCR. I mean that's been around in essentially the same format for almost two decades. If you ran an OCR algorithm on top of a handwritten sheet of paper, you'll get about the same accuracy that you get today, give or take a decimal point or two. Then you have these self-driving cars and you've got drones. The definition of AI that I'd like to impose on those is one of a continuously learning environment. So, as information comes in you have a learning module that tie traits to a closer and closer solution over time. The algorithms that underpin how our car drives are unaffected by the ambient environment of the car over time. It is a set of algorithms that have been imbued into that self-driving car to recognize various features, drive the car around those features, have certain traffic situations pre-programmed such that they operate in a specific way in response to those traffic situations. It isn't a self-learning environment that adapts based on an improved understanding of a traffic environment or a driving pattern or other environmental information that gives rise to a better and better, quote virtual driver. That's not what it does. You have to distinguish that. As sexy as it may be, to see a quote self-driving car moving around town, that is not an apt characterization of a AI’s application within that space. Drones are the same thing, you're trying to get from point A to point B and avoid collisions. Again, that's an algorithmic process. The drone itself isn't learning about its environment or adapting how it gets from point A to point B, it's been given a set of algorithms which is executing. The design of those algorithms, may be a bit better, if imbued with an AI predecessor process that helps it to more rapidly identify things in its environment. Sure. But AI within the drone or within the car, that's a stretch.
Kenn: Where do you see AI going into business and how do you use it in your company?
Zia: Our application of AI is in behavioral prediction. So, we're looking at how humans behave differently; customer to customer, agent to agent, in a large enterprise. And that is a continually evolving prediction of behavior over time in response to new information. So, as a customer calls in once, twice, three times we learn more and more and more about that customer. We have a better and better understanding of how the customer is likely to interact in the future.
Kenn: But let's look ahead a little bit. I think we're going to disagree on what AI is. But where we're both going to agree is it's going to make a serious impact on business. Where do you see it making the biggest?
Zia: I think that's a set of applications where: A) The economic output is very well defined or the social values are very well defined. And B) It's relatively easy to articulate at a core decision making level. So: Well-defined economic output, easy to articulate. What are some of those? Medical image processing. Right. So, abstracting on that field of OCR. Can you detect cancer more efficiently? In 3-D images as opposed to 2D images, a very very powerful application... Hydrocarbon mining and extraction with a three-dimensional view derived from bouncing sound waves off of substrates in geological layers; can you identify pockets of hydrocarbons more efficiently than just drilling in random spots over time? If you have applications like ours where you're predicting human behavior, can you use that to identify pockets of customers more effectively or pockets of employees more effectively within a large enterprise? These are all applications that are real world, exist today, making material if not transformational contributions to enterprises. I think those will be the breed that will succeed in the next decade.
Kenn: Zia, thank you very much!
Zia: Thank you very much. Thank you for having me.
Kenn: So, what are your views on Artificial Intelligence in business? And in particular, whether you agree with Zia that all we're doing today is the same stuff we did in the past? It's just algorithmic and not real AI. Or do you there is indeed something new going on?