Josh Johnston: Machine learning is one of the ways that we use to achieve artificial intelligence. It's definitely not the only way.
It's usually one tool that we want to use in combination with some other things. When I think about this, I look back at when I worked in self driving cars. A lot of the behaviors ... We didn't use a lot of machine learning. A lot of the behaviors were just rules. This is how we teach a 16-year-old to drive. You don't sit there and say, "Okay. Watch me drive around for a long time, and then you'll figure it out." Because a 16-year old's been doing that. They've been sitting in a car, watching somebody drive. They still haven't figured it out. They learn parts of it, but not everything. We teach a whole bunch of rules. We teach the laws, how you're supposed to drive. We teach this kind of behavior. Then, we go out, and with a couple hours of practice, we now say that this driver's picked it up. This is the same sort of thing that we do with fraud, where it really helps when we have fraud experts, who have deep domain expertise, and they know what fraud looks like and they know where to find it. Those are the sorts of things that we should be programming in explicitly. We should be using that domain knowledge and putting that into the system.
Now, where we use machine learning, is usually on perception, where there's a tremendous amount of data coming in and we want to extract some kind of abstract meaning out of that. We use machine learning in fraud solutions, where we have all of this data, we can pull out of every transaction that we see. We know what a fraud analyst is trying to look for. We try to connect those dots. We figure out how to get from this raw information to something that's useful.
That's how we use machine learning, in combination with our domain expertise, and knowledge of fraud, as well as business specific policies that individual merchants and payment processors would have.
