Artificial intelligence and machine learning has been a topic that’s been on our minds quite a bit lately at PrecisionLender.
In this two-part series, Carl Ryden and Jim Young discuss some of our favorite articles about AI and how it relates to the banking industry. In part 1, we’ll look at the two perspectives of adopting AI.
Jim Young: Hi and welcome to The Purposeful Banker, the podcast brought to you by PrecisionLender, where we discuss the big topics on the minds of today’s best bankers. I’m your host, Jim Young. Conversations with PrecisionLender CEO Carl Ryden are always interesting and usually wide ranging. This one about artificial intelligence, and machine learning, and banking was no exception, so we decided to divide this episode into two parts. Here’s part one and part two will be coming your way next week. Enjoy.
Artificial intelligence and machine learning has been a topic that’s been on our minds quite a bit lately at PrecisionLender. You may have heard us discuss it on the recent podcast that we had with Carl Ryden and Justin Lafayette of Georgian Partners. Just about every day here at PrecisionLender, people are sharing interesting articles, thought pieces about AI and machine learning. It’s generating a lot of discussion. We thought we’d pull back the curtain a little bit, let you in on some of those conversations. That’s why I’ve got Carl Ryden with me in the studio today. We’re going to talk a little bit about a few of these recent articles and some of the interesting and important points they raised. Carl, first off, for those who didn’t hear that first podcast, why is artificial intelligence and machine learning a topic that we’re talking a lot about at PrecisionLender?
Carl Ryden: Well, from the beginning when we started PrecisionLender, the idea was to build a product that had primary value to the folks who used it. We still believe that’s true. Whatever you do, you have to begin with that person who’s using it on the front line and create something that’s valuable to them in that it helps them achieve what they’re trying to achieve. In our case, it’s to help commercial lenders win better deals, build strong relationships while they do it and build a better brand for themselves and for the bank.
When you do that, if you do it in the right way, in a thoughtful way with a eye to the future, as I think we have from the beginning, you actually build up a unique and valuable set of data. The question is what do you do with that data? How do you take that data that you intuitively know is valuable because it’s unique and interesting and affects an important part of the bank, and then connect it back, and use it to enhance that value creation process?
One of the first places you turn is what do you do with data? A lot of folks would turn to benchmarking. When we compute averages and compute statistics, and say, “Well, here’s the average spread on the deal here, or the average spread on the deal there. Ultimately, we chose not to do that because it felt like we would see folks who would use systems like that. It actually destroyed value in the bank.
I’ll tell you the simple case of how. I call it the doorway problem. If you look at the average height of a person who walks into your bank, the average height’s six foot. Then you change your door height to be six two for some reason. Now that affects the acquisition process that feeds back into the system. Now you measure the average height two weeks later, and it’s five four. What happens is there is this dynamic. I’m an engineer by training and probably by nature, so this is a systems dynamics approach where when you take that and you feed it back, it actually causes a bad outcome. Never mind the fact that the data a lot of folks are feeding into these benchmarking systems was poor to start with.
What we wanted to do is instead take that data and use it as a means of extracting intelligence. Intelligence about what folks are doing, what they’re not doing, what they do in this particular situation. Then feeding that back in. That leads you to artificial intelligence and machine learning. Having a machine or a system learn the more you use it, learn about how it’s being used, learn what the best lenders are doing differently from the not so great lenders, what the ones who are winning are doing differently. Use it as a tool and a means of continuous improvement.
I think you see in a lot of this stuff, and I think you’ll see more of it crop up in every aspect of our lives. You have AI assistants scheduling meetings with x.ai and Clara Labs and those things. I think you also have Siri and Cortana, and even things we use every day like Spotify selecting music that it thinks we’re going to like. Using intelligence as a means of enhancing that customer experience for a bank as that end borrower experience is really a key thing.
Again, we wanted to get to where we distill down intelligence and feed that intelligence back in. Not just the raw data, because we saw problems with that.
Jim Young: One of the articles, first one on that list on here, was from the site Medium, which has a whole lot of interesting topics. It’s called Artificial Intelligence is the New Digital Divide. It’s by, and apologies in advance to the author, I believe it’s Enrique Dans, Professor at the IE Business School in Madrid. One of the first things that struck me in here, Carl, is he makes a reference to the AI Winter, which I had never heard of before.
I clicked on it in Wikipedia. It talks about a period when AI was sort of out of favor at universities and funding was getting cut. That was around 1984, which kind of blew my mind because I tend to think of, you mentioned Cortana and Alexa and those sort of things, you feel that’s cutting edge. We’re talking about something 30 years ago. Obviously, it’s been around for a long time. Why is it now the buzzy thing that we keep seeing everywhere? What’s changed that suddenly catapulted it into the forefront?
Carl Ryden: There’s a bunch of layers to answer on that question. I been working with computers since probably 1980. AI was always something that was cool, and futuristic, and everybody wanted to work on. It came in fits and starts. There was a lot of excitement, and disappointment, and disillusionment. Back in the 80s, when you first experienced the computer, you thought, “This thing is magic. It can do amazing things.” This is a lot of excitement around that. Then we kind of forget how dumb they are. What happens is in the AI first day of winter, folks realize, “Hey, these are machines. They are mechanical calculators.” They’re really good at calculation, and really good at calculating vast amounts of numbers.
The problem is when I say they’re stupid, I mean you can show a human a few samples, a few instances of something. The human can learn and recognize patterns because the human can draw on all of their lifetime of experiences, and all the other things that they hold in their brain. A computer can’t. A computer only sees the data that you give it. Consequently, a computer needs a vast amount of training data. That training data just simply wasn’t available.
The other thing is it needs to process that training data, it needs a lot of storage to hold it. It needs a network and a means of gathering that, and a means of gathering what we call validated truth. I need to know which samples are known goods. Examples of the data that I can use to train on. A computer needs a vast amount of data to train on because it is fairly stupid. It handles that, it compensates that by brute force. Lots of data, lots of processing power when it comes over.
Really what causes the first day of AI Winter was when someone has that realization there just wasn’t enough training data. Even if you had the training data, storing it was ridiculously expensive. Processing power was ridiculously expensive back then. Back in 1980, a gigabyte of storage, there’s a wonderful graph on this. A gigabyte of storage in 1980 cost about half a million dollars. Then the chart goes down every few years where recently, I think a couple years ago, it was free. Then the chart began to go negative. It said, “Okay, how much free data do I get?” It was, “Okay, Google will give you five gig free. Microsoft will give you 15 gig free.” Now it’s like, “How much does a gig of data cost?” Well, you get 15 gig for free. It’s even gone negative.
Same thing’s happened with Moore’s Law of Processing Power. The other thing is the internet as in the use of applications in the cloud has given us a massive set of training data where computer’s network connected systems, cloud connected systems, can actually observe behavior. Uber knows where folks are requesting cars. Uber can see that and track that data.
Even Microsoft with the email and other things can … One of the early cases was spam filtering. You can get folks training the algorithms by saying that’s spam, not spam, that’s spam, that’s not spam. You can actually get vast amounts of training data. That’s where it was successful.
Ultimately what mattered for ImageNet and image recognition was a huge library of known good data that Google could train on, and Amazon could train on, and other folks. That really is what broke us through to where we are now where it really is starting to take off and feed on itself. It’s past the tipping point of where this is going to be a thing. It’s going to be a huge thing that impacts almost everything we do.
Jim Young: That’s what the article talks about. It establishes the return of the use of the digital divide. Basically saying that going forward, there’s going to be the haves and the have nots. You either are on the side that get and understand artificial intelligence, machine learning, and know how to unlock it, or you get left behind. Let’s turn this to banks. Are they in danger of being on the wrong side of that divide, or is it some banks are on the wrong side of that divide?
Carl Ryden: What’s interesting about the article I think was interesting was it framed the problem well. That there’s a digital divide and you want to be one side. It doesn’t really tell you how to determine which side of that divide you’re on. I think that’s the rub is how do you know, and how do you put yourself on the right side of that?
I think there’s maybe two camps on this. We see this a lot in what we do. There’s kind of the back of the bank camp that really sees what they do, maybe not consciously but at least through their actions, it seems they feel that it’s a commodity business. That our job is to reduce costs, and take out costs, and improve efficiency. There is an aspect of that no matter what you do. You can also take that to the extreme where you destroy any chance of creating value.
It is truly a cost plus commodity business. I’m of the belief that in banking that’s not a place you can be. That you are selling the world’s most fungible asset, its dollars. Your money’s just as green as every other bank’s. If you sell the world’s most fungible asset in a cost plus commodity business, if that’s your choice, that takes you down a path that’s really a hard place to be.
The other side is that if you say, “Well, we’re actually not going to be that. We’re going to be in the value creation business. We’re going to be in a value minus, not a cost plus.” Then what you do is you take a different approach. The idea is how do we create a system that creates more value for our customers? Where there’s more value for us to share in.
He said, “What does that have to do with AI?” Well, it has everything to do with AI because the first camp, when they see AI, they say, “We’re going to replace humans. We’re going to take them out of the loop. We’re going to decrease cost, increase efficiency, take it down the commodity path.” Don’t get me wrong, there’s some of that. There’s a lot of cost and waste that can be taken out. You can’t solely do that.
The second camp says, “Okay, we’re going to be using AI to enhance the customer experience to create value for our customers, to create more valuable interactions with our customers.” That is the approach we take is that in what we do, our philosophy has always been to do everything … We use PrecisionLender in our system. We want to do everything that a computer can do well, well, so that … The important part is that so that, so that humans can do everything that humans uniquely do well. Because remember, the machines are still stupid. They just have more data to train on. We’re a long ways away from sentience and those things. Even the experts in the field will tell you that.
What we do is when you take that second approach, AI becomes a means of amplifying humanity, not replacing it. Amplifying humanity by removing all the stuff that computers can do better than humans, that are mere calculation … There’s another article. We’ll talk about that. It moves from calculation to prediction. What we’ve seen now is really at the heart of AI algorithms or prediction algorithms.
What you see is calculation became cheap. Storage became cheap. Training sets became prevalent because of that. Now I can actually get to where I can do more than basic calculations. I can start doing prediction. Now when I can do prediction well, I can start moving more of that prediction function. Computers can do that better than humans. Now what matters is the things humans do well, which are judgment, and empathy, and connecting to another human, and understanding their needs.
The way, at least internally, we think about it is build systems that amplify humanity, not replace it. They amplify humanity by taking the rudimentary tasks off. We don’t ask a lender to calculate an ROE in their head. We can calculate that for them.
Jim Young: That article that Carl referenced just a moment ago will be our first topic we discuss next week when we bring you part two of our conversation on Artificial Intelligence and Machine Learning. That will wrap things up for this episode. Remember, you can always find more about today’s episode at PrecisionLender.com/podcast. If you like what you’ve been hearing, make sure to subscribe to the feed in iTunes, Sound Cloud, or Stitcher. We’d love to get ratings and feedback on any of those platforms. Thanks for listening. Until next time, this has been Jim Young and Carl Ryden. You’ve been listening to The Purposeful Banker.
The post Podcast: Artificial Intelligence – The New Digital Divide (Part 1) appeared first on PrecisionLender.