Get new episodes right on your device by following us wherever you get your podcasts:
Click for the podcast transcript
Artificial intelligence or AI is the science of programming computers that can reason or think like humans. But in order to really understand AI, first we need to think about cats. At least according to Scotiabank’s resident AI expert, Yannick Lallement. He’s here this episode to outline what exactly AI is and how it can change how we think about work and do our jobs.
Key moments this episode:
00:16 — How cats are key to understanding AI
2:39 — A quick definition of AI
3:05 — Two examples of where you may already be using AI in everyday life without knowing it
4:48 — What is ‘augmented intelligence’ and how could it help us do our jobs more easily?
6:00 — How do banks use AI?
8:59 — What is ‘ethical AI’?
10:59 — Three ways AI could change how we approach work
Transcript:
Stephen Meurice: Artificial intelligence or AI is the ability to program computers that can reason or think like humans. But in order to really understand AI, first we need to think about cats.
[cat meow]
Yannick Lallement: So, let's take an example, imagine you want to create a program that will recognize cats in pictures.
SM: That’s cat enthusiast and AI expert Yannick Lallement.
YL: So, you give the program a picture and it has to tell you whether there is a cat in it or not. If you're going to program that by hand, what do you have to do? You have to start considering all the different ways that a cat can look like, right?
SM: Sounds simple enough. That’s how computer programs were made until AI came along. Pretty soon though, you start running into problems.
YL: A cat can be white, black, brown. It can have stripes, it can not have stripes. It typically has two ears, but maybe we can see only one ear. It has legs, usually four, but maybe it's a three- legged cat. There's lots of considerations to take into account if you want to hard code that. To the point that it's nearly impossible to hard code or hand code that.
SM: And that’s where AI comes in. Actually, a subset of artificial intelligence called machine learning or ML.
YL: Machine learning on the other hand, is going to perform that task by learning from a list of examples. And here your examples are going to be a whole bunch of pictures of cats.
SM: This method of learning from examples is kind of like how a kid might learn.
A two-year-old doesn’t need to have studied a textbook about cat characteristics to be able to say, “Hey look, a cat!” when they see a new one.
YL: That’s exactly it, yes. A kid would probably learn from one or two pictures of a cat. And ML is not at this point yet. You still need hundreds of pictures of cats. But that's not too hard to come by, right? So, you will be able to give your ML model hundreds of pictures of cats, the ML will learn and then you will have a model that can decide if a picture is a cat or not.
SM: Okay, so today’s episode isn’t just about cats. Yannick, who you’ve been listening to just now, is the Vice President of Corporate Functions Analytics and AI/ML Solutions at Scotiabank. He’ll explain why AI can change how we think about work and do our jobs. And how you may already be using it in your daily life. I’m Stephen Meurice and this is a cat.
[cat meow]
I mean, this is Perspectives.
Yannick, welcome to Perspectives
YL: Thank you, Stephen. Thanks for having me.
SM: Okay, in the intro we heard you give us a great breakdown of what AI is, using cats no less, maybe we can start things off again with just a reminder quickly of the definition of artificial intelligence?
YL: Yes. So, AI and more specifically ML — machine learning, is a way to create computer programs from examples. As opposed to create computer programs from scratch using a software developer.
SM: Okay – are there any other examples of artificial intelligence being used in the real world that we may be experiencing without even knowing it?
YL: Basically, AI is touching more and more aspects of our day-to-day life. So, you can see for example, each time you use your phone. The phone is going to predict a lot of things for you to make your experience using the phone easier. So, two examples. When you chat with your friend and you type, “how are.” The keyboard, the phone keyboard is going to show a button that says, “you” right? That's using machine learning. It has learned that the word “you” is often following the word, “how and “are.” And so that makes it easier for you to use the phone because you can do the same thing but you are going to type in fewer taps. Another thing the phone does for you using AI/ML is predict the next application you want to open. So, it uses your past habits, you know where you open applications, what time you open applications, what applications you open after another one and so on. It uses your own data to create a model of what is Stephen likely to open as their next application. And it's going to show you the predicted apps. And for me most of the time, the app I want to open is there. So, I don't need to rifle through my screens or my folders to find it. That's an example of ML. For artists, there is a brand-new category of model that can generate images based on text input. So, you can generate a picture of an astronaut on the moon on a horse, for example. In the past you would have to actually paint this or you know somehow draw this in graphic software. Now you can have AI models that have learned to associate pictures to words. And when you give them your own words, they are going to generate a picture that will match more-or-less those words. It's a lot of fun to use.
SM: So even artists are going to be put out of work by AI.
YL: I wouldn't say they are going to be put out of work, but it's likely that their job is going to change, right? Instead of working really hard to create from that picture of an astronaut on a horse on the moon. You can try the model to generate what you want and then tweak that. So, you start from something that is not from scratch and that's going to make your job actually easier. I don't think it will necessarily replace the job. And that's the general theme in AI/ML. A lot of times we create models that make people more intelligent in a way. That's AI as augmented intelligence. You have the AI taking care of some of the cases for you. Typically, the easier cases. And the person does the harder, more interesting things. So, the AI is really here to augment your own intelligence. I think that the artist situation is an example of that.
SM: Right. So, it gives people or jobs the ability to do some things very easily so they can focus on more advanced or more complicated things.
YL: That's exactly it. And we take advantage of that at the Bank quite a bit.
SM: Ok, well that's a great segue to my next question, which is why is a financial institution interested in AI? How do banks and other financial institutions use artificial intelligence or machine learning?
YL: There's a few ways. One is making it easier for our employees to do their job. One big project we have, for example, has to do with reviewing mortgage documentation that our customers provide in support of a mortgage application. In the past it's been done completely manually by employees that validate that the document that was provided to support the income is valid. And of course, the number that we have matches what you need for the mortgage that you are applying for. So, we can now use a category of AI/ML called intelligent document processing and we can use AI to look at those documents on behalf of an employee. And what the AI is going to do is essentially validate everything that can be validated, but surface to the employees the exceptions. The things that either of the AI could not find or could not understand or identified as being a discrepancy with the expectation. And then the employee can focus only on that part. And that's the more interesting part, number one. And number two, it makes their job faster, right? They can do more at the same time. And we use AI for customer satisfaction. So, we for example have another large project at the moment, which is a chatbot. The idea here is that customers want to self-serve. They don't want to have to call us or visit the branch. They don't want to wait in line. They want to have the answers they are looking for at their fingertips. And the most requested feature for our mobile app was the ability to chat with the bank. And so, the first part of the chat when you go to the chat in the mobile app is essentially a chatbot. That is going to see if the chatbot itself can answer your question.
SM: So, there are certain types of questions that come up repeatedly and so the chatbot can recognize those and has the answer.
YL: That is exactly it, yes. And we have observed so far that roughly half of the sessions that customer start with the chat are contained within the chatbot. So that means the chatbot is able to recognize what they want and give them the answer. Now, if the chatbot was not able to understand the question or was not equipped to give the answer, you will still be transferred within chat to a live agent, you will be able to chat with a person. So, it's a good way to have essentially the best of both worlds. Blending the human capabilities with the AI capabilities.
SM: How is a chatbot powered by a I different or better than a normal chatbot in terms of how machine learning plays into it?
YL: The chatbot we have created uses the latest generation natural language processing models. Natural language processing is the part of AI/ML that deals with understanding what people are talking about. And it is for that reason that it is pretty good at understanding what our customers are talking about and it is able to automatically, on its own answer a large number of questions.
SM: Okay. From a consumer perspective, what do you have to be careful about when it comes to AI? We hear about ethical AL, what does that mean? And do people need to be concerned?
YL: So, any model you create based on past data is going to learn the biases of that data. Now it's not necessarily a big problem because we are talking about the model here and a model is essentially a mathematical object. And so, we have multiple mathematical techniques to control for that bias, to correct that bias, to measure that bias. So, we know exactly what we are talking about. And we can essentially, I’m simplifying a bit, but we can essentially eliminate it. You cannot magically erase bias in humans. Whereas correcting bias in ML models is a lot more tractable as a problem. So, it's really not something that people should be too concerned about, provided of course that the organization creating the model is itself an ethical organization. We actually are the first bank in Canada to develop what we call an ethics assistant. And the ethics assistant is essentially a checklist that works like a checklist for a pilot when they are going to take off. To make sure that the plan is configured exactly right. To make sure that no problem is going to develop later to make sure that you are ready for takeoff. The ethics assistant helps our AI/ML practitioners make sure that the models they are going to produce actually check out. That there are no ethical issues with them. That there are no biases with them.
SM: Right. What's it looking for exactly that you're trying to eliminate from the process?
YL: We talked about models typically learning bias from bias data. that can happen, right? The data we typically learn from has been produced by humans. But we specifically look at gender, race, all of those biases — age. And we make sure they are not there. If we find that there is a little bit of it, then we can correct for it again using mathematical tools.
SM: And what's the next big thing on the horizon for AI?
YL: I think that's a bit of a too general question because it's really a business-by-business question. For companies that make large equipment, for example airplane manufacturers, AI has a completely different role from anything we have talked about till now, right? AI can help the airplane manufacturer sift through the incredible amount of data that planes produce and they can use that data to detect developing problems and trigger preventative maintenance on the plane so that they don't have to be grounded. Computer programmers, their life is also going to be transformed because AIs are starting to predict what they are going to write next in their computer program. So, it's not always right, obviously, the AI cannot generate the entire computer program at this point. But the AI can again, augment you by predicting what you're going to type next in your program and make it faster for you to program. So, there's lots of examples. There's the artist example. For health care, if you want to classify medical imagery and you have a whole bunch of medical images and some of them have been diagnosed by a doctor as non-pathological and others have been diagnosed as pathological.
SM: Pathological means there's an issue, basically something that might need further diagnosis or investigation, right?
YL: Yes. So, what you can do is use machine learning to look at this entire set of images.
Some pathological, some not. That's your data set. And the AI, the machine learning is going to learn to recognize if a new medical image that it hasn't seen before is pathological or not. That's an example. The entire idea is that instead of having a software developer code something, which is not always possible. In the example of the medical imagery, nobody knows how to do that. But we can instead use AI/ML to learn from all of those examples and then create what we call a model that will then diagnose on its own whether new medical images, previously unseen before, are pathological or not. We see everyday new applications and it has started to transform the world, but I think it's going to transform the world much more in the future.
SM: Well, that's a really great place to leave it. Yannick, thank you so much for joining us today.
YL: Thanks for having me.
SM: I've been speaking with Yannick Lallement, Vice President, Corporate Functions Analytics and AI/ML Solutions at Scotiabank.
This is our last episode before we take a break for the holidays. Thanks to everyone who listened in 2022 and we’ll be back January 5th with new episodes.
The Perspectives podcast is made by me, Stephen Meurice as well as Armina Ligaya and our producer Andrew Norton, who is actually still working on identifying cat photos. Okay, how about this one? Andrew is this a cat?
Andrew Norton: No, that is a toupee, Stephen.
SM: No, that’s a cat.
AN: Huh, close though.
