Executive Interview: Jordan Jacobs, Chief AI Officer, TD Bank Group

897
He co-founded Vector Institute to help keep AI talent in Canada, is
Applying deep learning to enterprise data within TD Bank,
And is delivering explainable AI within a regulated bank environment

Jordan Jacobs is cofounder of Layer 6 AI and was co-CEO prior to its acquisition in January 2018 by TD Bank Group where he is now Chief AI Officer (Business & Strategy). Jordan is also a founder of the Vector Institute for Artificial Intelligence, a concept Jordan initiated with Tomi Poutanen, and Professor Geoffrey Hinton, and Professor Richard Zemel.

Jordan previously co-founded Milq, a cultural content discovery platform that was an early adopter of using machine learning for personalization. Jordan spent 15 years as a tech and media lawyer, advising tech entrepreneurs, Grammy and Oscar winners, and sports teams in complex transactions and financings, first at a large firm in Toronto and, for 10 years, at his own firm. Jordan received his JD from Osgoode Hall Law School in Toronto. He recently spent a few minutes with AI Trends Editor John P. Desmond.

Jordan Jacobs, Chief AI Officer, Business & Strategy, TD Bank

 

Q. Layer 6 was acquired by TD Bank in January 2018. Could you talk about why you sold the company to TD Bank and can you describe the mission of Layer 6 today?

A. When you’re six years into a startup there are many different considerations when you’re thinking about how to go forward. Our primary focus was to continue to do the most interesting work possible. If you want to hire the best people, how do you keep them interested? And truthfully, as a software company, it’s a challenge because you do one or two things incredibly well, and then you hope to sell them 10 times and 100 times and 1,000 times over again and that builds an incredible business. If you’re one of the best people in the world doing something that is really, really unique, that can get boring.

So, selling the company to TD allowed us access to a large amount of  data and many different use cases so that our team could work on not just one or two or five things but ultimately tens and even hundreds of things over time. It was very important to be able to satiate the intellectual curiosity of our team and to make a much bigger impact by doing that on a bigger platform.

And as for the mission, inside TD our goal is really to change the experience of how people interact with a bank so that your experience is personalized. It’s like your relationship with the bank going back to the ’50s, ’60s, or ’70s when you walked into a local branch and the branch manager knew you and your family, they knew your business, and they knew how to help you on a personal level. Right now, bank customers interact with many different products and in many ways. However, they almost never do this in person. It would be impossible to offer that level of service without AI connecting each of those interactions together and acting as a so-called ‘brain’ in the background. Then the bank becomes your trusted service provider on the financial side of your life; it anticipates your personal needs, and it helps you solve them in a way that you didn’t have to ask for necessarily.

We’re also working on things outside the bank, which we’ll be able to talk about more in the coming months. Those projects are not related to banking at all, but they were important to us when we sold to TD. We needed to have TD believe in the mission that we could use the talents of our people and AI to change things on a societal level. We’re particularly focused on some issues in healthcare.

Q. It will be good to hear more about those when the time comes. Could you talk about your role at Layer 6 and TD Bank, and how about how you communicate using AI to the larger organization, your title and who you report to?

A. TD is a very large bank, one of the biggest banks in the world. There are about 85,000 employees. It’s much different than our small startup the bank acquired. We’ve been able to maintain Layer 6 as a separate brand and physically stay separate from the bank, which allows us to retain the feel of a startup, which is very important. At the same time, they were very supportive of giving us a significant enough role so that we could be change agents across the bank. My partner, Tomi, and I were co-founders and co-CEOs of Layer 6. Each of us is also Chief AI Officer. I am Chief AI Officer (Business & Strategy) of TD Bank Group and Tomi is Chief AI Officer (Technology) of TD Bank Group.

In terms of the banking level, we’re senior vice presidents reporting to Michael Rhodes, who is one of the group heads. Michael reports to Bharat Masrani , TD Bank Group CEO. Michael is in charge of innovation, technology and shared services. Reporting into that group in the bank allows us to touch many different business units. We’re essentially one of the core functions that is shared across the bank. TD is a leader in digital banking applications in North America and the expectation is that will only grow over time.

Q. Could you talk about the primary AI technologies that Layer 6 uses and what types of application you’ve had the most success with the bank?

A. Our innovation was to take deep learning, which was largely invented in Toronto, where my two co-founders studied with Geoffrey Hinton, a technology that is being used to power computer vision for self-driving cars and speech translation in your phone, and apply deep learning to enterprise data. We built a software system that could do prediction and personalization in one end-to-end solution, as opposed to having to create a hybrid system.

We did that in one system with deep learning at its core, but the flexibility of the system is that we can take in all different types of data and use different methodologies. So convolutional neural nets can be trained on image data, but we can also take in, for example, audio data, textual data, geo data and time series data — we made some significant innovations on that front. It can then be trained end-to-end to produce an output. A software layer is built around that for deployment on-premises or in any of the clouds, working in real-time on a massive scale.

All of those things are a challenge and it took us essentially five years to do that work, but the result of it is that we have a very unique software package. Right now, it’s being used exclusively inside TD, and it’s being used across the bank for lots of different applications; for example predicting customer needs when it comes to buying a home and how to best serve them. Also, predicting loan defaults, powering personalized experiences through the digital app, and anticipating customer needs and delivering personalized experiences throughout  their financial journey. The software integrates the different customer touchpoints whether it’s through an app, in person through a wealth adviser, or on the phone back into the central brain, with a repository that connects all the different data touchpoints.

Q. It sounds very impressive. Let’s talk about a major challenge today with machine learning, which is the difficulty in explaining the reasoning that a decision has been made or why it was made and how it was made. How does Layer 6 enable its predications to be explained and how useful is it?

A. This is something that has been a very big challenge in deep learning, which is often thought of as a black box. For us, being focused on banking was a problem because the regulatory requirements necessitate that you can explain why you made the decision that you did. We must ensure there’s no bias in turning down people for loans, for example. It was very critical that we focus on the ability to explain our predictions. We put a lot of energy into it and developed a system that is extremely comprehensive. It is very accurate at explaining why a decision was made and what factors went into it both on a macro level, from a cohort of all people or a big segment of the people, and right down to an individual level.

And I can’t get into the technology behind it but it’s very accurate. We’re able to use it on any of the predictions that we make, whether they are personalized recommendations or other kinds of predictions. It has been a very important factor for us being able to work with the different business units inside the bank. We’re able to show them why things are happening the way that they are, rather than just asking them to trust it. People are not generally satisfied when you tell them it works and not to ask why. They want to know why. So being able to use explainable models to show why something happened is very important.

We wanted to make sure that we test it in lots of different scenarios and so far, it’s performed exceptionally well. It’s an innovation that we’re very, very proud of; it’s been very difficult for lots of other organizations to get there. It makes a big difference, particularly in a regulated industry like banking. It’s able to prove that the decisions you’re making are the right decisions. Not just from the outcome the AI is producing but also, it’s unbiased, it’s fair, and what the thinking was that went on inside the machine.

Q. We will be interested in learning more about the wider availability of that capability in due time. Shifting to the race to become a global leader in AI, Canada was the first country to release a national AI initiative. You’ve played an important role in that as one of the founders of the Vector Institute, which works to advance AI research and commercialization. Could you talk about the personal path that brought you to AI and the role that you played in the initiatives that Canada has taken?

A. Layer 6 is essentially the technology that we built inside another startup we created in 2011 called Milq. The idea behind Milq was to use machine learning, deep learning in particular, to provide personalized content delivery and discovery around culture, specifically music and video content. These things that were not objectively quantifiable; they are taste-based. Someone might like one song but not another song by the same group, might like one group but not another group that other people would think fits naturally into that category. Similarly, on the film side or TV side or food or fashion, these things are extremely personal and nuanced. I had been fascinated by what was going on in Canada in the development of AI that was not being used commercially.

Our idea was to build a company that used AI to understand people’s taste and to deliver cultural content to them based on their extremely unique, nuanced taste. This was before the big tech companies were using AI for this. Our co-founder of Milq, Don MacKinnon, is still running the company in New York, and we were fortunate to be introduced to Tomi Poutanen who is Layer 6’s co-founder and who had studied with Geoff Hinton back in the ’90s. He later worked in large search organizations at Microsoft and Yahoo on the West Coast. And Tomi had a lot of experience building what became the biggest machine learning platform in use when it launched in 2011, which is Bing’s search engine ranking algorithm. Our first hire was a PhD from the machine learning lab at University of Toronto that Geoff Hinton led. We really spent a lot of energy focused on these problems.

Along the way, ImageNet [image database] happened, which caused the big explosion of interest in AI. Over time, Geoff Hinton’s graduates ended up being the heads of AI at many of the world’s leading tech companies, such as the head of AI at Apple, at Facebook, at Uber, at Elon Musk’s OpenAI organization, and lots of other organizations that all came out of this one lab in Toronto. And as an AI company based in Toronto, that was fantastic except that we saw that a lot of the talent was leaving. And we also knew that suddenly there’d been an explosion of global interest. Companies, particularly the West Coast tech companies, were buying up the talent and then the professors who were training that talent. So, there are even fewer people coming into the field because the professors were starting to disappear from schools.

So for us, it was how do you double down in this area that was really started here, and how do you build a top five research organization globally? How do you massively scale up the number of students to meet the demand? We devised this concept that became the Vector Institute by bringing it to Geoff Hinton and Richard Zemel who’s running the machine learning lab at the University of Toronto. They agreed we needed to do this, and we jumped in together, widened it out to some of the other key faculty in the University of Toronto — Raquel Urtasun, Sanja Fidler, Daniel Roy, David Duvenaud, Roger Grosse and some others. We got lucky and met the former CEO of TD Bank Group, Ed Clark, with whom we worked very hard to try to raise money and do this in a way that would solve some of the problems both locally and globally in the field.

We were very fortunate to ultimately be supported by federal and provincial governments in Canada. Our efforts spurred the federal government into this pan-Canadian AI Strategy, which also invested in Montreal and Edmonton (which are world-leading centers as well). What came out on the other side is an explosion of growth at the research level. Now, we have MILA in Montreal being led by Yoshua Bengio, and the Amii [(Alberta Machine Intelligence Institute] in Edmonton being led by Richard Sutton, who is one of the founders of reinforcement learning that DeepMind is based on.

Those three organizations have been able to hire many more people to scale up the number of students they’re training, and around that, now we’ve seen an explosion of commercial activity, particularly in Toronto and Waterloo and Montreal. It’s very gratifying to see that. It was a lot of hard work to be doing this at the same time as we were birthing Layer 6 out of Milq and running two companies. But it’s been really an incredible experience to be part of and witness.

Q. Congratulations, it was a good plan.

A. Thank you.

Q. We talked about machine learning being invented at the University of Toronto led by Geoff Hinton, and it’s been in use for some 25 years and has seen a lot of evolution. How far can the roots of machine learning take us before another new paradigm in AI algorithms will be needed?

A. We have a long, long way to go, but Geoff would be the first one to say do something better. We do feel that with deep learning, that we’re in the first inning of the baseball game, so to speak, in terms of the impact this is going to have in the world. We haven’t solved self-driving cars yet; that’s one big application of it. We’re at the beginning of transforming healthcare from reactive to proactive, which I think is the transformation that will be the biggest one coming out of AI. It’ll be an incredibly profound thing for everyone around the world because it will mean that people in rural areas have better access to healthcare, and it will extend people’s lives so they won’t just live longer but live healthier. I think those applications are just at the very beginning.

Transferring all the research that has been going on a long time into applied work is at a very early stage. We have a long way to go to realize all the benefits that this will bring us. At the same time, researchers don’t sit still. They’re looking to push the boundaries on the next thing and the next thing after that. So even Geoff is doing that; it’s interesting work that pushes beyond what he’s done before. Progress doesn’t stop, it will continue, and no one is resting on their laurels. We are at the very beginning of an incredible time in human history.

Q. Finally, do you have any advice to young people or mid-career professionals interested in pursuing a career in AI?

A. I think people should absolutely go for it. This is a field that will only grow in importance academically, on the research side and in companies. If you are a young person looking at doing a startup, joining an existing startup or a scale-up, or joining an enterprise that is doing this or if you are going into school for it, I strongly encourage it. For the foreseeable future, it is the leading edge of tech. For people interested in creating the future, it is absolutely the place to be.

And that holds also for mid-career professionals. There’s no reason why people can’t make a move into a different area mid-career. It does take the sacrifice of having to go and put yourself out there. Go to conferences, meet people who are in the field, ask them if you could take them for coffee. Not everyone will say yes, but some people will and that can help start to open a door.

For people who are looking at a career change, an incredible amount of literature is out there, both professional literature such as research papers and on social media. Most of the people in this field are very active in sharing the developments that they’re interested in. It is easy to follow key people on social media. If you’re diligent about it, you can quickly start to learn a lot and get a sense of who the people are and what the technology is that is shaping the future. So, I think for both young people and people mid-career this field is wide open and going to grow enormously. It’s really the place to be focusing your energy if you’re interested in the future.

Learn more at Layer 6 AI.