Mike breaks down some of the key AI solutions being used at BNY Mellon and shares his thoughts on the many possible future implementations of these emerging technologies. He also unpacks how their enterprise AI team chooses what to prioritize before addressing the importance of demystifying AI capabilities — both within and outside of an organization.
Joining us today to provide insight on how to put together a credible AI solutions team is Mike Demissie, Managing Director of the AI Hub at BNY Mellon. We talk with Mike about what to consider when putting together and managing such a diverse team and how BNY Mellon is implementing powerful AI and ML capabilities to solve the problems that matter most to their clients and employees. To learn how BNY Mellon is continually innovating for the benefit of their customers and their employees, along with Mike’s thoughts on the future of generative AI, be sure to tune in!
Key Points From This Episode:
Quotes:
“Building AI solutions is very much a team sport. So you need experts across many disciplines.” —Mike Demissie [0:06:40]
“The engineers need to really find a way in terms of ‘okay, look, how are we going to stitch together the various applications to run it in the most optimal way?’” —Mike Demissie [0:09:23]
“It is not only opportunity identification, but also developing the solution and deploying it and making sure there's a sustainable model to take care of afterwards, after production — so you can go after the next new challenge.” —Mike Demissie [0:09:33]
“There's endless use of opportunities. And every time we deploy each of these solutions [it] actually sparks ideas and new opportunities in that line of business.” —Mike Demissie [0:11:58]
“Not only is it important to raise the level of awareness and education for everyone involved, but you can also tap into the domain expertise of folks, regardless of where they sit in the organization.” —Mike Demissie [0:15:36]
“Demystifying, and really just making this abstract capability real for people is an important part of the practice as well.” —Mike Demissie [0:16:10]
“Remember, [this] still is day one. As much as all the talk that is out there, we're still figuring out the best way to navigate and the best way to apply this capability. So continue to explore that, too.” —Mike Demissie [0:24:21]
Links Mentioned in Today’s Episode:
Mike Demissie 00:00
But it's also important that we have diverse set of use cases to really show what's possible because nothing builds confidence and awareness more than seeing solutions live and in action.
Rob Stevenson 00:15
Welcome to how AI happens. A podcast where experts explain their work at the cutting edge of artificial intelligence. You'll hear from AI researchers, data scientists, and machine learning engineers as they get technical about the most exciting developments in their field and the challenges they're facing along the way. I'm your host, Rob Stevenson. And we're about to learn how AI happens. Here with me today on how AI happens is a man with tons of experience in the AI space. He was a managing director of architecture as well as robotics transformation. And the SVP of business transformation, lean robotics, and AI over at State Street. Currently, he is the managing director over at the B and y Mellon AI hub. Mike Demissie. Mike, welcome to the show. How the heck are you today?
Mike Demissie 01:05
Good to be with you, Rob. Nice to be with you.
Rob Stevenson 01:07
Yeah, really pleased to have you here on a Monday morning. Thanks for taking the time to do this loads to talk about because the NY Mellon has a ton of awesome applications of AI that you can explain to us. But you've also this company's been around for a long time, right founded by the $10, founding father, Alexander Hamilton, less. And so I'm always really interested when companies like this that have been around a while are also technology rising and adapting to the fast moving pace we're in as opposed to startups that come out of the gate with AI, lots more legacy things to change, and probably a lot harder to institute AI at a legacy company that's much bigger than it is a startup. And that's what I'm kind of looking forward to talk with you about today. But before we get there, would you mind sharing a little bit about yourself, your background, and kind of how you wound up in this current role?
Mike Demissie 01:53
Sure thing, Rob, I started out my career actually as an engineer, and my first opportunity was building hydrogen powered electric cars, which is quite different from what I do today. But that was an amazing experience, and have had the good fortune of working on emerging technology, emerging capabilities and different industries over the course of my career. So you know what you find me today, but there was a common thread along the way, which has been, I've always been really fascinated and interested in new technology and applied manner. So not just in a theoretical abstract way. But how can you use them in the best possible way and actually bring them to life, which is what I did in my first foray, and, you know, building out hydrogen powered cars, and what we do today in applying AI in financial services. So it's an awesome opportunity to be in and you're constantly learning and new capabilities, and I'm making good use out of them. So that's the makeup of the team. So our goal is twofold. One, apply this powerful capability across the organization to solve problems that matter for our clients and our employees. And to make it easier for the rest of the organization to tap into this capability in a responsible way. So that means finding the right tooling, the right governance, the right framework, and mechanisms, actually, from an opportunity identification to solution building to implementation as well. So that's the mission of the central Enterprise team. And with the expectation that in the not too distant future, this capability is going to be embedded in everything that we do. So really, we're laying the foundation and showing by doing so we can get there faster in the most responsible way.
Rob Stevenson 03:42
Gotcha. What about the Mike demissie of it all? Where do you come in.
Mike Demissie 03:45
So given we have these very diverse, complementary set of businesses across the organization, it's really important that we have one cohesive approach on this very powerful transformative technology and capability, how we are going to charter across organization. So my role in that is really assembling that central team that allows us to demonstrate that in a credible way, help our colleagues actually see the opportunities and their respective organizations and functions, and be effective and actually transform our services for our clients, taking advantage of these amazing capabilities in our hands. So that's my role, and really inspiring folks and getting the right practitioners at the right problem sets, so we can be ready for our clients tomorrow.
Rob Stevenson 04:37
I want to ask you a little bit about your role specifically, just because it's an exciting time to be in the AI space. We're seeing lots of different ways to bring the skill set to market whether it's you know, on the icy side and engineering or in data science, what have you. Your leadership role is interesting to me too, and I was hoping you could kind of explain a little bit about you know, quote, office space, what is it you'd say you do here how Have you kind of characterize your role? Sure,
Mike Demissie 05:02
maybe before I dive into that, it probably just for the audience's benefit, just to give a little bit of context about being white male, and as a firm, as you said, you know, in founded by Alexander Hamilton, so we've been around for a long time, 239 years to be exact. And, you know, for us to really be there for our clients, we need to continue to innovate. So it's something that we truly see as a core as part of our DNA. And when you look at us as a firm, we play a unique role in financial markets, we touch more than 20% of the world's investable assets. Were the largest custodian with more than 46 trillion assets under custody or administration, as well as the sole clearing agent for US government securities, that's, you know, more than $10 trillion every single day, our payment business alone processes more than two and a half trillion dollars worth of transactions every single day. So everything that we do, we do it at scale. And behind all these transactions and all these activities in our clients, looking for insights, how we can actually help them to make more data driven decisions. And that's how we see our role increasingly with these capabilities, you know, where they will take us so our clients and us will be able to make more data driven decisions with AI. And that's why we're quite excited about that. So we have a central team, that is really finding the right problem statements across the organization to apply these emergent capability. But at the same time, demystify the capability make it real for people see what it can do. So we can really actually identify with more and more opportunities across the services that we offer, who
Rob Stevenson 06:43
comprises the enterprise AI team, like what kind of skill sets make up the core of the same?
Mike Demissie 06:49
Sure, building AI solutions is very much a team sport. So you need experts across many disciplines. So at the heart of it, we have data scientists, and these are specialists, you know, typically with Advanced Math and Science Degrees, who with a solid grasp of state of the art algorithms in this space, you also need domain experts with deep understanding of the underlying business, the controls the processes the KPIs, and you need product and design experts who play an important role in identifying the unmet needs, outlining a solution, and being able to envision the best experience for the end user as well. So that's an important role. You also need experts in data as well as engineering who can really find the right way to connect to the right sources, and applications and deploy these solutions as well. Another important role I should mention is model risk management, which is people who have some expertise from an understanding of the data science, who can help us actually scrutinize do independent validations of the solutions before they go to production as well. So it really is a very much a team sport. So you have all these different folks, not all of them need to sit in the central team, for instance, model risk is as an independent function that sits in a second line, but you really need everyone to, you know, take an AI solution from an idea to implementation. Since
Rob Stevenson 08:18
it sounds like this team's product, if you will, is solutions for within the business, they can maybe even run like a business within the business, right? If they are serving like lots of different areas of the company, in ways that other departments tend not to write their departments, you know, tend to be a little siloed and then have less Interplay just as a base, as at the base quality of their existence, I guess, this team goes all over the business. So it's curious that they they can kind of operate in this way. And I think it's kind of exciting. Like, it's you have like this internal consultancy, if you were on this team, you get a different kind of challenge every day. And you're it sounds like your directive, if you're on this team is just like, go out and find ways to make people's lives easier using various AI and ML tools. Is that fair to say?
Mike Demissie 09:02
That's fair to say? I think it's encompasses many aspects. One part of it is really identifying the right opportunities, as you said, Rob, to deploy this capability and solve problems that really matter that matter for our clients that matter for our employees. It also means building that solution. So there's going to be so the data scientists that I spoke of earlier, would need to look at whatever is available in their arsenal and see what's the best way actually to build a solution and deploy it in a scalable fashion. The engineers need to really find a way in terms of okay, look, how are we going to stitch together the various applications, you know, to run it in the most optimal way. So it is not only the opportunity identification, but also developing the solution and deploying it and making sure there's a sustain model to take care of it afterwards after production as well. So you can go off to the next new challenge. Maybe that one way to kind of just bring it to life is you know, and share with you some of the use cases, for instance, that we've been working on. As I mentioned, you know, everything that we do being at scale, if you look at, for instance, in one of our solution, it is an AI assisted trade settlement analytics, a small fraction of trades generally fail to settle on time. But that's a costly proposition for clients. There are potential fines in certain markets. So given that we process more than 100 million trades, you know, any given year, we felt we had enough sufficient data that we can actually look into for any underlying patterns that would help us identify whether a trade is likely not to settle on time, so you can take action. And that's exactly what we've done. So we have an AI solution that gives clients a real time insight, which trade maybe having, you know, risk of certain settlement issue. And this allows them to actually take action instead of finding about the facts in after the fact and avoid costly settlement failures. So that's one, we're also using predictive analytics. for our own use internally, we have several solutions focused on a segment that we call AI assisted liquidity management, this is enhancing our forecasting abilities to make better informed investment decisions, how we manage deposits, and other things. And this is, you know, liquidity management forecasting is a familiar challenge for clients as well. And our clients are actually looking to leverage what we have built for our own use. And that's something that we're actively exploring as we speak, changing gears a little bit, you know, giving you a completely different example, that we have an AI assisted securities lending solution, this is helping our traders determine optimal pricing for, you know, to generate more yield for our clients. Because we, you know, one of the roles that we play is lending those assets on their behalf, so they can generate additional income, and being able to have insight on the price movement allows us to do that at higher utilization. So there's endless use of, you know, opportunities, and every time we deploy each of these solutions that actually sparks ideas and new opportunities in that line of business. So to your point earlier, you know, that really becomes a connecting thread across the organization. Given
Rob Stevenson 12:23
that there's endless opportunity, how does the team prioritize and decide where to spend its time?
Mike Demissie 12:30
That's a very good question. And it says, it's actually, you know, not an easy challenge. I think, going back to the heart of the opportunity identification, it's a very much a collaboration effort between the AI practitioners, the product specialists, engineers, and other team members. And we, in all cases, we ask upfront, what's the value of the solution? What would be different? How would it work? What would you do with that in for instance, in a predictive signal, and what happens when it doesn't work? That's also an important thing that we have to have an answer for, what controls do you need to have in place. So we scrutinize all of that upfront, and it's very much a hypothesis based iterative process in terms of you know, how we identify the opportunities and prioritize the right ones. With emphasis on impact, we have a framework that gives our partners our stakeholders guidance on the type of problem patterns, these capabilities tackle. I mentioned a few of them in the predictive analytics space, there are capabilities in anomaly detection. So you can really, actually, you know, find something unusual and large amount of data that can help you to manage risk better, there are capabilities that help you to process unstructured and large volume of information that may be coming to us can be client instructions. And queries are how you can really get to the heart of it, and you know, action it as well. But it's also important that we have a diverse set of use cases, to really show what's possible, because nothing builds confidence and awareness more than seeing solution live and action. So we're very intentional about that as well,
Rob Stevenson 14:08
that is a really interesting piece, because it's so clear to the individuals developing it, or you, for example, why someone should want the solution, or why the solution works. You have the challenge of you mentioned earlier, making sure that it is utilized properly. It doesn't just become you know, shelfware, so to speak. And then also there is like this need for internal PR, kind of because, again, like the people listening to this podcast, and the people who work on this kind of tech understand intimately why it's so good. But then you know, the your other people in non technical departments or who have not really seen what this can do, might still need a little bit of convincing. So what does that process like for you? How are you kind of going on that internal PRs and make sure people understand and then see the opportunity in the same way that your team does?
Mike Demissie 14:55
That's a really unimportant part of the journey. We talk about it as a demystify In this capability, and the best way to we find ourselves to demystified is actually talk about what problems it solves, it's really not about the technical jargon, or how the sausage is made, there would be a small segment of the population that would be interested in that. And we make all the resources available for Pete, if people want to dive in deeper, but really abstracting that away, and talking to people where they are and say, This powerful capability, here's what it would mean. And your specific area, here are the type of problems actually that we managed to solve. And having that continuous engagement, we really find it to be quite, quite helpful. And we've been getting similar feedback from our colleagues across the company, not only it's important to just raise the level of awareness and education for everyone involved, but you can also tap into the domain expertise of folks, regardless of where they sit in the organization. Because once they understand that, they can actually better guide you, they can guide you better in terms of where it can be used, as well. So really, and we're seeing, so the fruit of that is actually more and more problem statements and powerful problem statements. And we're seeing that lists grow by, you know, every month, and it's something that we continue to invest in. So you're absolutely right, demystifying, and really just making this abstract capability real for people, there's an important part of the practice as well.
Rob Stevenson 16:29
You mentioned, for those people who do care how the sausage gets made, you do also provide the resources for people to go check out, what does that typically look like?
Mike Demissie 16:38
Yes, so it's a combination of access to our practitioners, actually, where and where you can really come to, and really go at a much more technical level, how a solution is built. So you have that we do host the what we call these, this for our community of practice, so everyone in the company can join. These are free forums, some of them are tailored towards the business audience, some are more on the technical side of things. And you can self select in terms of which one you want to go to. So if you're curious about getting into this algorithm selection, what are the trade offs that people are making? What are we excited about, actually, some of the emerging capabilities like Gen AI, it's almost impossible to talk about AI and not going to judge AI these days, then there are places where you can go to and do that. At the same time, we also encourage our community and practitioners across the organization to share because each of us, I mean, come across useful insight, a different perspective, a different point of view. And we really find it valuable to really actually have that debate that discussion as well. So there are ways for people to share different papers, different, you know, insights, and lectures, or even some events that are happening that you know, so everyone can tap into that as well. So that's really an important piece of us continuing to stay current, which is essential. And this fastly, fastly evolving space.
Rob Stevenson 18:09
So for those of you out there who had generative AI on your house, AI happens podcasts bingo card, you can go ahead and mark that off. But it is everywhere. Right now, like you say, I'm curious is being why Mellon going to be utilizing it.
Mike Demissie 18:22
Look, it's very early days, I think, if I take a step back, there's really good reasons behind the excitement about generative AI, right, like, I mean, when you think of it, it's really trained on general knowledge, making it very versatile across domains, and it's very multi dimensional. And you can make it intelligent on your specific domain on your proprietary datasets and make it very powerful assistant and ALA Lansingin a large language models have revolutionized how we interact with AI, right? I mean, we can simply ask question in a natural language way and get an answer. Being able to engage AI conversationally in has made it more accessible. And you know, it's public availability is democratizing AI, access and utilization. So there's really good reasons why people are excited about it. And you can count us in that category as well. So recognizing still is early days, we see the potential for to enhance being 1 million products and services and a number of ways and it can enhance our clients and employees experience as well. So we have some, you know, we have a growing list of actually potential use cases, including, you know, streamlining, knowledge management capabilities, making content easily accessible for employees or clients. And then as you can imagine, this can be done for training purposes for customer support and others, assisting our engineering efforts in coding, right, like can do really a remarkable job and it can accelerate that effort and application development. Being able to summarize and distill down information for actionable insights, again, it can accelerate those types of activities as well. And also generating first drafts of, you know, various types of documents. So, we are really excited about the possibilities. We're exploring it in a very deliberate risk managed way. So we can really get the fruit of this type of capability. But also being mindful of what any incremental risks introduce us and how we can manage that better.
Rob Stevenson 20:27
I'm glad you said first draft, because that, to me assumes like, okay, it's not published ready, it's not ready to stamp and prints. Like, let's put a human in this loop and make sure this is good. However, the fear based AI, headline du jour is that language learning models are creating this content that is now going out on the internet. And now AI generated content is becoming part of the corpus training those same models, and they can't seem to tell it apart. Are you following this narrative at all? Yes, yes. So I guess I only have a question, but I just wanted to gather you react to that sounds like what you said, like we're looking at a first draft Here you are, you know, weary of this possibility for you don't want this generative to just like run rampant, but outside of being yml. And I'm curious, in your take on that, is there some kind of guardrail that you could foresee that would be like, Hey, how do you stop an AI from training itself on its own output?
Mike Demissie 21:22
Look at any powerful capability? You know, you really have to think about what are the proper guardrails that need to be in place, right, and some of those guardrails may be in a fairly obvious for everyone to see, for instance, how the need for us to control against bias in terms of you know, like, the data that can be powering the, you know, these types of solutions. So we always talk about the responsible use of AI. And so we're very mindful of in terms of, you know, what, that we need to be aware of those risks and put the right, you know, guardrails in place. I don't think we have all the answers today, in terms of you know, what those things would be? Because, as we talked about earlier, we're still discovering, what are the potential applications of it? What are the potential uses of it? And when you do that, then you would have to look at and you have to ask yourself in terms of what can go wrong, you know, what can go off rails? And what are the mechanisms that you would need to put in place me to, you know, protect yourself against that. And enterprise applications, that very common theme is we always talk about it as an assistant. So it really implies there is a human in the loop. It really is just accelerating, increasing our productivity. But we have the right controls and guardrails to make sure like the decision making and everything else is checked, and you have the right checks and balances in place. But when you're talking about looking at the future, how this capability can evolve. I don't think we have all the answers. I think it's just required, all of us to keep asking the questions, having the right dialogue, bringing the constituents and finding what those guardrails would be. So that's what we value and something we'd like to see more of.
Rob Stevenson 23:05
This is a challenge because someone like you and a company like being why Mellon cares enough to put up those guardrails and to really understand the way technology is taking shape. But just, you know, a random person with a mid journey account isn't probably not right, it's kind of impossible to just ensure everyone using this technology has the sort of respect for it. So I don't know if I have a question here, it's just a problem. It's like when with large organizations like the NY Mellon, I am a little more assured. But when it just gets into the hands of the public, who knows how someone's going to push that button.
Mike Demissie 23:36
That's why we need to continue to have this dialog and whether it is from the regulatory standpoint in terms of accountability for firms that are putting products out there. So there's multiple ways to tackle that. But I think it's the right time to have that dialogue and discussion at this point to really just protect ourselves from the downsides of these powerful applications that can also do a lot of good and can solve a lot of challenges that we have and you know, across society, yeah,
Rob Stevenson 24:07
yeah, of course. Well, Mike, we are creeping up on optimal podcast length here. And at this point, we just say thank you so much for being here. This has been really, really fun having you before I let you go, I just want to ask one more thing. For the folks out there forging their careers in the AI ml data science, etc. Space, what advice would you give to them?
Mike Demissie 24:25
Sure. The first one is really curiosity and continuing to learn. Remember, still is day one, as much as all the talk that is out there, we're still figuring out the best way to navigate and the best way to apply this capability. So continue to explore that to less have bold vision, and a lot of times with these type of transformative capabilities. The first thing that may come to mind may be some incremental changes, but the capability can do far more than those incremental change. Just Just take us back to the early days of the Internet, right? Like I mean, so if you look at the initial applications, they were basically trying to replicate whatever we had in bricks and mortar oil into the online world, and not necessarily completely imagining what can be possible, what new models, what new offerings can be made possible with the internet, I would say it's the same thing with AI as well. So I think the need to really just explore beyond the immediate future in terms of, you know, work that can take us as well. So you know, be bold, and explore this capability. Third pieces is really, really important to going back to the piece that we talked about, to have an expansive view, and asking yourself, you know, what can go wrong? And how can we protect ourselves? How can we really do the right thing for society, just regardless of where we sit. So I think that's not a responsibility, we can outsource and just give it to one government entity or regulatory body. I think it's an all of us as practitioners to find the right way of using these capabilities. So that would be that my third ask Rob,
Rob Stevenson 26:05
that's great advice. Mike, thank you so much for being here today and sharing a little bit about what's going on over there at being yml. And and just your general takes on the space. I've loved learning from you today. So thanks for being here.
Mike Demissie 26:15
Thank you, Rob.
Rob Stevenson 26:16
How AI happens is brought to you by sama. Sama provides accurate data for ambitious AI specializing in image video and sensor data and notation and validation for machine learning algorithms in industries such as transportation, retail, e-commerce, media, med tech, robotics and agriculture. For more information, head to sama.com