Guy Nadivi: Welcome everyone. My name is Guy Nadivi and I’m the host of Intelligent Automation Radio. Our guest on today’s episode is Bret Greenstein, Vice President and Global Head of Artificial Intelligence for Cognizant Technology Solutions, a global MSP with 270,000 employees worldwide. Prior to Cognizant, Bret spent over 30 years with IBM where he led their Watson Internet of Things Offerings and with a pedigree like that, and given his deep domain expertise in the fields this podcast focuses on, we absolutely had to have Bret on the show and he was gracious enough to take time out from his very busy schedule to join us today. Bret, welcome to Intelligent Automation Radio.
Bret Greenstein: Thank you very much, Guy. It’s a pleasure to be here.
Guy Nadivi: Bret, you’ve spent most of your career with IBM and then you left Big Blue to join Cognizant Technology Solutions, and Cognizant while being a large global firm with over a quarter million employees, probably isn’t one of the first companies that comes to mind when you think of AI.
Nevertheless, Cognizant is taking an interesting approach to AI by applying Darwinian principles to its machine learning efforts with what it calls Evolutionary AI. Now there are currently three basic machine learning paradigms, Supervised Learning, Unsupervised Learning, and Reinforcement Learning. What are the main principles behind Evolutionary AI and do you think it might become the fourth paradigm of machine learning?
Bret Greenstein: I do. You could think of this as a natural extension of Deep Learning and Reinforcement Learning as well. You can think of it as an extension of that. The key principles are, and we’re able to build models even from almost no data, bootstrapping a model itself, letting AI learn from whatever limited rules or limited data you have to begin to create a surrogate of your business. And then to use the surrogate to explore possible solutions and then improve the model. So this evolutionary technique, genetic algorithms, allow us to take a model that might be reasonably crude at first and then refine the model, improve it over time, and then to use simulations with the model to create optimal decisions.
And you’ll find that most learning techniques usually find strong local optimal decisions, or they can exhaustedly search an entire space for possible solutions, but it can take an almost infinite amount of time depending on the complexity of the space. We use an approach that allows us to do population-based learning. So as we’re exploring the solution space, using our surrogate models, all of the simulation, we can explore every possible opportunity in parallel. And then the population learning allows us to learn from every exploration, stopping ones that are inefficient or suboptimal and putting our focus on ones that are more optimal. If I could just for a second more, the result of that is we can often learn with less data in significantly faster time to get to the same, you know, optimal outcomes, or better.
Guy Nadivi: Interesting. Now, there’s been a lot of hype around AI the past few years and in some quarters even fear, and when I read about the accelerating developments taking place in the field, it reminds me of what the economist Rudiger Dornbusch once said about economics where things take longer to happen than you think they will, and then they happen faster than you thought they could. With that in mind, what do you see as some long promised AI capabilities that will be happening faster now than we thought they could?
Bret Greenstein: So you’re right, it is a combination of them. We’ve all been watching this space for our entire careers, always one day away from the big breakthrough, but it is accelerating and it’s accelerating because we’re finding that instead of AI being generally intelligent and replacing all of us as leaders in whatever we do, it’s actually extremely specialized intelligence and it’s in those specializations that it’s moving very, very fast.
There was only a few years ago where the idea of machine vision could recognize a dog or cat, or hot dog or not a hot dog, these kinds of very simplistic cases, and now it’s already generally recognized that image recognition can provide better diagnostics of x-rays and radiology than a human being, more accurate. It doesn’t replace it. What it does is gives the radiologists new tools to find things they might’ve missed, or to validate things they assumed or figure it out themselves. So in many ways the specialization of AI has progressed much faster than I think everyone thought.
Look at conversational AI. We all saw, Alexa and Google come out and we’re like kind of impressed that you can talk to a computer. That’s been a promise for decades, but you can already see the advancements with Google Duplex for example, where the language is so natural that it’s, you know, can’t be discerned from speaking to a person which is now requiring solutions to identify themselves as AI when you’re talking to them. Who would have ever thought all that discussion about Turing machines and we would actually have to have systems tell you it’s an AI so you wouldn’t confuse it with a person. That’s where we’re at now.
Guy Nadivi: Okay. Now those are some interesting use cases, so I want to go back for a moment to Cognizant’s Evolutionary AI initiative. Can you speak about some of the more interesting use cases your team has applied Evolutionary AI to, and the results you achieved?
Bret Greenstein:Of course. So let me, before I do that, let me just back up on really your opening. When you think of AI, AI has its technology foundation layers. What the APIs and services for natural language processing or image recognition, or other specific cases available through all the cloud providers along with the tools to build up an AI-based system and feed it with the data that’s needed.
These are generally provided through cloud services and cloud providers and the advancements in those are accelerating every day. Amazingly broad movement in technology. Then you’ve got companies like ours which apply AI to business problems. So we take the technologies available there, we apply our own IP use cases, industry knowledge, and then deliver outcomes. So that’s what makes us unique in the space of AI is that we have IP like Evolutionary AI. We work with the technologies from the cloud providers and application providers, and then we build new business processes and transform processes using AI.
So in that light, we’re using Evolutionary AI specifically in two areas. One is for people who’ve built deep learning models, neural networks of any type. They’ve generally been built by people, architected by people, optimized by people, which means you’re constrained by the number of data scientists, PhDs you have to work on a problem.
We can take Evolutionary AI and optimize and improve the architecture of models of neural nets to make significantly more accurate outcomes. So we’ve been using that in conversational AI, in image recognition, in other forms of machine learning patterns to build significantly better, optimize models off of whatever people built manually. So we’re going to take the work of data scientists and improve it. This helps them be more productive and to deliver higher accuracy.
The other thing we’re doing with it is we’re using Evolutionary AI to find optimal outcomes. So in a business process, if you’re trying to decide how to price things or where to put things on shelves or how to staff your business and you have different goals of revenue and profit and customer retention loyalty, we can take all those parameters and all those goals and use Evolutionary AI to find optimal outcomes, which are sometimes counterintuitive outcomes, which are a way for people to take it to the next level. A lot of AI is really good at predictive analytics. We’re raising this to be prescriptive to help business decision makers know what to do to get the very best possible outcome.
Guy Nadivi: So, with the idea of prescriptive AI in mind, right, what do you think are going to be some of the biggest disruptions we’ll see in three, five, or 10 years from now with respect to automation, AI, and machine learning?
Bret Greenstein: I think it’s going to be the access of AI to business decision makers on a broad scale. While the technologies themselves will get faster, we’re going to see continued improvements in performance in cloud based on a GPU and CPU performance and virtualization capabilities. You’re going to see significantly better performance for raw execution of models. But the faster pace, and of course you’re going to see improvements in data science and the tools to accelerate the creation of models, the pipelining of data, the scaling out of AI services. But we’re going to move from an era where a small number of people, data scientists, data engineers, other experts are creating AI, to 10 or a hundred times or a thousand times more people who will be using AI, using AI to make real business decisions every day.
So, we worked with quite a few analysts and they consistently predict more than 100% growth every year of the number of projects each client is working on in AI. And that’s because it starts out as one project and then you realize I can apply this more generally and it gets broader and broader, but you’re also seeing it more accessible.
So today it’s still a lot of work to build, going back to that retail example, a store optimization engine using evolutionary AI. It’s still a fair amount of work involved in that, but these things will be packaged up and become more repeatable at which point any store manager anywhere should have access to this. Being able to feed it the data, their historic data, their supply data, their staffing and resource and cost data, and local data and get optimal models. So I think in the next few years we’re going to see AI being useful by business leaders the way that spreadsheets became useful, it’s just going to happen significantly faster.
Guy Nadivi:And so with that growing ubiquity, whenever we have an AI expert like yourself on the show, I always like to ask them the following question. Over the long term, do you think that AI and machine learning will ultimately augment more people or replace more people?
Bret Greenstein:So I think it’s going to create more jobs because it creates more value and as long as human beings run businesses and create value in market, which means grow the economy, it creates more opportunity, new jobs, new style of jobs. Now there’s definitely going to be individual jobs which will be able to be done by AI, some augmented by AI and some that can’t be replaced. And so I think for individuals looking at constant education is really important. We built a data science academy within Cognizant so that all of our employees have access to training and skills so they can continue to learn new skills and grow as AI becomes a part of how we all do our jobs. So, and I think a lot of companies are going to have to consider how they do in continuous education of their workforce as they embrace AI.
It also creates all kinds of new jobs. There was never a time long ago where you had people cleaning up data to the degree we’re going to be. Managing data, its integrity, cleaning it up, bringing new insights, creating data marketplaces, this is all new stuff. As we move from data warehouses to data lakes and modern data architectures, this is all because of AI and then the roles in AI. For example, we have conversational designers in our AI team. That was not a career anyone could choose in college. That didn’t exist a few years ago, but now we have people who design amazing customer experiences using conversational AI and they’re like the web masters of natural language and it brings all kinds of new design skills and social and psychological skills in building great user experience and doing it with conversation.
Guy Nadivi: Okay, now you just spoke a lot about the data driving AI and machine learning, but the underlying models AI solutions are based on are built upon algorithms, and late last year a Harvard Business School published an article calling for the auditing of algorithms the same way companies are required to issue audited financial statements, public companies anyways. Given that AI developers can incorporate their own biases into algorithms, even unintentionally or unconsciously, what do you think about the need for algorithm auditing?
Bret Greenstein: So I think it’s a little more subtle than what you described and that is when people do auditing of anything, they audit results. They don’t always audit the algorithms, although they can. I think in the case of AI you’re looking at systems where … I don’t think you should be looking at neural network designs to figure out is it fair or not, but I think you should be accountable for the output generated and whether it’s providing biased outputs and biased results.
There’s already technologies which some we’ve built, some of the cloud providers have built that are for bias detection and remediation where you can compare the output of an algorithm to the expected distribution and assess whether it’s created some inherent bias. There’s also more subtle forms of bias that can be incorporated that are harder to detect, but I don’t think this is something where you crawl through the code or crawl through the algorithm the same way you might think. I think this is one where you have to look more at the output and the behavior of systems rather than the code itself. The algorithm, the data that feeds it, the fine tuning the parameters, all can influence the bias of an overall system. The same way that human beings can be inherently biased. You can hire someone who seems brilliant, great resume, and they’re your recruiter for you, and they just have a bias you couldn’t see.
You’ll know it if you look at their output, but you wouldn’t know it if you tried to inspect their brain. So inspecting the brain of an AI is probably not the best approach, but looking at the output, understanding how the system is behaving, having testing criteria and design in principles around how you manage it and govern it over time is probably the best way for us to recognize and reduce bias. In addition, introducing a diversity among the people who create systems. Also, looking at the time frames of data. A lot of times people are training systems on very old data, which is itself inherently biased because people were more biased years ago. And so you have to really look at a few of these principles when you’re designing systems to make sure you’re looking at it with fresh eyes.
Guy Nadivi: Interesting perspective. There’s a lot of excitement, Bret, about AI for young people entering the workforce and even for more established professionals thinking about a career change. I’m curious, what kinds of skills does Cognizant covet the most when hiring talent for automation, AI, & machine learning?
Bret Greenstein: So there’s the core skills around the ability to create algorithms and models and to operate businesses using AI as well as the data engineering in order to make data available on the forms needed for AI. So those skills are all extremely hard. Those are the creators. But increasingly we’re looking for people who are AI aware with business point of view because ultimately we’re transforming business processes, call centers, supply chains, retail store operations, diagnostics in healthcare, insurance underwriting, reconciliation and banking. Those are not data science skills. They require data science, but they also require a subject matter and domain expertise in those processes. So I’m very excited to start it, to see universities and online education and graduate degrees beginning to be aimed at, I’ll call it AI awareness, AI understanding for business people, not just for technologists. And so as we become algorithmic thinkers, people who can recognize the value in the use of algorithms and data for your business, that skill is going to become extremely important.
And I don’t think most companies should delegate their AI transformation to data scientists or technologists. It’s really going to come down to business leaders who understand the value of this and how it works and what it means. I’ll give you a metaphor. When the internet came out, most CMOs knew they had to do something, so they took their catalog and they stuck it on the web and they called it a day. And those companies are mostly out of business. There were other people, webmasters, HTML geeks and others who recognized that the web could be the front end of business and they created companies like Amazon and others that wouldn’t have existed otherwise.
And so those people saw the web for what it was, which is a form of interaction, information sharing, and engagement and they transform business around it. So we need people who see business in terms of the data that make up the business and what algorithms could do with it. Those people will transform every business process.
Guy Nadivi: So there are concerns cropping up about the misuse of AI machine learning. And I’m curious, Bret, if you see any economic, legal, or political headwinds that could slow adoption of these advanced technologies, or is the genie out of the bottle at this point to an extent that they just can’t be stopped and perhaps not even effectively regulated?
Bret Greenstein: There are really strong lessons to learn from the past on this. And I think the idea of saying the genie’s out of the bottle, it can’t be stopped is irresponsible. And a lot of that happened in the web and it led to companies taking advantage of or abusing public trust on the web, which led to privacy and other implications and now a lot of backlash. I think all of us need to learn from how technology can get ahead of policy and good judgment, and I think we’re already seeing that in terms of AI where there’s significantly more talk around data privacy, on ethical and responsible AI, and a lot of talk at the government level. We’ve been involved with the World Economic Forum for example, as well as we recently speaking at, I spoke at a conference for Politico specifically around responsible and ethical AI and it’s implication on the workforce.
I think it’s getting discussed because I think we’re all smarter than we were and I think we recognize collectively that this is not something you just unleashed and see what happens. These are specialized skills and capabilities. They transform business. There’s implications to it and we have to go ahead in recognizing this needs to be regulated, it needs to be managed. Companies need to be responsible and how they manage it. We have a, for example, a council for responsible AI at Cognizant specifically to make sure we have cross functional leadership looking at the projects we do, the projects we don’t do, and then we’re thinking about how we use it to help and support our brand.
And if there’s anything to be learned from recent history, it’s that brands rely on how they’re perceived by the public. The use of AI can enhance your brand or hurt it, just like the use of any other technology. And so we’ve been spending a lot of time with clients discussing that and helping them to see their way through it.
Guy Nadivi: Overall, Bret, given your high-level perspective, what makes you most optimistic about AI & machine learning?
Bret Greenstein: I think the people who work on it are so much less hype than the marketing. The people who actually do the work are very grounded in what it’s good at and what it’s not. You’ll often hear media or other people talk about general intelligence and robot apocalypse and all that stuff. And that’s fun to talk about because it’s pop culture kind of stuff, but when you get down to it and talk to real data scientists, they’re not worried about that.
They’re focused on accuracy of algorithms, improving data, getting more forms of data to work with, how they build ecosystems of insights, obviously regulations. They’re thinking in very grounded terms and I think business leaders are embracing this. I haven’t met a business yet who doesn’t have some degree of investment, and they’re all trying to figure out how do they responsibly get into this? How do they do projects that deliver business outcomes, not just experiment?
And that feels very different to me than the dotcom bubbles and hype that came around from “everyone must be on the web”. I don’t think people generally believe everyone must use AI. They’re trying to figure out where it applies, where it delivers value and they’re focused on that. So we don’t get asked to do a lot of frivolous projects. We get asked to do things that deliver real outcomes, and I think projects grounded in outcomes are going to be the way that all companies embrace this responsibly.
Guy Nadivi: Bret for the CIOs, CTOs, and other IT executives listening in, what is the one big must have piece of advice you’d like them to take away from our discussion with regards to deploying AI & machine learning at their organizations?
Bret Greenstein: While data is something that’s important to be managed, almost all the customers of IT have told us they need access to data, and not just the data that runs the business, but all the data about the customers, supply chain, environments they operate in. So local data, geospatial data, social data, IOT data, all the other stuff that CIOs may not have had to manage historically are now relevant to AI-based systems for business.
And so embrace the fact that data accessibility and the forms of data is going to be a never ending agenda for you now, and we can’t just manage it and protect it. We’ve got to really unlock it, in ways that reach the business needs. And then the business leaders themselves are going to continue to ask for more, more help on data science, more help on governance, more help on access to data. And this is part of the new normal. I don’t think it can be controlled under one person. I have not seen that pattern happen very often. It seems to be pockets all over companies and we have to figure out ways to help them, govern them, steer them, enable them, without limiting them.
Guy Nadivi: I think that’s going to be a real eye opener for a lot of the CIOs and CTOs listening in.
Bret Greenstein: I think so. If I could just for a second more … So many CIOs are transitioning and pivoting to this already. So, they’re driving agendas for data modernization, creating a much more modern architecture for access to data across their enterprise. And so that’s being driven from CIOs as well as Chief Data Officers, which are sometimes in the CIO office. I think that’s great. I think it shows the connectedness between the CIO and some of the business buyers who are driving an AI agenda. That kind of teaming moves IT from being a cost to an enabler of revenue and growth. And I think that’s probably the best thing that could happen to the world of IT.
Guy Nadivi: All right. Well, it looks like that’s all the time we have for on this episode of Intelligent Automation Radio. Bret, it’s always great interviewing someone with the kind of deep domain expertise you possess, because that usually means we’re going to learn something intriguing and you definitely didn’t disappoint today. Thank you very much for coming on the show. It’s been great having you on.
Bret Greenstein: Thank you so much, Guy. It was a pleasure for myself as well.
Guy Nadivi: Bret Greenstein, Vice President and Global Head of Artificial Intelligence for Cognizant Technology Solutions. Thank you for listening everyone, and remember, don’t hesitate, automate.
