David Yakobovitch
You are listening to the HumAIn Podcast. HumAIn is your first look at the startups and industry titans that are leading and disrupting artificial intelligence, data science, future of work and developer education. I am your host, David Yakobovitch, and you are listening to HumAIn. If you like this episode, remember to subscribe and leave a review. Now onto the show.
David Yakobovitch
Listeners welcome back to the HumAIn Podcast today our guest is Matthew O’Kane¹, he’s the head of AI and analytics for Europe at Cognizant ², for many of you who don’t know Cognizant is a fantastic company that works with digital transformation in the advanced analytics and AI space we do a lot of work with them, both personally, and with our companies and organizations. So Matt, thanks so much for joining us on the show.
Matthew O’Kane
Excellent, thanks a lot, David. It’s great to be here.
David Yakobovitch
Thank you, just to get us started for our listeners, we’d love to learn a little bit more about your background and the work that you’re doing at Cognizant.
Matthew O’Kane
I lead our an AI analysis practice across Europe. It’s about 1500 people we have in our analytics team across Europe and we’re supported by another probably two or 3000 people in India in our offshore centers. So it’s a big team and everything we do is everything from helping clients around their data problems, through to business intelligence analytics, machine learning and AI.
So it’s really about the end to end, and to be honest I was quite looking at my background started off in banking. I finished a math and stats degree, got interested in statistics, joined banking and realized there was tons of data I could play around with apply predictive models to.
I wrote my first neural network probably in 2001. I might be sure my age a little bit there, but almost 20 years ago now and I never realized how important AI in Analytics will become as it is today. It’s just unbelievable. So I started off in banking and I moved into consulting probably just over 10, 15 years ago now and it’s probably around the last recession considering we’re probably going to enter another recession now I started doing a lot more work around machine learning and applying machine learning to various problems.
I can remember explain as people, it was in probably 2006 or 2007 what a random forest model was I don’t want to try and explain this thing I said, it’s like a decision tree with a thousand decision trees and the looks I had from people and I try to explain how this thing works and now everyone knows lots of people on the street you could probably ask they know about #randomforest.
So that’s my data science all the way through, obviously I joined Cognizant a year and a half ago to really drive what the next level is around analytics and AI, how clients and are really scaling AI across the companies and it’s a big engineering effort now. Hence why we’ve got a big team of people who do all the things you need to get started on around AI.
David Yakobovitch
I remember when I was also very much in the analytics space before I got involved in AI and I was working with Random Forest and I was working with Cart and all these models and this was the big thing back in the, the early 2000 teens and anywhere from like 2000 to 2012.
Like carts random forest, these were the big models that was state of the art in fact, all these decision trees almost like the one we just looked at with the MTA offline a few minutes ago about do you go left or right do you take the train or not? These were some of those basic models, but we’ve seen such an evolution in this space with analytics and AI and I’m sure it has a lot more state of the art that cognizant is working with today.
Matthew O’Kane
I still go back to the underlying machine learning algorithms have been around for a long time so when I was working in the fraud area for banking, if you’re looking at fraud it’s a problem where you’re looking at a huge amount of data, you have many variables you’re looking at and the number of frauds are really small.
So naturally using machine learning was even 20 years everyone understood you had to use machine learning to solve this problem. It’s just the fact that some of the models have got more sophisticated and computing power has come along and cloud computing power has come along too to help us actually power these more and more.
David Yakobovitch
And now what we’re looking at as you mentioned with cloud and compute, I mean, there’s so much that’s possible with the models that we previously worked with and now we’re also in a new norm a new world order if you will, around COVID which has been in everyone’s mind for at least the last couple months and I had the opportunity to listen to Harvard Business review and the Harvard Business School they have their podcasts.
They’re talking about their predictions, both of them and Wharton and a lot of these leading universities are like, is it going to last two more months? Is it going to last six, four months? We can make predictions on black swans or black elephants, but it’s not necessarily something that’s easy to think about which I’m sure we’ll get more into here but I want to hear your thoughts about, are we moving into a recession? Are we in a recession? What does that spell for analytics and AI?
Matthew O’Kane
I alluded to earlier, we are definitely in a second recession. It was at $2 trillion as it you guys said the U.S. putting in, that’s has that ever happened before? I can’t imagine it does.
David Yakobovitch
No. in 2008, they put in about it was like 600 billion or so and now we put in the first two try and another two and a half trillions about the past.
Matthew O’Kane
It’s amazing so I mean that money it does kind of grow on trees I guess in some way but it hasn’t yet impact it has a massive impact and we’re obviously going to enter a large recession at this and the unfortunate, I don’t think anyone can really imagine it’s going to be any better than that and I really to be honest when I started this year I felt we were going to enter a recession even without COVID.
Just the way the economy was going global economy, so we’re in that state and I’ve been in that time before, go back to 2008, 2009, and the type of AI and the type of work you can do within the ISP will change dramatically I think it’s unfortunate to say things like revenue generating opportunities for AI are going to be less on the priority list for at least the next year, I’d say, and it’s probably going to be more on cost reduction, etc unfortunately
David Yakobovitch
I remember last year, which seems a lifetime away now it’s 20 2019. I remember when I was listening to Yoshua Bengio and Gary Marcus and a lot of these leading thought leaders as well and the AI space saying are we coming up on the fourth AI, winter? And this was before anything COVID we have the constant cycles of hype and reality and all that moves between and everyone was saying, no, maybe there’ll be a little cooling we have a lot of exciting technology, and on the show today we will be talking about still some of that state of the art technology and best practices used at Cognizant, but what do you think AI looks like in this new recession?
Matthew O’Kane
If we say it’s moving from revenue generating opportunities to cost optimization opportunities so most organizations when they exit this. I was talking to one particular client the other week who said we’re not selling anything at the moment.
Our clients aren’t buying anything from us at the moment but we still have a lot of fixed costs that we have across business and how we can reduce those fixed costs, how we can use AI to actually automate manual processes etc.
So I think we’re really going to see a big shift towards automation around AI, and we’ve seen a lot of clients are working at the moment looking to apply AI in new areas they probably hadn’t thought of before back in operations and in a claims handling department in a customer service center so we’re going to see a lot more of that why do hope is we don’t see a lot more of the more sinister side of AI which is around observing people and breaking privacy.
We’re probably not seeing huge about that in China, we’ve probably points it’s fairly positive I think so, as long as we don’t move to that side, but the automation and the fact that automation means less jobs in a recession and it takes away human effort, we have to square up for that is going to be the reality of the moment.
David Yakobovitch
So we know that data privacy has been a big topic in the last few years, GDPR passed CCPA passed, prior to the COVID, New York was planning the New York privacy app similar to CCPA and some of the challenges are exactly what you’re sharing with us, they map is how will data be used?
I read this example in the last few weeks about some mortgage company out in California that was already creating a checklist based on people who missed rent payments both for their companies and as consumers and immediately a lot of people in the AI world, stepped up and said, I think this is I don’t know if it’s unconstitutional or what but there’s a lot of great ethics there.
Matthew O’Kane
Yes, and I don’t think privacy is going to go away. It still seems to be top of priority when I talk to a lot of clients at the moment, we’re just trying to solve privacy problems by Webex and by remote working and by email rather than face-to-face but it’s still a big issue and coming out of this if you’re going to apply more data and AI to your business, you just, the privacy aspect goes up and up rarely so it’s always going to be top of the agenda.
David Yakobovitch
And thinking more beyond just our data security and privacy, of course as we’re in this recession that now is almost definitive that we’re just waiting for the governments to actually say we are in a recession, but it seems that it is definitive there is job consolidation and job loss across the board one of the big important things for organizations to do today is to manage those fixed costs, what can we do to not only survive, but thrive during this economic time so that we can come out ahead and grow stronger from that.
What we’ve seen at least that internally at the organization I work for Galvanize is fixed cost is all about how can you have resources lend a hand you’re working together you’re in this startup mode where everyone’s contributing, seeing how can we build products or how can we put out fires together or even redeploy resources in different areas. What are you seeing on your end at Cognizant.
Matthew O’Kane
I think it was about a year ago we produced a report around the symbiosis between HumAIn and AI and I think it’s quite interesting here is that if you look at every job it’s. What are AI’s good at? What is the machine learning model very good at, and what’s a human very good at, so this is the work that a couple of my team members Mike and Sean put together.
And if you look at it there are still fairly distinct areas where humans are good and certain tasks and the machine learning are good at a task so it’s really about taking another look at every process you have and re-imagining it within this new digital AI world and unfortunately is that a lot of companies are going to be looking at these processes and looking at the cost of them and say, how do we balance this and reduce the number of humans we have work on this process and probably put them into other uses?
Frankly this is certainly a crisis that has created significant demand in some areas and a drop in demand in other areas and that’s how it’s going to play out going forward so we need to be shifting humans to the right areas and shifting out to the right areas really.
David Yakobovitch
Shifting AI to the right areas, a lot of it is about humans and AI working together. We talk a lot about it on the show with different startups and different ventures in that space and humans and machines is what it’s about I think one is the report you just shared Matt, about the symbiosis, at Cognizant you, your teams come out with a lot of reports on research I mean what else have you been uncovering or insights during this time?
Matthew O’Kane
The Humans AI Space is fascinating to me and I do see this end results for example some of the work we’re doing with an engineering firm that’s that sends engineers out not the moment but when they where they’re allowed to leave the house, they send engineers out to fix big water systems essentially in buildings now, typically if you send this engineer out to solve the problem then they’re not the expert there’s only about five experts in the entire company.
But by taking some of the knowledge from those five experts and turn them into some models like that decision tree you were talking earlier but a little bit more complicated you can kind of infuse the insight and the knowledge from the five SMEs into the day-to-day work that the engineers are doing and they can use augmented reality to actually see something they can see the water pump in front of them.
They can see some information, that decision tree in the background helping them fix the problem and I really see that that’s a really exciting vision for the future of AI machines working together so, it’s about thinking more about those ways where the human still needs to do something, the AI needs to do something who does what best really.
David Yakobovitch
I completely agree with you there, Matt, we’ve seen a lot of the reports last year from cognizant and other organizations that talked about the #futureofwork and what is jobs 3.0? Or humans and machines and we’re seeing that now where traditionally the financial analyst at Goldman Sachs is no longer just working in Excel, but they might also be using #NLP and computer vision to process loans faster and to go through different documents. So it’s not necessarily about job elimination but it can be about increasing that efficiency to release human efforts to more creative and mind challenging tasks.
Matthew O’Kane
I think we were talking about something before the show around Snorkel. Snorkel is one of the it’s coming out of Stanford University it’s a weak supervision technique which when I first read week supervision, it’s kind of you’ve got to work out what that is but what Snorkel does it allows a human to essentially to take what’s in their brain and turn it into a model.
That’s the way I describe it that’s the way it describes a few clients that we’re using with at the moment it allows your experts in the organization, your best claims handler your best salesperson your best engineers we’ve talked about earlier to take what they have and their understanding and turn them into a set of rules this is called data programming and these rules can then be turned into a neural network model using the likes of Snorkel and others, and the mandate of use cases for this, it’s just absolutely amazing because a lot of problems we have in businesses there isn’t training data ready available.
So if I want to categorize some things as fraud or not for let’s say for an insurance company I may not know exactly which cases are fraud and which aren’t for upfront, but I can turn to an expert a claims handler that has been doing this for 20, 30 years and say, can you write me a list of rules which you think are actual fraud non fraud rules and then we can turn those fraud and fraud rules into #neuralnetwork model it’s just fascinating.
It it’s one of those technologies that just works and we use it with a lot of clients at the moment we also noticed that Google has picked this up. I think they’ve got a version called Snorkel DryBell they seem to be using it throughout their throughout the company as well so, it’s an area where it’s interesting. It’s human and it’s AI. The AI is very good at processing all the massive data, but it doesn’t have the intuition that’s held inside of an expert’s hat so it’s how do we combine those two together?
David Yakobovitch
What I love about what you’ve been sharing with us today on HumAIn map is that explainable AI is basically coming of age the last couple of years everyone’s been talking about how do you uncover models? How do you discover different features?
How do you know what’s happening? And from what you’ve been sharing at Stanford, and even with Google we’re seeing today that there’s people building on top of model so a model comes out of Stanford called Snorkel, and then Google says let’s take a different case study and then maybe even Facebook or Amazon says, let’s try a different case study and then researchers at #Cognizant say let’s try another case they keep on building on each other’s efforts, one thing we’ve been seeing even in the time of COVID right now is about open research and around open science and about communities collaborating further together that’s not only being seen here, but I think it’s across all industries. How do you see open source and open research with your teams?
Matthew O’Kane
It turns around to the ethical AI space as well as the fact that if the research you’re doing and what you’re developing isn’t open and people can’t go into get help and look at it and look at your code and understand how it works.
We’re shifting away from the black boxes of this world and we’ve seen that with platforms with Hadoop and the other plethora of big data platforms we’ve seen this more of AI is the fact that if we’re sharing round different models and modeling techniques; I just love the fact that Microsoft, Google a lot of our partners at Cognizant are spending a lot of money and to have teams working creating really interesting #machinelearning research and then just open it to the community and it allows those to add on top of it.
As you said but a lot of what my team does is take the complex research and a client problem and try to fit the two together and that’s usually the hardest thing to do is actually getting something that impacts clients business really.
David Yakobovitch
One of the things I love about the companies you just shared Microsoft and Google particularly is I’m a big fan of working AI demos because when you’re looking at business practitioners you don’t always know what’s going on in the wild and Microsoft has on their website AI demos that microsoft.com place where you could actually go hands on with AI.
They have AI route planners, they have text analytics, all these pretty cool features. One of my favorite demos is the video Indexer, the video indexer I’ve used that before to actually go into a video and using computer vision see where a person is in a TED Talks or Microsoft keynote. It’s pretty incredible to see where technology has gone.
Matthew O’Kane
I like what Microsoft is doing about it’s not just about #algorithms and #code things like that we have to convince we the executives in our company if you work for a company you get how do you translate the fact that Snorkel could change their their business or some new #deeplearning could do to the actual outcomes that an exact would be interested in and I think what Microsoft is doing is fantastic just making it really simple try it out yourself see the results visually as to where to get people understand that.
David Yakobovitch
So let’s dive deeper into ethical AI and use cases but more around the government level, we’ve talked a little bit about GDPR and CCPA and you’re here in London and the United Kingdom and the Government has definitely been pretty vocal about what they’re looking to do I know that we’ve seen in The United States, particularly the department of defense came out with their call for AI ethics we’ve seen the roam call for AI ethics we know in the EU, there’s been the EU ethics both through the commission and the parliament what are you seeing from the UK Government specifically?
Matthew O’Kane
Ethical AI was find quite interesting because this is a problem that’s been around for 20 to 30 years. I’d argue so I started my career building credit scoring models and you have to be extremely careful about what information you put into those scoring models.
I think we saw something on the news, it was it was a US story around the Applecart were given a different credit limit to the wife was to the husband and I remember reading that thinking this isn’t because of AI this has always been a problem I was having this problem 20 years ago so the problem isn’t necessarily new but what’s really nice is that again like a lot of things in everyone has understood how important it is so the UK government has been doing.
They’ve used their cheering institute which does a lot of research on AI, they’ve used that to develop a set of ethical AI pieces good set of standards that cheering is pulled out and then each department now we’re working with in the UK government is infusing ethical AI into every single machine learning model or project that they run so one of the organizations we’re working with at the moment in the government every single time we build a model we’ve developed a set of criteria that has to go through before it goes live into a government use case and it’s fantastic that they’ve taken the lead on this and we’re working collaboratively to make it happen.
David Yakobovitch
It’s incredible to see that there’s so much collaboration happening everywhere in the world we think in the U.S. between what we’re seeing with big tech and what we’re seeing with different organizations like the Allen Institute for AI in Seattle and Open AI in San Francisco but it as much progress that’s happening.
In the United States we’re seeing those results and that’s evidence of the efforts occurring in the United Kingdom as you mentioned the Turing Institute sounds like it’s from one of the fathers of AI and the work is here as a collaboration between government and private practice and universities.
It does have to start with public private partnerships and then move into creating policies and standards so that we’re being responsible we’re being ethical we’re having traceable standards were having governable standards I think they’re all part of the conversation one conversation that I’ve been sharing recently with some of my podcast guests and students I work with is about design thinking and data science.
And one of the things that are as often not talked about is who in the organization does it come to about thinking about ethics and traceable standards? Does it fall on the product owner? Is that to the software engineer the data scientists the day, the engineer there isn’t much of that responsibility taken today yet, I personally think everyone should share in that responsibility have new jobs standards, or job requisitions that say thinking about ethics or responsible AI Systems. Loveto hear your take on that map.
Matthew O’Kane
In Europe, we’re quite lucky because the GDPR regulation itself doesn’t have a large part of its regulation that’s focused on AI so there are a lot of things that can be drawn from #GDPR and that has meant that over the last couple of years that organizations have been setting up that accountability network.
Who’s accountable for data privacy it’s just really about moving that from data privacy now to more around ethical AI as well so a number of companies we work with you’re the person who is ultimately responsible around breaches and privacy is also taken on the ethical AI responsibility but that’s just the accountable person that’s the person who probably goes to jail if they get something seriously wrong but the responsibility you’re absolutely right it’s from the data scientist all the way through to the product engineer if the business where we’re actually applying the AI as well because they’re making different decisions now that responsibility has gone all the way through the organization I think.
David Yakobovitch
And going deeper into the organization so I think one thing that we’re discovering today is that data can be biased. I mean we mentioned earlier to in the show about that startup in California that’s basically flagging people if they’re not making their mortgage payments.
That one’s a little bias some of them are less obvious than that we talk about classic cases in hiring that you’re not aware that the model is looking at women versus men, whether they should be hired but then that’s discovered later in the road and so that can mean anonymizing data, or normalizing #data, or different standards to improve that.
But I think one thing that you and I were chatting about offline is as we’re moving or in this recession now separate recessions are exceptionally good at making it obvious that data’s biased like a few months ago we weren’t really thinking that certain data sets were biased but now it’s like it’s so clear so what do you mean by that?
Matthew O’Kane
I mean again looking back at the last recession, I remember working with some investment banks and there was this change in view from an investment bank because they’d seen a recession so they’d seen the data look differently for a certain period of time so every time you built a model you say what would happen if it was reflected back in that recession let’s test it on 2008 data so we’re going to have a nice set of data which is very different to any of the data around it in time.
So if you are a let’s say you’re looking at call centers a lot of call centers suddenly have thousands of callers over this period and have less people to answer the phone so you’ve got a nice set of data, which looks completely different from the debt you had the year before.
So we should start to use that and make people read on sand that data is always biased if you looked at that data without realizing #COVID etc was happening you probably wouldn’t realize that’s the reason why so there’s always something behind data and there’s something generating that data, I think that’s the that’s my takeaway anywhere.
David Yakobovitch
And generating that data is always the beginning case of any project. I’ve had my design thinking standards for #datascience where I really break it down into five, six stages.
I think first ones, you collect your data then after that you’re going to start looking at refining your data, then expanding it and then start with the modeling and then the maintenance; all these five stages. And organizations can collect data, maybe they can use Snorkel, maybe they can create synthetic data but then what can they do beyond that? How can we get AI into organizations too? Be adopting AI today I mean one of the big statistics we’ve seen in the last couple of years is every organization says it but about under 20% of organizations have actually deployed AI in 2020 so what can we do to start getting more towards that deploying?
Matthew O’Kane
I’ll start with the bad news cause I was talking to someone the other day and they were looking at Google searches for AI and the search AI has kind of come off the top priority for a lot of people at this point in time quite obviously people are talking about remote working and obviously COVID etc, but that still doesn’t mean that in the next few weeks a lot of execs in companies people that are budget holders you can control where AI is used we’ll be really thinking of how they get out of this crisis and what’s what how can they take off and how can they accelerate and improve business results so, now is the time if you’re if your death size hasn’t dropped that much obviously done.
David Yakobovitch
In the last couple months so looking at as we’re talking about in this show about the Google trends I mean, from February to March AI dropped at least 25% in total listenership or viewership because well people fairly enough have a lot of other things on their mind.
Matthew O’Kane
You’ve just brought this up and we’re looking at the graph now and it has suddenly gone back up again and the last week or so maybe that’s some good days maybe some of the people that realize actually is it is important but I do see a lot of companies now they’ve worked out how to operate remotely they’ve got their people safe they’ve kept the work going and now everyone’s at home and everyone’s thinking there’s more time to think and that’s a very good time to open up about ideas about how you could be scaling AI in the organization how you can really get going and change things so I think now is definitely the time to have that conversation.
David Yakobovitch
And scaling AI is always a good time because when there is more downtime or the opportunity to digitally transform you want to think about everything that you can do. I know that cognizant recently had a post where there was a lot of talk about scaling AI beyond the pilot stage with actionable steps. This is actually one that you worked on yourself. This came out in late 2019 so we’ll definitely share that in the show notes for the audience but what were some of the actionable steps you also took out for this piece that you published?
Matthew O’Kane
There are lots of steps you can take around not just trying to solve for one proof of concept or one pilot a lot of companies are stuck in that pilot stage around AI and you’ve got to think about 10 pilots when you start thinking about 10 pilots around them one pilot you start thinking in the way you need to think to scale AI, to be honest there’s one area that I’m slightly changing my opinion on over the probably over the last year it’s how important getting the right data platform is before you can do AI.
A lot of clients have jumped over the data problem a little bit too much, and they see a lot of them going back and saying our data is just not in the right format it’s not the right platform it’s not the it’s not modeled correctly so a lot of clients that are going back and saying we need to solve our data we modernize our data create the right governance model around it usually move on to the cloud that’s what most clients doing is moving their data into cloud enabling it and then really scaling the AI. I thought you could probably do both at the same time but more and more I’m seeing you really have to have that debt foundation before you moving move manually with a lot of AI now.
David Yakobovitch
And we’re going to see a lot more efforts there this year, especially now as organizations collaborate together with open science. We’re going to start seeing that AI into organizations as you’ve been sharing with us today on the show map and as we’re continuing thinking about humans and machines working together, we’re moving in a cyclical nature where we’re going to see additional feedback where we’re able to come up with real, tangible results from these pilots what do you think is next that organizations should be thinking about?
Matthew O’Kane
I’m going to reflect back on some of the things we’ve talked about; Where are the biggest impacts from a use case point of view? It’s easy to get caught in the interesting or get caught on the shiny interesting use case that would dazzle someone but what will companies care about? They’ve really got to reduce costs, reduced errors, all these things that are dragging their business down, if we can really help in that area we can really speed up growth in the local companies so it’s maybe unfortunate I’m going to have something quite boring, do the boring stuff first and then we’ll get as we fix the companies we work for and then we can get back to that growth trajectory and get onto the more exciting things really.
David Yakobovitch
In due time we definitely will get back to that growth trajectory right now is definitely a time of working together to embrace this new norm one thing we talked about recently on #HumAIn with Eric Adams from New York City is about we’re in a dawning of a new moment in #digitaltransformation.
So we’re going to see some positive signs from COVID, that hopefully will help organizations not only in juror, but come out better from there Matt O’Kane Cognizant you’re the head of the European practice for AI and analytics. Thank you for being with us today HumAIn.
Matthew O’Kane
Thanks a lot, David.
David Yakobovitch
Thank you for listening to this episode of the HumAIn Podcast. What do you think? Did the show measure up to your thoughts on artificial intelligence, data science, future of work and developer education? Listeners, I want to hear from you so that I can offer you the most relevant trend setting and educational content on the market.
You can reach me directly by email at david@yakobovitch.com. Remember to share this episode with a friend, subscribe and leave a review on your preferred podcasting app and tune into more episodes of HumAIn.
Works Cited
Companies Cited