Artificial intelligence research: Dr. Michael Chui, partner at the McKinsey Global Institute, speaks with CXOTalk about his latest report on AI, automation and the impact of technology in the workplace.
Chui leads McKinsey’s business and economics research arm in analysis of Big Data, Web 2.0 and collaboration technology, and the Internet of Things. He’s a frequent speaker at global conferences and consults for high-tech, media, and telecom industries on strategy, innovation and product development, IT, sales and marketing, M&A and organization.
Michael Krigsman: AI is everywhere. There is a lot of hype around AI, but what’s the impact of artificial intelligence, machine learning, autonomous technologies on the economy, on jobs, on organizations? These are key questions and, today, on Episode #268 of CxOTalk, we have truly one of the world leaders, one of the world’s leading researchers who is here to talk with us about these topics.
I’m Michael Krigsman. I’m an industry analyst and the host of CxOTalk. I want to say a quick thank you to Livestream for providing our video streaming infrastructure. Livestream has been supporting us since the beginning, and those guys are great. If you go to Livestream.com/CxOTalk, they will even give you a discount on their plan, so thank you to Livestream.
Without further ado, I want to introduce Michael Chui, who is one of the leaders of McKinsey Global Institute, which is the research arm of McKinsey. Michael, I don’t know if that’s a correct or an incorrect introduction to McKinsey Global Institute, but welcome back to CxOTalk.
Michael Chui: Thank you, Michael. Thanks for having me on, and that was a perfect description of MGI.
Michael Krigsman: Tell us about MGI, McKinsey Global Institute. What does it do, and what kind of research are you engaged in?
Michael Chui: Well, as your watchers know, McKinsey & Company is a global management consulting firm. The McKinsey Global Institute is part of McKinsey. It’s an investment by our group of global partners around the world to do research, quite frankly, on topics that matter. We’ve been around for over 25 years as part of McKinsey [and], for most of that time, have done work on productivity, country competitiveness, labor markets, [and] capital markets.
For the past few years, we’ve added another research leg, which is around the impact of long-term technology trends. We’ve looked at data and analytics. We’ve looked at open data. We’ve looked at Internet of Things. Increasingly, now we’re looking at artificial intelligence, robotics, and automation technologies and their potential impact on business, society, and jobs and employment more generally.
Michael Krigsman: When you focus on artificial intelligence, robotics, these kinds of technologies, do you have a particular perspective or a lens through which you look at these topics?
Michael Chui: We have multiple lenses. First, we understand we need to understand the technology itself, and so it’s really interesting. I mean the term “artificial intelligence” is not new. It was coined almost, I think, over a half a century ago, so it’s been around for a while, and so understanding the underlying the technology.
Then, really, it’s not just the technology that we’re interested in. We want to know what this means for business, what this means for jobs, what this means for employment, what this means for economies, and so we really try to take that lens on it. What’s the impact of those technologies that underpin some of these trends?
Michael Krigsman: It’s the business impact of these technologies. Now, is there a common definition that you use? Just as a baseline, maybe you can help define this for us.
Michael Chui: Defining artificial intelligence, you could go for hours debating it. Roughly speaking, we would describe it as using machines to do cognitive work, to do the work that comes about primarily because of our brains. But, as it turns out, even from my graduate research studies, we know that not all of our intelligence is just trapped in our brains. It’s also part of our bodies, et cetera. And so, we understand that, in many cases, artificial intelligence itself might enter the physical world and be things like robotics and autonomous vehicles, et cetera. But, it roughly has to do with intelligence and then the machines that instantiate it.
Michael Krigsman: I don’t want to spend too much time on the definitions. However, when you talk about machines doing cognitive work, you’re sort of leaving wide open a big question that I just have to ask you to elaborate on.
Michael Chui: What’s the question that we’re leaving wide open?
Michael Krigsman: What does that actually mean to do–
Michael Chui: Yeah.
Michael Krigsman: –cognitive, because machines are not thinking, right? They’re going through many, many patterns–pattern, pattern, pattern, pattern–and matching. It’s not really thinking, but yet we use that term “cognitive.”
Michael Chui: I think it’s completely fair. Like I said, we can continue; we could talk for hours and hours about whether or not this is that or that is this. Roughly speaking, if you think just in economic terms, and you think about the work that we all do, there’s work that we think of as involving intelligence. Now, there are all kinds of different types of intelligence, too. In some cases, we think about logical problem-solving. As you said, pattern matching, recognizing patterns, but also developing new patterns.
As we think about what it takes to be an effective worker in a lot of places, whether you’re a salesperson, whether you’re a customer support representative, whether you’re a manager or an executive, part of that is understanding human emotion as well; being able to process that and then respond appropriately. We all think that those are aspects of intelligence in addition to the ability to potentially pick up something, which has irregular shapes, irregular objects, and then maybe add a certain amount of squishiness. You don’t think about that as being something that you use your brain to do, but what we found is, in robotics, those are some of the hardest things to do, and so that often gets incorporated into this definition of artificial intelligence. Some people separate out the physical from the pure cognitive; but, in many cases, those things can’t be separated.
Michael Krigsman: Finally, on this definitional point, is there a meaningful impact in the fact that you define AI in this particular way? How does that address, shape, inform, or circumscribe your research?
Michael Chui: Yeah, well, we take a pretty broad definition of what we mean by AI. Quite frankly, sometimes we use other words such as automation to capture the fact that some of these things don’t seem to involve a lot of what you would call a higher level of intelligence. I think, as we think about it, the sharp cuts between whether it’s AI or not are less important than what the impact might be.
If you don’t mind, I think there are a bunch of other words. People talk about cognitive computing, which is perhaps just another brand name for AI. There are some additional distinctions, so particularly recently. When I started in AI, a lot of the AI that was developed was purely algorithmic in a certain sense, being able to program an if/then type, and large amounts of if/then type statements.
Machine learning is quite occurrent now in that roughly the difference being between traditional computing and machine learning is now you’re training a system rather than programming it. The important thing is, what are the training sets? What are the ways in which you maybe reward or punish is another way to think about machine learning? Then deep learning, particularly, is another where a lot of the recent advances have been.
To a certain extent, that’s oftentimes synonymous. It is really a subset. Deep learning and machine learning are subsets of the overall artificial intelligence. That’s where a lot of, as I said, the recent revolutionary or at least striking developments have been and where a lot of energy is going into on the technical side.
Michael Krigsman: Of course, understanding the technology is kind of a basis or a foundation, but your focus is on the business implications of this, looking at industries and looking at companies. Can you share with us some of the broad, high-level conclusions of what your research has uncovered?
Michael Chui: Yeah, well, a few things. One is, our research is ongoing. This is a topic I don’t think that we’ll publish one report and we’ll be done. Just to preface, we can keep diving into different details.
First of all, as we started to try to understand what the overall potential for these technologies might be that we roughly call artificial intelligence, they are huge. They affect potentially every sector, potentially every function. One reason for that is a lot of the potential applications of AI really are extensions of the work that people had already started in data and analytics. And so, we’ve been looking at nearly 500 different use cases of artificial intelligence across every sector, across every function.
Sometimes what we say is, these traditional analytic methods, whether it’s regression or what have you, gets you this much impact. But, when you could add the multidimensionality of additional data or these additional deep learning techniques, you could increase, for instance, forecast accuracy or increased OEE or decreased waste, a number of these things, which these use cases allow us to do. You could think of AI as just being another turbocharged tool for your analytical toolkit. I think that’s one broad finding, which is that there is almost no part of the business that this couldn’t affect.
Another piece, though, is we’ve been surveying thousands of different executives in companies all around the world. My colleagues who serve clients on these topics also have very direct contact with people who are thinking about or are using AI. One of the things that we know now, as we sit in December 2017, as we’re talking, is that it’s very early. While there’s this huge potential for improving economics, both in the top line and bottom line, that a very small percentage of companies have actually either deployed AI at scale or within core business processes.
Now, that’s changing every day as more and more companies develop this capability, learn more about the technology, and then also be able to embed it within the processes of an organization, which in some cases is the hardest thing to do. As we sit now, we’re just very early on this learning curve. It’s a steep learning curve, but we’re early. And so, I think those are two things, which a little bit intention, so much potential, but we’re early.
Michael Krigsman: I want to remind everybody that we’re speaking with Michael Chui, who is one of the leaders of McKinsey Global Institute. Right now there is a tweet chat going on on Twitter using the hashtag #CxOTalk. You can ask Michael questions. Please share your comments and your thoughts.
Michael, we’re at this phase where it’s clear that there’s all of this potential across many, many different, almost every sector of industry, as you described. Yet, very few companies have deployed artificial intelligence technologies at scale. That means that at this stage there is also a large understanding gap and execution gap. And so, how do organizations fill those gaps? What steps do they need to take?
Michael Chui: Well, I think, in terms of steps that need to be taken, it’s not the first time when we’ve seen a new family of technologies enter the scene, which has great potential and where it’s fairly early. If you think back to cloud, if you think back to mobile, these are technologies which are transformative but take time to actually embed into a business.
First of all, as we were talking about, you need to have an understanding of the technology itself. You need to raise the waterline. That can often start with the technologists. We’ve said this so many times when we’ve talked about any technology. It’s not just the technologists who have to understand something about it because you really do need business leaders writ large to start to understand what’s possible. If you don’t understand the art of the possible, you’re not going to be able to identify and prioritize opportunities. I think that understanding the technology is number one.
Secondly, again, because we’re at this early stage where the potential is increasingly being recognized, we’re starting to see more and more vendors show up at our doors carrying AI solutions in their bag. That’s great. I think that’s part of taking those meetings as part of that education. But, I think often a failing would be to be captured; to find something so exciting that you just want to go ahead and do it. I think making sure that, again, not that you may need to take years to do this, but very quickly understand the portfolio of possibilities so that you’re actually spending your time on the types of solutions which will drive the needle for your business.
For instance, I talked about the 500 or so use cases that we looked at. Just two broad categories where we’ve seen huge amounts of potential: one is basically on the customer-facing, sales, marketing, [and] customer experience side. This is everything from improving your next product to buy recommendation to being able to segment a customer who comes to you so that you provide them with much more personalized and customized interactions. Then another broad category, which let’s just describe it as operations improvement, and so whether or not it’s identifying waste, reducing inventory costs, managing your energy costs, logistics, supply chain, et cetera. It’s another broad, broad category.
Again, if you understand what type of business you’re in, you’ll understand which of these needles, if you tuned it up by 10%, 15%, 50%, which one will actually drive the most value for you. I think understanding that and then, of course, the execution against it is so, so important because, again, that last mile, right? We see this pilotization, you know, pilotize-itis, maybe, right? It’s great to run a pilot. It seems to be successful, but the real hard work then is how do you, day-to-day, change that process which is really going to drive benefit at scale? That’s the hard work we do on change management every day, but this is a different set of underlying technologies. We’ll need to learn how to do that in every business.
Michael Krigsman: What you say is so packed. There are about 12 different things I’d like to ask you simultaneously, and I wish we had about 3 hours. We have to focus because we have so limited time. We have half an hour left, and the time is just flying by.
What are the common threads? As you look at industries, what are the common threads? At the same time, we have a question from Twitter, and maybe you can weave this in. Scott Weitzman from IPsoft is asking, “How does the company understand technologies from a purchasing perspective?” In other words, okay, how do companies think about buying this? The industries and absorption are the topics here I would ask you to elaborate on a bit.
Michael Chui: Well, let’s start with industries, as you said. Again, if you take these two broad categories of value potential, there are a lot of industries which much of their value gets driven from their customer interactions. If you’re a retail company, if you’re a consumer package company, et cetera, it might make more sense to look at the value of AI and those types of functions. On the other hand, if you’re driven by your operational effectiveness, if you’re in the business of manufacturing, delivering and shipping products, for instance, if you’re in logistics, then perhaps those operational needs. Again, I think those are, at least at the top level, one way to think about it.
I think another common thread that we have found is the following, which is, I think oftentimes you discover a technology which has a potentially transformative impact. You say, “Gosh, isn’t there a shortcut? Can’t I just jump and use that to compete?”
As it turns out, what we’ve discovered with AI, particularly because of the need for large training sets of data that, in fact, we’ve discovered a high correlation between both sectors as well as individual companies, which are further along on their digitization journey, so the ability to use digital within their core processes to improve process effectiveness. There’s a high correlation between that and your readiness for AI, and so I think one of the other common threads we’ve discovered is it’s actually quite difficult to accelerate past your digitization journey. You need to be on the digital journey in order to enable yourself to be ready for AI. I think that’s another finding. If you want to accelerate your potential impact with AI, you need to accelerate your move along the digital journey.
To the questioner’s question in terms of how do you think or how are customers thinking about purchasing, I think a couple of things. One is, because it’s early, developing and understanding from the customer perspective is more important than ever. Again, I’m assuming the question is coming from a place of, as a technology vendor, how do you sell more effectively? The first thing is sometimes what you need to do is to educate the customer so, in fact, they can be a good customer [and] understand the value that you’re potentially providing.
The other piece of it is, as we think about the impact of AI, certainly, there are impacts of AI which are improving the IT function itself, so whether or not it’s using AI to improve cybersecurity, using AI to improve customer support for IT. Look; the vast majority of the value of AI is coming from improving business metrics. Then what I’d say is then it’s going to be a simultaneous sale.
From a procurement standpoint, you need to make that case to the business. You also need to make that case to the technology function as well, simultaneously. I think that’s something else that we’ll see more and more of, which is that the business will need to understand this in order to buy AI.
Michael Krigsman: When you say that the vast majority of the business benefit of AI comes from metrics, I’m assuming that what you mean is businesses need to see the practical results or are you getting at something else?
Michael Chui: Yeah, I think business needs to see the practical result. When I mean practical result, we’re talking about things like increased conversion. We’re talking about things like decreased maintenance costs, increased uptime, higher priority prospects or higher potential prospects being identified. Literally, the things that drive the business, the metrics that drive the business are the ones that people are going to be looking toward AI solutions providing.
Michael Krigsman: We have a question from Twitter from Mitch Lieberman about a topic that I know is very important to you, and that is the issue of AI and the economic impact and the impact on jobs and the workforce.
Michael Chui: This is a topic that we recently published. Actually, we published two reports. We published one in January, and we just published one at the end of last month, our last one entitled Jobs Lost, Jobs Gained. A couple of things: one is AI and these automation technologies more broadly. Again, some of the potential impact is for these technologies to automate activities, which we currently pay people to do in the economy.
Again, we looked at things at the level of individual activities, not just occupation, so 2,000 different activities we pay people to do in the global economy. Roughly half of the time that people spend being paid at work is on activities which theoretically could be automated by adapting technologies, which exist today. That sounds scary, right? That’s a large percentage, but we’re not predicting 50% unemployment tomorrow partly because it takes real time. It takes real time on the technology side to develop the technology.
You actually have to have a positive business case, typically, when technologies are first developed, whether it’s a self-driving car or an artificial intelligence algorithm. It tends to be relatively expensive. That cost declines thanks to Moore’s law. You need to net that out against the cost of human labor, and that’s different around the world.
In any case, 50% of the world’s activities potentially might not be automated for another 40 years, so 2055. Although, we have a scenario which is 20 years earlier and a scenario that’s 20 years later. We do know that increasingly activities, which we pay people to do, will be automated.
The question then is, will there be enough demand for human labor, even net of the things that might be automated? Our report from last month suggests yes. If you look at a number of different potential catalysts, whether it’s increasing prosperity around the world, another billion people entering the consuming class in the next couple of decades, whether you’re talking about aging, which is a troubling thing because we have [fewer] workers, but on the other hand it drives the need for healthcare. We have roles for people to develop and deploy the technologies themselves.
We’re going to see increasing, hopefully, investment in infrastructure to help the consuming class, but also fix and improve the infrastructure we have. We’ll see changes in energy mix and efficiency, and potentially even a lot of what’s currently unpaid work in the economy that’s many times done by women at home, whether it’s childcare, cooking and cleaning, increasingly enter the market. If you look at all of those things together, and then even you netting against those the activities which AI and robotics might do, we still see plenty of work for people to do, enough to basically offset the effects of automation.
The broad question, though, is, okay, if you think mass unemployment isn’t going to be the problem, mass redeployment might actually be the problem. As much as we want the education system to get better, it actually works fairly well. What we think is potentially the great grand challenge for the next couple decades is, how do we retrain millions of workers who will be displaced by the technology? We need them to keep working to have economic growth, and yet, at scale, retraining of people past their first two decades of life is something that I dare say we haven’t completely solved yet. That’s something we really badly need to work on.
Michael Krigsman: Clearly, there is a very broad set of stakeholders that need to be involved in thinking through these issues.
Michael Chui: That’s absolutely right. The breadth of this includes everyone from government and policymakers, business leaders, and then individuals themselves. People have said, “Look. What’s the one thing we need to do?” Given the breadth of this problem, I do think it’s potentially an all of the above. We need to think about, what are the public policy levers; what are the business levers?
I’m encouraged greatly by a number of business leaders. You think about these heartless capitalists, but they understand that Henry Ford thing about, if you don’t pay you’re your workers enough, they can’t buy your products, right? We do hear about business leaders saying, “How do we retrain our workforce? In fact, how do we even retrain our workforce if sometimes their next job won’t be with us, but we know that we need to have a compelling employee value proposition?”
Then a lot of people have said, “How do we enable workers to take charge of their own careers, take charge of their own training?” Give them the tools and the information they need so they can retrain because this will affect all of us. It doesn’t just affect a sliver of people who are low-wage front-line. We find activities even for people with graduate degrees; 20%, 30% of what they might do might potentially be done by machines, and so they’ll have to think about what they do differently as well.
Michael Krigsman: To what extent do business leaders need to think about these retraining issues now at this very early stage where they’re just beginning AI pilots in very narrow processes?
Michael Chui: I do think that this is something that demands some immediate attention. Now, it’s not necessarily because things are going to happen overnight, particularly with regard to AI. But, if we think about automation technologies more broadly, then, in fact, we are starting to see these things, whether it’s robotic process automation, whether it’s physical automation from a manufacturing plant, in logistics, or in a distribution center. These technologies are coming into play today.
Again, another point that we sometimes make is, while we’ve described this as being a multi-decades trend, that this will take time in macro, it will happen quickly for individuals. It will happen quickly for individual workers. By the way, it takes time also to understand retraining. We described this as a grand challenge. Usually, grand challenges aren’t solved overnight, and so I do think business leaders engaging on this question about retraining their workforce on a continuous basis at scale is something that is a question that ought to be top of their mind when they’re starting to think about their workforce strategy.
Michael Krigsman: What is the difference between retraining now and retraining following the wave of outsourcing that took place some time ago?
Michael Chui: Yeah, when we look to history, in some ways there are encouraging pieces there, whether it’s outsourcing, whether it’s the broad-based move from agriculture to manufacturing, the increasing move for manufacturing services. In flexible marketplaces we’ve found people do find new jobs. Now, there is a separate question about whether or not they’re being paid as much as we would like them to be in order for them to continue to consume and be part of the economy. Broadly speaking, we’ve been able to do that.
Are there differences now? A couple of things: One is the breadth of potential impact here. Like I said before, this is not just some assembly line manufacturing worker. This is not just one role or another. In fact, all of us will have this impact our lives, and so I think being able to do that. The scale, also, particularly, again, past the first two decades of life, mid-career retraining we’ve found to be a challenge as we’ve looked at different programs. There are successes out there. But again, we need to build on those successes to be able to see what we can do at scale.
Just the level of investment. I had the privilege of talking with a workforce development board recently in a large city in the United States. We just sat back and compared the amount of investment being made in K-12 or K through college, even, and compare that to the level of investment that we’re making in people who have left formal schooling. Again, it’s a huge, vast difference.
Now, that said, this has been a story that’s continued. Again, when we started this process of moving from farm to factory in the United States, there was no universal high school. That was a social movement. That was a set of investments that we, as a society, decided to make. I think our question now, as we think about artificial intelligence, as we think about the impact that might have on the workforce, the fact that we’ll need to retrain so many people, what are the analogous, but different sets of investments that need to be made so that we can take advantage of the benefits of artificial intelligence, and yet continue to have both dignity, as well as income, that comes with work? We don’t believe, in fact, if the machines do all the work and people aren’t working at all, we won’t have enough economic growth in addition to the other benefits that work gives you.
Michael Krigsman: I want to ask you about demographic issues. Before that, we have a comment/question from Twitter, from John Nosta who has been a guest on this show and is a brilliant thinker about the future of healthcare and the impact of these technologies. John asks, “What about the potential for a guaranteed minimum income that changes an entire notion of what is a typical full-time job?”
Michael Chui: Yeah. This idea of a universal basic income, guaranteed minimum income, et cetera, is capturing a lot of currency. I sit in San Francisco here and, as it turns out, there are a lot of people talking about it there. There are lots of arguments for it.
One of those arguments is if we think the machines are going to take everybody’s job and we’re going to have mass unemployment, we need to make sure that everybody has enough income so that they, in fact, can feed themselves, et cetera, and feed their families. I think that justification or that rationale for universal basic income gives up too early because that assumes mass unemployment. In fact, what we say is we do need mass redeployment, not mass unemployment, just to make sure that we have enough economic growth going forward.
Our point of view is that we’ve looked at the past 50 years of economic growth. Half of that has come about because of more people working. Because of aging, we’re going to lose a lot of that. One way to think about it is we just don’t have enough workers. We need all the AIs, robots, et cetera working, plus we need people working to have economic growth. Again, if you think UBI is based on the fact that we’re going to have mass unemployment, I think you’ve given up already and, in fact, you need to move.
The other thing that I think is also helpful, again, as we modeled out the potential impact of AI and other technologies, plus these additional drivers, we might continue to see this increasing income dispersion or income inequality. You might ask, “Look. We just need to make sure that people get paid enough.” Well, then again, if you want to look at it from a public policy standpoint, maybe you could target the types of subsidies such as the Earned Income Tax Credit, which both incent work as well as provide additional income to people. I think, thinking through all of those possibilities.
Now, that said, UBI for a place that’s a developing country, again, it might put the floor in place that allows people to have a lot more freedom in terms of what they’re able to do in their job. But, in a developed country, both because of the expense, as well as the fact that it isn’t targeted towards trying to get people working, I think it’s challenging for that reason. That said, the overall point, another overall point that we found from history, which we hope will continue is, while we don’t think everyone can completely stop working, the working week has declined, on average, by double-digit percentages over a matter of decades and centuries.
Hopefully, we all can actually have more time for leisure. By the way, leisure actually drives new activities, new occupations. That’s something else we need to do. We need to continue to generate new activities and occupations. Hopefully, the workweek will continue to decrease over time. At least, for the foreseeable future, we don’t see it going to zero.
Michael Krigsman: We have questions backing up on Twitter. Steven Norton from the Wall Street Journal has a question that I want to get to in a moment. The CxOTalk social media account [laughter] has a question that I want to get to. What about the issue of demographics? Where does that come into play?
Michael Chui: Yeah, demographics is a really interesting and a number of powerful factors. Again, we cover some of this in the report we published last month. First of all, countries vary greatly in their demographics. For many countries, they’re aging, and that basically exacerbates this question; we don’t have enough workers to continue the economic growth that we’ve enjoyed for so many years. The reason why we have better lives than our parents and our parents had better lives than our grandparents, et cetera, is because of this economic growth and half of it coming from more people working.
Germany’s workforce is declining. Japan’s workforce is declining. China, with a population of a billion and a half people, their workforce either is or, depending on who you ask, will shortly begin declining. Those are countries which simply don’t have enough workers to underpin economic growth. Again, one of the implications of that is AI and robotics can actually be some of the workers, can fill in for that gap in terms of just numbers of people who are available to work.
That said, there are other countries like India, countries on the African continent, et cetera, which are very young, and their demographic pyramid looks very different. We’re concerned at some point about the fact, well, gosh, what if automation AI, these technologies, come into play just as they need to create even more jobs? That’s absolutely true in India, for instance; another 150 million people needing jobs going forward.
Again, we modeled out all of these potential drivers of additional demand. By the way, we picked seven of them. We know that there are more, so even our modeling is limited. Particularly in those countries which tend to be young, those are countries also which tend to have high aspirations for their economic growth. They start relatively low on the GDP per capita scale. As a result, that will generate lots of demand for human labor, as well as robotics and artificial intelligence. Even in those countries, we see the potential for lots of work as well, work to be done.
Again, that comes back to the question of retraining and education. Can we get people into those jobs? Then, can you actually deploy those technologies in a way because, as I said before, AI and robotics require an underpinning of moving on the digital journey? Even those countries, which are developing and young, will need to move on the digital journey in order for them to take advantage of these other technologies and improve their productivity while they’re generating new jobs for people as well.
Michael Krigsman: We have several questions from Twitter, and I’ll ask you to answer these relatively quickly because I’d love to get to as many as we can. Wow, my mind is just being blown by the things that you’re saying, Michael. [Laughter]
I want to remind everybody; we’re talking with the amazing Michael Chui, who is one of the leaders of McKinsey Global Institute. His research into AI, the impact of AI on jobs, the economy, is really pretty extraordinary.
We have a question from Steven Norton on Twitter. Steven Norton is a reporter for the Wall Street Journal. By the way, I want to always encourage reporters to ask questions. We have such amazing guests on CxOTalk. There are a lot of stories here to write. Steven Norton asks, “What role does user experience play in [the] adoption of AI in the enterprise? Is user experience today sophisticated enough to be useful to help employees outside the data science and IT departments?”
Michael Chui: Hey, Steven. Thanks for the question. I love your stuff, by the way. A couple of things: One is, I think we are starting to see really interesting things in AI, particularly in voice, et cetera, where it’s a lot better than it used to be. At the same time, is it perfect? Not even close. There’s that very visible frontline use of AI for user experience.
There’s also the effective use of AI and these technologies a bit in the background, so making sure that, in fact, you’re getting a personalized type of response. I think, in some cases, that’s getting a lot better. In fact, you can’t even see that it’s working. I think there’s a little bit of a distinction there, but in both cases, I think AI continues to improve the potential of improving customer experience. Some of it is sometimes laughable, but we do see grading improvements in that over time.
Michael Krigsman: Another question from Twitter. This has become, like, 20 questions here. Mitch Lieberman asks, “You’ve mentioned hundreds of use cases. Where does medical fit in? And, in general, which use cases have the greatest impact?”
Michael Chui: Yeah. The healthcare use cases are extremely powerful. Lots of value potential there. In fact, I’m sure the questioner has seen lots of interesting academic results being published day-by-day. We see a recent result out of Stanford where the ability for a deep learning system that was trained on radiology scans to be able to diagnose pneumonia better than trained radiologists, for instance. Lots of interesting work going on in dermatology.
Primarily, these are places where deep learning’s effectiveness and image processing allows it. That’s a natural place to effectively go, but we are seeing applications across the healthcare space, whether it’s in claims and provider, as well as looking at population health. Number one, healthcare is a huge part of the economy, particularly in the United States. There are lots of potential for improvement. Over time, hopefully, there’ll be more and more data that can be used for training. I think the potential there is just absolutely tremendous.
Sorry, Michael. I’ve forgotten the second part of the question.
Michael Krigsman: The second part of the question is the ability of user interfaces, user experience to enable non-specialists in the enterprise to make use of the data and these AI technologies.
Michael Chui: Yeah, I think this is one of those things where, just like many other technologies, oftentimes the benefits to the end user are embedded in the system, but you don’t have to understand a convolutional neural network in order to be able to speak to a voice interface. I think, over time, what we’re going to find is a little bit of the AI inside.
We see a lot of the technology CEOs saying, “We’re going to be an AI company.” That doesn’t necessarily mean that someone is going to have to be on the command line and set up the system themselves. Rather, that it’ll be embedded in the systems that operate, and we’ll just benefit from better systems.
Michael Krigsman: Okay. In the spirit of plowing forward fast, the @CxOTalk Twitter account asks an interesting question. She asks the question, “What differences have you seen since you published your report at the end of 2016 and early 2017?” What’s happened this year, and where is it going? Where do you see it going, the changes and the rate of change as well?
Michael Chui: In some ways, what’s happening is predictable because we’ve seen these technologies, whether it was from data and analytics, whether it’s Internet of Things, [or] cloud. We’ve talked about mobile before as well. We’re starting to see it permeate in terms of awareness, particularly amongst business leaders, at least people asking, “What’s AI?”
Then there are all kinds of funny, naïve things as well, right? I joked at another conference. There was a time in the data and analytics journey where you’d hear a business leader ask their IT person, “Please buy me a Hadoop,” and now, “Please buy me a neural network, and how much does it cost?” Right?
We’re starting to see that. That’s part of being along the hype cycle, being part of a journey. I think we have good or increasing awareness now. But now, I think, understanding the technology, being able to actually deploy it in business, we’re on the cusp of doing more and more of that. Clearly, the born digital companies, online companies, et cetera, have embedded these types of technologies deep within many of their processes.
We’re starting to see incumbent. Let’s describe them as incumbent sectors because we have startups in every sector. But, incumbent sectors are starting to think about, how can I use these technologies to improve my performance to compete better? That’s really starting to happen. It really is a process of understanding, but then practice, right? This isn’t something you just read about on the Web. It’s something you actually have to do because, at the end of the day, that last mile challenge, “How do I embed that at scale within the process of an organization?” that’s the hard stuff.
Michael Krigsman: Finally, we have about three or four minutes left. Picking up on what you were just describing as incumbents starting to pick these things up, what advice do you have for incumbents who are looking at this and they say, “Hey, we need to do something.” What’s your advice to these folks?
Michael Chui: Yeah, number one is, dedicate some time and resources to understanding the technology and its potential. I mean I should have said they should read our report, but [laughter] but I’m not going to do the commercial. I think it really is starting to understand what that potential is. Then I think the same sort of test and learn philosophy, which was effective in data and analytics broadly, I think that’s something which is true here, too.
Another thing I think, which is also true, is particularly for the technologies which are working well today around machine learning and deep learning. They’re based on having training sets, so data. Again, I think being sophisticated about having a data strategy is really important.
I had the opportunity to speak with Andrew Ng, for instance, who is one of the pioneers in deep learning and machine learning, overall. He talks about some of the leading companies in the deployment of AI, really spending time on these multiyear views of what data is important to be collected or have access to so they’ll be able to compete going forward, and they’re playing these multiyear. He describes them as multidimensional chess games to have access to the data which matters.
Michael Krigsman: What are the largest challenges that you see these businesses are facing right now?
Michael Chui: Well, one of the largest challenges now is particularly on the human talent side. We saw this with data scientists previously. Again, to a certain extent, we talked about many of the AI use cases being extensions of the analytics use cases. The analytics challenges with regard to talent now are extended to the challenges around AI as well, and so huge amounts of war for talent in terms of people who understand these technologies deeply.
Of course, that’s changing, too, as more and more people take advantage of online resources, enroll in classes, et cetera. Again, supply and demand are constantly evolving. Right now, demand is so high, and supply is relatively limited. One of the biggest challenges is just having people onboard who can do it.
Then another challenge is around data because, again, many times you need these large training sets. In many cases, labeled training sets, and that’s a really interesting job effect too. It’s been covered a lot, but take for example self-driving cars. They have these cameras in them, and they do a lot of their sensors through cameras. There’s a growing number of people whose job it is to annotate video feeds from self-driving car pilots to help train these cars themselves. Again, there’s a real challenge around data, and that’s something else that companies, which are on the forefront, are really spending some time and energy addressing.
Michael Krigsman: Mitch Lieberman says that he’s doing the commercial for you or for Michael Chui and McKinsey saying, “Go read the report because it’s really great.” For sure, I will send that.
Finally, in the last one minute, changing gears entirely, what do you think about Bitcoin? [Laughter]
Michael Chui: [Laughter] Bitcoin is fun to watch. I have an app that shows me the Bitcoin’s dollar conversion every day. Let me leave this with people just for fun, for those who remember their computing history. We talk about Bitcoin as an application of blockchain. I wonder if blockchain is a little bit like risk computing.
Michael Krigsman: Ah-ha. In what respect? Yeah.
Michael Chui: It seems to have transformative impact in terms of its potential performance, but perhaps the legacy systems, either technical or business, could incorporate them, and so it doesn’t have as much [of a] disruptive impact at the business level so much as incremental.
Michael Krigsman: Well, that’s very interesting. I wonder how the blockchain folks would think about that. That’s a whole other discussion. Blockchain folks, what do you think about that?
We are out of time. Wow! Michael Chui from McKinsey Global Institute, thank you so much. This has been a super, super speedy fast 45 minutes. Thank you so much, again, for taking time in coming here and sharing your experience and research with us.
Michael Chui: Thank you, Michael. I have enjoyed being with you again.
Michael Krigsman: I hope you come back.
Michael Chui: I’d be delighted.
Michael Krigsman: Everybody, you have been watching Episode #268 of CxOTalk. Our guest has been Michael Chui, who is one of the leaders of McKinsey Global Institute. Truly, their research is extraordinary, and so I urge you to check it out.
Next week, we have a show on Wednesday, Episode 269 with two interesting folks: Minette Norman, who runs Engineering Services for Autodesk, she’s responsible for 3,500 engineers; being joined by Tamara McCleary, who is one of the most well-known digital transformation experts. Join us on Wednesday. Next Friday, there’s no show because of the Christmas holiday.
Thanks, everybody. Thanks, everybody, for joining us, and we will see you again next time. Bye-bye.