AI, and in particular Machine Learning, is having a huge impact on the business world. MIT scientist, Dr. Andrew McAfee, discusses the future of Artificial intelligence and the ways in which we underestimate its power and influence. If you are a digital leader, wanting to understand how the progress of this technology is changing business, the economy, and society, this is the podcast for you. Find Dr. McAfee's full talk from ParadigmShift at www.thoughtworks.com/paradigmshift
The dominant strain of Artificial Intelligence is Machine Learning and I think it will continue to be for a long time. I can’t think of an industry that is not being affected by right now, and that will not be affected a lot more in the not-too-distant future by this new category of technology.
What we’re learning is how incredibly powerful Machine Learning systems are at different kinds of classification. What I’m really excited about is that Machine Learning systems are breaking out of the sandbox of playing games and they’re doing really powerful things in the real world -- diagnosing disease, translating among languages, detecting fraud.
To use Machine Learning systems strategically, companies must have a deep-wired cultural preference for making decisions analytically, as opposed to based on gut, judgement, intuition, seniority, charisma, politics, or table-pounding skills. I’m really interested in the hard work organizations are doing to minimize these things and make decisions on the best available evidence.
Adapting Machine Learning for decision-making can cause wrenching changes for a lot of organizations. If the best available evidence comes from a Machine Learning system, as opposed to a person that’s been doing that job for 20 years, if that system is better the companies should go with it.
The world of sports can confidently prove that older, more established organizations that aren’t born digital can still shift to making decisions analytically. In the past 20-25 years, in almost all the major sports, the smart, well-managed teams are bringing in data-minded people to do a ton of analysis to inform decisions about players, strategies, tactics and anything to do with the sport.
If you ask a clear question and the analytics team comes back and says you should do the following four things differently, a lot of organizations do zero of those four things differently. All these processes of the status quo and inertia and politics come into play and companies come up with 24 reasons not to do those four things differently. You just see this happen over and over.
If companies invest in Machine Learning capabilities to change the way they do human capital processes, performance goes up. It lets talent shine through. It lets diamonds be detected more easily.
There’s nothing about Artificial Intelligence or Machine Learning that overturns the fundamental rule of computing which is garbage in, garbage out. If you train these systems on biased or flawed data, they’re going to give you biased and flawed decisions.
If you’ve been a historically underrepresented member of a community, technology will give you chances to manifest what you can do. You can learn more quickly, you can make your skills and abilities more visible, and if companies change their decision-making processes you will have a better shot at realizing your full capabilities.
Tania: You've probably heard in many ways or at least questioned what it would mean to augment humanity with machines, but did you consider that machines don't just think faster than humans, but they use completely different reasoning and logic to solve problems in ways that we can't even imagine. And this means we're less able to predict and control for those outcomes. This is a theory that was shared by Dr. Andrew McAfee at our annual executive conference ParadigmShift. Dr. McAfee or Andy as he prefers it, is an MIT scientist who studies how technological progress changes business, the economy and society.
Welcome to Pragmatism in Practice, a podcast from Thoughtworks where we share stories of practical approaches to becoming a modern digital business. I'm Tania Salarvand, global head of marketing for Thoughtworks, a global software consultancy. I recently had a chance to sit down with Dr. McAfee. Here we talk about the future of artificial intelligence and why he says we're still underestimating its power and influence.
Andrew McAfee: My name is Andy McAfee. I am a scientist at MIT and along with my coauthor Erik Brynjolfsson, I co-founded the Initiative on the Digital Economy, which is part of the business school at MIT.
Tania: Well, thank you for joining us. We talk a lot about AI, what it means, what it doesn't mean. I'm just curious as you've clearly done a lot of studies around this and maybe talk to a lot of executives and clients about this. What are the processes where AI is truly having an impact?
Andrew McAfee: The huge area where AI is having an impact right now, and when we say AI, we basically mean machine learning. That's the dominant strain of AI. I think it will continue to be for a long time. So what's machine learning good at? And the basic answer is it's an extraordinarily good classifier and that's kind of a weird word. It doesn't sound that impressive, but it can classify pictures of animals as cats versus dogs. That's a little bit impressive, especially for cats that look kind of dog-like and dogs look kind of cat-like. It does a pretty good job at that.
It can look at an image, classify what's in it and label that image with pretty decent accuracy. It can listen to these sounds leaving my mouth and classify the words that I'm speaking and transcribe speech pretty accurately. It can look at videos of tomatoes in the field and classify ripe versus not ripe. It can look at transactions come across a merchant's website and classify fraudulent versus non-fraudulent. So in the business world we do a whole lot of classification. It's actually a really, really important category of work.
And what we're learning is just how powerful machine learning systems are at different kinds of classification.
Tania: I know you reflect a lot on this. We still need human brains. It's a very valuable piece of the puzzle of asking the right questions and exploring in the right areas. You also explain this through games and gaming. Obviously it's something that people can relate to and maybe that's part of it, but is it also because we're not as educated around what AI and machine learning is today and what it can do for us?
Andrew McAfee: We're getting rapidly smarter about the whole thing, but the reason that games resonate so well is twofold, I think. First of all, they're great test beds for all these technologies because you're playing by a pretty well defined set of rules. We all know the rules of chess or the rules of poker and the outcomes are very clear. Is the system any good? Well, how well did it do at playing chess or poker? So it's this very defined environment. But what I'm fundamentally excited about is that machine learning systems are breaking out of kind of the sandbox of playing games well and they're doing really powerful things in the real world.
Diagnosing disease, translating among languages, detecting fraud. I can't think of an industry that is not being affected right now and that will be affected a lot more in the not too distant future by this new category of technology.
Tania: And that brings up a really good point. Everyone's going to be using it in some way, shape, or form. Actually, we're probably all using it now. We just don't know it or impacted or touching it in some way. As you think about organizations who've embarked on this challenge of better data, better machine learning, better analytics, how to use it differently, what are some of the unique characteristics of the talent needed to do this beyond the data scientists and the mathematicians? What are those characteristics that are important?
Andrew McAfee: It's a cultural preference, a deep wired preference in the organization, in the company for making decisions analytically as opposed to based on gut judgment, intuition, seniority, charisma, table pounding skills, all these different ways, politics, that organizations make decisions right now. In the really interesting companies that I've had the chance to study and to learn from, it's not that all those things are gone. Those things are part of human organizations. They're never going to go away entirely. But what I'm really interested in is the hard work that organizations are doing to minimize those things and to say gang, these are important decisions for the company.
We are going to try hard to make them based on the best available evidence. And if that evidence comes out of a machine learning system as opposed to the person that's been doing that job for 20 years, if that system is better, we're going to go over there. These are difficult conversations. These are wrenching changes for a lot of organizations. And one thing that I see is that younger, more digital native geekier organizations can have an easier time of it. And so one thing that I hear from executives at more established older companies is, are we out of luck? Do you have to be born digital to be analytic like this?
And the reason I can say no with some confidence is the world of sports. The world of sports, as long as we've been playing modern sports, the games have been decided, managed, played. The decisions have been made, the strategies have been formulated by a bunch of very experienced people who grew up playing the sport. And in almost all the major sports over the past 20, 25 years, that has started to change. In some cases, change a great deal and the smart, the well-managed teams are bringing in very geeky people doing a ton of analysis to make decisions about players and strategies and tactics and anything having to do with the sport.
Then the hard part is they got the rest of your organization to listen to those super geeky parts of the sports team. I know baseball best. This sport has been played for a hundred years by one set of rules, basically one set of strategies and with decisions made by a very well identified group of people. My sport has changed over the past generation, largely because of the geek invasion. I think that's wonderful. I think baseball is a more exciting sport now than it was a generation ago and I love watching the changes unfold in front of me when I watched the sport.
Tania: That's very interesting because it's another game that people can relate to and it's an easy way to express it. We forget about how that data plays into, as an observer, our ability to consume that information
Andrew McAfee: And it hasn't changed the sport of baseball. It's still a recognizable sport. There are still players and coaches and managers and all that still makes sense and we still need all those people, but they are influenced in addition to all the other things that have historically influenced them. They're also influenced by this ton of data that is being analyzed and sifted through and massaged by some very, very geeky people, a lot of whom never played baseball at a high level. And it used to be the case that if you wanted to make an impact on a Major League Baseball team, you had to put in a long apprenticeship and be a very, very good player yourself.
That's just not true anymore. And I think that's a really exciting development.
Tania: As I think about C-suites boards, folks who are making some of these decisions at least at the top down organizations, I think everyone knows and recognizes and wants data capabilities of some sort within the organization. I recently talked to the head of a team that was doing kind of the ML and data components of the team and they said, yeah, everybody bought into it. They said, go do it. We did it and now they have no idea what to do with it, the it. I'm just curious what your experience has been, working with or advising or coaching, or understanding what the differences between someone saying yes, go do it and whatever that it might be for them and then how are they actually leveraging it?
Tania: What do we need from these executives?
Andrew McAfee: I think when there's that kind of failure mode, one or both of two things could be going on. Number one is they didn't specify what they wanted that team of geeks in that category of technology to go do. So, okay, go do ML is absolutely kind of a useless thing to say. Help us figure out who to hire better than we're doing right now. We all know there's this terrible war for talent going on. There's amazing advantage if you can find people who will be a good fit, good members of the organization, but they're undiscovered right now.
They're not as heavily scouted or recruited right now. So help us do hiring better is one thing that you could turn an analytics team loose on. And I think they'd do you a great deal of good, so that's one failure mode. The other one is let's say you ask that clear question and the analytics team comes back and says, okay, you should do the following four things differently. A lot of organizations say thanks a lot. We're going to do zero of those four things actually differently. And then all these processes of the status quo and inertia and politics and all these things come into play and you can come up with 24 reasons not to go do those four things differently.
You just see this happen over and over. And for me a big component of leadership going forward is to say, gang, if we decided that we wanted to invest in this capability, we wanted to change the way that we do human capital processes in this company, then we're going to do it and we're at least going to do an honest faithful experiment and see if it works out better. And if that first experiment didn't give us perfect results, is that an excuse to throw the whole thing away or is that a reason to go try to figure out what didn't work perfectly and go do another experiment?
Andrew McAfee: I don't know a single scientist who only does successful experiments in their career. Having something fail, having a hypothesis not work out, you don't then leave and go do something else for a living if you're a scientist. You then go back and why didn't that one work and refine what you think about the world.
Tania: That really is a definition of a good scientist is to be able to explore and dig and dig and dig.
Andrew McAfee: And Richard Fineman had, he was so articulate about so many things. He said, look, let's be clear on how science works. You start with a guess, you start with a hunch and any scientist who says anything else is lying to themselves or to you. It's okay to have a guess or to have a hunch. The science part comes in. When you go try to rigorously test your guess or test your hunch, prove yourself wrong. That's your actual goal as a scientist. And if you've banged on your idea as hard as you can and you can't prove yourself wrong, that's when things get really interesting.
Andrew McAfee: So we think we should change the way we hire people at this company. We did it and we've noticed higher levels of satisfaction from the employee, the bosses seem to be happier with what's going on, our turnover has gone down. That looks like a pretty successful experiment. Let's go do more of that.
Tania: And so I guess if we think about AI still being a little bit of a buzz word, but more and more people becoming comfortable with this concept that it's necessary, what it's doing, how it's going to work, it's still controversial. What are maybe the one or two questions you get asked all the time from folks who are trying to maybe embark on this journey?
Andrew McAfee: I get asked all the time if we're going to bring the Terminator? And the answer is no, or it's theoretically possible. But here's something else that's theoretically possible. A guy that I've learned a lot from is Andrew Ng, who's just one of the rock star artificial intelligence researchers. And when he gets asked the question about killer robots and big scary AI and aren't we going to build the Terminator and Skynet is going to become self aware, he says it's theoretically possible so are so many other things. Worrying about that is like worrying about overpopulation on Mars, which theoretically possible, it's a long, long way off and wow, do we have better things to worry about.
Andrew McAfee: So I get asked about the Terminator all the time and then there's a whole set of very good questions about ethics and bias and how do we make sure these systems are actually making good decisions because there's nothing about AI or machine learning that overturns the fundamental rule of computing, which is garbage in, garbage out. And if you train these systems on biased data or flawed data, they're going to give you biased decisions and flawed decisions. That's absolutely true. What we don't do a good enough job with is saying compared to what.
Andrew McAfee: And the more you understand human biases and the flaws in human reasoning, the more you want the technology. Our brains, these amazing things, holy Toledo, are they full of bugs and glitches and biases. The fascinating thing to me is that we make different kinds of mistakes than the machines do. That to me implies this really wonderful partnership where they will backstop us and keep us from making really stupid mistakes and we will backstop them and keep them from making really stupid mistakes. And as a result, decisions get better, we become more productive, we treat people more equitably.
Andrew McAfee: That's fantastic. You want more objectivity. You don't want your biases to rule the day. We've known this for a long time. There was this famous innovation a while back when they looked at high level symphony orchestras around the country and there were very few women on them. And they're like, wow, when you look at people taking music lessons and performing well at lower levels, you see a ton of women playing very well. Why is it that there are so few women at the high levels of symphony orchestras? So then they said, you know how you get into a good symphony orchestra is you have a live audition.
Andrew McAfee: And the way they did live auditions is the person would walk up, sit down and start playing in full view of the judges. They made one change. They made everybody take off their shoes so that a woman's high heels would not make any different sound than a man's shoes. They turned the judges around and they put a screen up. So then you had blind auditions for high level orchestras. The percentage of women in high level orchestras went up immediately and has stayed high ever since. It's just as beautiful example to me of if you can debias the decision and if you ask the judges before are you guys a bunch of sexists, they would've said, no, no, absolutely not.
Andrew McAfee: So you just have to get the biases out to let the talent shine. And the reason I'm super excited, one of the reasons I'm super excited about AI and machine learning, it has different biases than we do and I believe it has fewer biases than we do. That will let the talent shine.
Tania: That's fantastic. And that's exactly what I was going to end on is to just get a sense of what you're excited about because you've been doing this for a while, you've been very successful, lots of research. You're a scientist so you're continuously learning. It sounds like that's what's really driving you to continue.
Andrew McAfee: I just come across all these examples of when we can actually be more analytically minded. Performance goes up and hey I'm at a business school. I'm all about performance improvement. But the other benefit, as a human being, I'm trying to cheer for the home team. The other profound benefit is that it lets talent shine through. It lets diamonds be detected more easily. So if you've been a historically underrepresented member of our community, man, you want the technology and it will give you chances to manifest what you can do.
Andrew McAfee: You can learn more quickly. You can make your skills and abilities more visible. And if companies change their decision making processes, you will not have an inferior shot. You will have a better shot at realizing your full capabilities. I can get out of bed and get excited about that.
Tania: It is exciting. Thank you so much, Andrew. Andy, if I may ...
Andrew McAfee: You may.
Tania: ... For joining us. It's been wonderful to have you here and wonderful to talk to you about this because I think there's so much opportunity that we're yet to explore.
Andrew McAfee: I agree.
Tania: Thank you.
Tania: Thanks for listening. Be sure to watch the highlights from Dr. McAfee's keynote, why we're still underestimating artificial intelligence at thoughtworks.com/paradigmshift. If you enjoyed this episode, help spread the word and give us a rating on iTunes, Spotify, or wherever you get your podcasts.