Deep learning hope and hype: MIT Technology Review’s Will Knight
Why researchers at the year’s biggest AI conference focused on how to keep human bias out of computer algorithms.
Both the progress and the hype around cutting-edge machine learning techniques were on vivid display at the December 2018 NeurIPS Conference in Montreal, Quebec, says Will Knight, MIT Technology Review’s senior editor for artificial intelligence. One big question hanging over the meeting, he says, was how to detect and reverse the sexism, racism, and other forms of bias that seep into machine-learning algorithms that train themselves using real-world data. Participants also previewed the coming generation of chips designed specifically to support deep learning—a field where US manufacturers face growing competition from China. Separately, Will looks to the most exciting AI trends for 2019, including the generative adversarial networks (GANs) being used to generate authentic-looking photos and videos.
This episode is sponsored by PwC, a global consulting firm in 158 countries with more than 250,000 people. PwC transforms business outcomes and results, helping companies use digital and emerging tech to reimagine their business, from strategy and operations to tax and finance. In the second half of the show, Scott Likens, PwC’s New Services and Emerging Tech Leader, shares details from a new PwC study on the main trends in artificial intelligence that business leaders need to know about in 2019.
Business Lab is hosted by Elizabeth Bramson-Boudreau, the CEO and publisher of MIT Technology Review. The show is produced by Wade Roush, with editorial help from Mindy Blodgett. Music by Merlean, from Epidemic Sound.
Show notes and links
Will Knight: “China has never had a real chip industry. Making AI chips could change that.”
PwC Essential Eight Technologies
PwC 2019 AI Predictions: Six AI priorities you can’t afford to ignore
Elizabeth Bramson-Boudreau: From MIT Technology Review, I'm Elizabeth Bramson-Boudreau, and this is Business Lab, the show that helps business leaders make sense of new technologies coming out of the lab and into the marketplace.
This episode is brought to you by PwC, a global consulting firm in 158 countries with more than 250000 people. PwC transforms business outcomes and results, helping companies use digital and emerging tech to reimagine their business from strategy and operations to tax and finance. Later in the show I'll be talking with Scott Likens, who heads PwC's new services in emerging tech practice.
Scott will share details from a new PwC study on the main trends in artificial intelligence that business leaders need to know about in 2019. But first we're going to look at some news from an important conference in December on machine learning and computational neuroscience. Our own Will Knight, Technology Review's senior editor for artificial intelligence, was there.
Will, thank you for taking this long trip down the corridor to have this discussion.
Will Knight: Thank you for having me.
Elizabeth: So we're going to talk about A.I. and about your beat. But I'd love it if you would tell us about the conference you were at not too long ago in Montreal.
Will: Sure. Yeah. I'll tell you about this conference called NeurIPS, which is the biggest AI conference of the year. And for me it really sort of captures the excitement, the real progress and the hype around AI. So this is an event that maybe 10 years ago was just a few dozen researchers, all of whom happened to be working on deep learning, this this trend which has now taken the entire tech industry by storm. And so a few years ago the big tech companies realized that this was going to be a phenomenal technology. So this is, I should explain this is technology which involves using a large simulated neural network which you train on data, and then it can do stuff. Originally they were demonstrating you could do things like recognize faces and images and recognize the words people are speaking, but people are applying it to absolutely everything now. So a couple of years ago, the big tech industry realized that this was going to be huge. And fast forward a decade, and now there are literally 10,000 people coming to this event. The tickets sold out in 15 minutes, believe it or not. Lots of people couldn't get tickets. There are just dozens and dozens of big stands. You know the auditorium, the main hall, was absolutely full of people. And it touches on all of the major themes that are very very important including as I say the sort of hype cycle.
Elizabeth: So what were the most interesting or maybe surprising things that you learned at NeurIPS?
Will: One of the most interesting things I think was that there was a huge focus on diversity and bias. So, the industry is impressive in the sense that it's taking some of these social issues on board and really you know they recognize that the technology is going to be so pervasive and have such a big impact that they need to start thinking about those. And so you know it's not something that maybe computer scientists are normally used to wrestling with. But that's kind of impressive. I think it's also important for companies, anybody trying to commercialize the technology. You have to think about the issues such as bias, because you don't want to train systems that are going to then behave in biased ways. That can be a huge, that can cause huge PR problems. It can cause your technology not to work as well.
Elizabeth: It can cause bad outcomes too, right? You can completely miss an opportunity.
Will: Yeah, right. You miss opportunities. You know companies like, even Google, which is probably one of the most advanced A.I. companies on Earth is experiencing problems with some of its, has experience problems with its image recognition systems that are biased maybe towards certain races. And as you say, no matter where you're going to be deploying this you might end up with results that are not representative. They're not the ones you want to act on. So it's a very important issue and that was something that really struck me from the conference.
Elizabeth: All right so tell us more about some of the people that were telling stories at NeurIPS or showing their technology. What are the two or three things that you saw that you know that you've come back in to tell the family and tell the folks in the office about.
Will: I don't know about tell my family but maybe! One of the things is that hardware is cool again. So you know we have had a number of decades where there's a lot of interest in algorithms and in big data sets. But now people are realizing that the computer chips themselves, which are normally you'd think are just too costly, too hard to really innovate that much on, people have realized that, and it reflects the fact that AI is such a big deal that it's worth redesigning these chips. So you have a whole bunch of chip startups. You have Intel also talking about its chip, its new chip designs, all designed around deep learning. It's an amazing kind of turnaround and it's going to affect the whole chip industry and has, as we write in a recent issue of Tech Review, this is something that affects the global innovation landscape, with China trying to improve its chip industry.
Elizabeth: So you're referring to an article that you published in our China issue, which is out in January-February. And in this China issue, you write an article about China's push to develop an AI chip. Talk about that race, and is that some of what makes hardware cool again, that there is in fact, there are stakes to be to be fought for?
Will: Absolutely yeah. So, catching up in the chip industry has been a huge challenge for China. It's incredibly difficult to do. But the fact that you've got this big technological shift like it once in a sort of generation one means that it's now possible, that the playing field is leveled a little bit. And so you have a huge amount of activity in China with these companies coming up with AI chips. Both ones to run on devices and ones to run in the cloud. And so yeah this is, there's a lot of, the current trade war is largely focused on chips. But one thing I think a lot of people aren't realizing is that the game may be changing slightly and China may be able to get a bit of a leg up in terms of some of the new chip architectures.
Elizabeth: That'll be interesting to watch. What are the companies in China that folks listening to this should begin to gain familiarity with?
Will: Well there are several startups that are there still, I say they're relatively small, they're unicorns, so they're worth billions of dollars, which show you, tells you how much the chip innovation industry is worth. So one is Cambricon which is backed by the Chinese Academy of Sciences. Horizon Robotics is another one. I spoke to the CEO of that company. Bitmain is another one which is kind of fascinating. The company started out making chips for crypto mining, for Bitcoin mining, and has pivoted, as everybody in China has had to, away from the crypto mining to AI chips. So there's a lot of activity going on there and it's just been fascinating to watch this, because the other thing that I really discovered in reporting this is or that was brought home reporting it, is that chips are so fundamental to other areas of innovation, advances in chips are going to power self-driving cars, 5G networks, cloud computing. So this is really a foundational technology.
Elizabeth: You mentioned hype and hype around AI versus the reality. So what's this about. Tell us more.
Will: Well I think this is a really fascinating subject. As someone who covers AI, and I'm very interested in the history, the industry has been through these cycles of boom and bust. And seeing what's happening now, I have mixed feelings, because there's amazing progress, and there's amazing commercial benefits coming out of the technology that's been developed. But there is clearly a huge amount of hype and I think you can see some of the AI researchers clearly, including some of the organizers of NeurIPS, are worried about this kind of snowballing and there being unreasonable expectations. And I think it's actually also, practically, a really important question, because companies will rush in, and rightly so, to try and commercialize the technology, but there's a lot of hucksters out there trying to sell their latest AI and there's a lot of unreasonable kinds of expectations. And so if you move too quickly and if you aren't careful, you might mess things up. But again you want to be left behind. So I think it's kind of one of the most fascinating things to me, is, how do companies who maybe don't have that expertise, how do they benefit from this without being sucked in and misled by some of the hype.
Elizabeth: And that is exactly what our listeners want to know because they're following along with the AI stories. And maybe some of those stories are hype stories or maybe they're actually giving it to them straight. So what would you advise a business executive listening to think about when it comes to commercialization and the way AI is likely to play into their company's day-to-day?
Will: I think that's a great question and I think one of the things I've noticed is it always seems very important to have a mixture of AI experts and domain experts in any place where you're applying the technology. So if you take something like medicine, when you see just machine learning experts trying to do something, I think that they'll they often run into problems and so having some combination of expertise seems very important. I recently actually spoke with Andrew Ng, who's a real pioneer in A.I. and has worked both at Google and Baidu in trying to, in helping them commercialize the technology. So he's really at the forefront of that. And he came up with a playbook for how to how to use AI. And he told me that a lot of CEOs he's talked to are making real mistakes. They're pouring money in and making errors. One of the things that he said is. just because you have a huge amount of data doesn't mean you are necessarily going to get value from it. And you need to really sort of think about how you're going to use it. He also suggested having small example cases where you can test out the approach rather than kind of revolutionary knows your entire company because sometimes there may be. You know it may just be the case that your company's not going to be totally AI first. You know you may use it in certain areas. And I think also just exploring different techniques because as we've been talking about deep learning is incredibly important, incredibly powerful. But sometimes other techniques might be might be quite useful as well.
Elizabeth: Yeah, I mean, I have to say as a CEO I'm constantly getting emails from people who want to sell me one or another quote unquote AI solution to some imagined or projected challenge that I face. And I'm guessing that many people listening to this are also recipients of those kinds of email. So how do we think about, where do we even go? I mean I can think of ways of using AI in MIT Technology Review. We are doing certain things with our data. But I don't think it's going to be the silver bullet. And I guess the question is, for companies that do have tremendous amounts of data, and that are aware that their competitors are gaining ground through the use of AI, where do you suggest people look?
Will: Yeah, I think that you know building up expertise in a careful way and combining it with your existing business equities is probably the key thing in any situation I think. So for example when you look at Kaggle, which is this famous platform for data science. Google bought it. It is really sort of a foundation for a lot of people who go into AI and the people who do really well at that are not people who are the greatest, necessarily just the greatest mathematicians or algorithm developers. They usually go off and they research the problem in a very very deep way so they understand this particular, you know, image recognition problem or optimization problem. So I think that that's really key to remember and companies are just going on and on about AI are probably not, to me is always a red flag. I think you need want to talk about where they are actually, how you're deploying it and where the results are coming from.
Elizabeth: So, Will, what are the other things that business leaders need to be thinking about doing kind of foundationally, as they get ready to either set their strategy or set the strategy for their AI strategy.
Will: Right, well, one point brought up by Andrew Ng was I think a very good one which is figure out you know what your strategy is regarding your data. And that may well involve figuring out how to label it, how to get it, how to collect it, how to make sure it's all in a comparable format. which is a not inconsiderable challenge also.
Elizabeth: Can you tell us why labelling data is so important.
Will: Yeah. So you need to label data especially with something like deep learning because without having labels applied to it—say if you're if you're trying to have a machine learn to recognize images of dogs in photographs—without those previous examples it doesn't have anything to learn from. It's just raw data. There are approaches which are unsupervised, which don't require labeling. But in most cases, the most powerful examples of deep learning, you have to go in and mark up your data to give the system examples. And then going forwards it's able to perform often as well as a person in order to do things. But say if you were in a medical imaging business, it's vitally important, you can't just throw your data into this algorithm, you need to have people go and label the data, often in great detail, so that the algorithm can then can then learn from it. So yeah I think having a data strategy is probably an incredibly important thing. You need to you need to figure out what the opportunities are and definitely think about how you go about labelling and managing it.
Elizabeth: So given that it's now 2019, Will, what are you excited about when it comes to your reporting on AI for this year?
Will: So I'm one thing I'm very excited about, I think everybody is really excited about, is the emergence of this technique called, this tool called GANs which are generative adversarial networks. We've written about them explaining about that, if you need if you want to learn more about how they work. But essentially what they're able to do is learn from data and then synthesize examples of that data. And practically speaking that means they can do things that are kind of astounding, like generate artificial celebrities, generate complete scenes from scratch. And it has huge practical applications for things like video gaming and CGI. But it's also kind of profoundly interesting in terms of just allowing people to kind of manipulate and generate fake video. And that obviously has some slightly concerning implications when we consider fakery and fake news. But it's a pretty amazing tool and something that has only emerged in the last few years which is extremely exciting and I'm fascinated to see where that goes next, I'm sure
Elizabeth: You've done some stories, including a video, if I'm not wrong, using a technique like this.
Will: That's right, I did this for our issue dedicated to politics, and one of the things that's amazing is that, these tools people have created some you know off the shelf software for using these tools, so even someone like me was able to swap one person's face onto another in a video in an almost seamless way. So I took Ted Cruz and put him on Dancing Paul Rudd just to show that it's possible to do this. And again I looked into the idea of computer forensics and one thing that's kind of amazing is that the experts in this field are quite concerned because they the very nature way which GANs works means that it's adversarially trained. It's trained against means to detect fakery. So it's maybe virtually impossible to detect these things. Which may change the way we come to view video in the future, for example. When you see something pop up on social media it's probably going to be quite important not to assume that that's the ground truth, right? Because it could be generated by anybody and these things are getting better and better.
Elizabeth: What else?
Will: So another topic that I think is going to be very important, we're seeing just because AI is such a powerful technology, such an important technology, this is becoming a real political issue. And not just in terms of governments pouring a lot of money into it. You've heard about China doing that. The US is getting ready to put more money into it. And Canada and so forth. But people are starting to think about, the technology is so important so powerful that we're going to have to have some standards. And I was recently at an event, sorry I was at NeurIPS, where at a secondary event at the G7, the Canadian Prime Minister Justin Trudeau came and talked about this plan for a panel based on the IPCC, the climate change panel, about AI that that Canada and France are putting together. And they want to they want to bring in other G8 G7 members to talk about issues such as you know how do we use AI for surveillance, how do we deal with things like bias, should we have AI weapons. So it's going to be really fascinating to see how that pans out and to see, especially given the rise of China which is, in some ways there's rising tension but maybe there's going to be opportunities for collaboration and we're certainly as a global society going have to figure these things out. It's going to be fascinating to watch that.
Elizabeth: Will, thank you so much. I look forward to reading your coverage in 2019.
Will: Thank you for having me.
Elizabeth: This episode of Business Lab is brought to you by PwC, the global audit and assurance tax and consulting firm. Among many other things, PwC helps clients identify opportunities to put technology to better use. The company compiled a list of the Essential Eight technologies that matter most for business today. Artificial intelligence was at the beginning of that list, not surprisingly. And just recently the company issued a list of six predictions for the most important AI strategies that business leaders should be considering in 2019. Here to talk with us about those predictions is Scott Likens, who's PwC's new services in emerging tech leader. Scott, thank you for joining us.
Scott: Thank you for having me. Excited to be here.
Elizabeth: So this role of new services and emerging technology leader really connects nicely with what we do and what we cover MIT Technology Review. And I know that you and colleagues at your firm advised business leaders to pay attention to the Essential Eight. And those were AI, augmented reality, blockchain, drones, IoT, robotics, virtual reality, and 3D printing. That's a great list. It definitely coincides with our coverage here. But notably, I note that AI was the very first on the list.
Scott: It was.
Elizabeth: And I'm wondering if that is specific or if that's alphabetical order or if there's some larger rationale for that.
Scott: You've figured us out. It's alphabetical. But it's easy to have it as an alphabetical, because then you can't argue about what's the most important of the eight. And that's one of the messages we're really standing behind, is these eight—while they are emerging technologies, they're the essential eight. You have to be thinking about all of these no matter what your business is.
Elizabeth: To what extent do you see the advances in AI undergirding the other emerging technologies on your list, in a sense?
Scott: So this is the year we actually came out with our convergence themes, because it became time for AI to really be the amplifier versus you know just the solution. So if you think about the investment in AI, again, for decades of course, but in the last five years we've had access to data you know at scale that we've never had before because of cloud And you don't hear the term “big data” anymore but, you know, the sense of data coming from everywhere. But if you started to look at the maturity, because of that, we started to look at IoT to get more data from devices. We started to look at virtual reality to embed or create an experience. And AI was absolutely critical. So I think that's been the pivot. There's been this acceptance that AI is something we're going to do. But now we have to actually build it into, how do we create the mechanics of the value behind AI. And we needed those endpoints. And I think a lot of the other Essential Eight gave us those end points that made it very real to a business or to a customer or to an employee.
Elizabeth: I've heard many AI proponents say that when AI is used to automate repetitive or tedious tasks, or sort of rote mechanical activities, that humans will be freed up to do things that humans are best at, like analysis, or like connection, person-to-person, either customer care or medical care whatever. Do you see that that as actually what's happening? Or are you seeing some companies using productivity and efficiency increases to shrink the size of their white-collar workforce?
Scott: Yeah let's just go right at the issue: Is AI taking away jobs? It is. In the sense of the jobs changing. That means the job could be gone. But we're seeing the investment in moving the workforce up that curve of what we're doing. You know I can say we as PwC realized the job of an accounting is changing. If you think back 30 years, when the spreadsheet didn't exist, the spreadsheet came out. It changed the job of how we do everything about accounting. We're in that same curve today. And I think the knowledge workers have to realize it's not a bad thing. We're going to up our game as humans and do the new stuff. And I think we're helping people through that transition. The workforce of the future is going to look different. We're going to use mind plus machine to be more efficient. And if you look at our AI predictions, in 2017 we looked at the impact on GDP by 2030 of AI was around $15.7 trillion. About half of that was in operational savings, but the other half was in really opening up new markets, new products, new businesses, new ways for us to actually engage customers. So there's a good upside there. So I think we have to balance the fact, and just accept the fact that jobs will change. Which means some go away in the sense of the mechanics of what they do. But those folks, those individual resources could actually should come along in that journey and that workforce of the future will always use automation and technologies to their benefit and make that pivot.
Elizabeth: So if roughly half of the economic benefit that AI will have for us will come from savings and greater efficiencies, and the other half will come from new products and services, can you give us some concrete examples of what you have in mind and say what areas are ripest for AI-assisted improvement.
Scott: Sure. So I think the two that jump out at me, and I just spent the last five years in China, which was pretty amazing, if you think about what happened around AI. There's huge investments because China had a market that had tons of data right. They started with collecting and getting data and then made these big investments in AI to actually understand consumer behavior at a really granular level. So when you ask what the specific examples, let me start with, the easy one for me is financial services. So whether it's banking or insurance, we've lived in this world, obviously it's a highly regulated industry and there's a lot of rigor that needs to be put around it. But at the same time the products haven't really changed quite honestly in insurance maybe a hundred years. Think about the ability to craft a very bespoke coverage or protection for an individual. It's really hard to do today. AI is giving us the ability to actually understand the consumer behavior very intelligently, and could actually allow us to create bespoke products on demand in this real time world we live in and you're starting to see that already.
I think retail and consumer is another easy one. If we think about the way products are manufactured today, there's ideation and innovation but then there's products pushed out to the market. And we try and find customers to buy them, versus really creating, again, bespoke products based on customer behaviors. As you start to combine things like 3D printing or on demand manufacturing, AI could be almost improving products that I've bought specifically for me, because it understands much more about what I'm going to want based on the data it has access to about me as a person.
Elizabeth: I'd be very interested from your vantage point in understanding why you think it's so hard for leaders to get their heads around AI and sort of really go at the obstacles that are in their way.
Scott: Yeah, sure. I think two things jumped to my mind. One is the structure of most organizations don't support AI. I think we went through a wave of what I call big data on to analytics, which was tough enough, because we started to try and understand organizations across silos of business units or operating units. So we got there. Now we're in this wave of AI, which can break down those barriers. So we're seeing executives embracing this this momentum around AI, but not understanding how actually to tap into it. So AI has also been you know quite academic in the sense of, it was a very unique skill set and we understood that it could maybe help our business but we weren't able to tie that to a return on investment. So we have to actually change the entire model across the organization.
So I think that makes it really hard, and executives are struggling around how do we actually you know elevate AI. So one of the things we asked them in our in our current survey, 59 percent plan to really invest in the data and AI specifically across the entire organization, versus having to do it individually within business units. So pulling it out in a way that they can actually use it across the organization. But it's a tough operating model. It's a very unique skill, sometimes. And where it sits, we asked again the executives, and there wasn't a clear winner on, did it sit in a business unit, did it sit under the CIO, was it a center of excellence? It was it was pretty evenly split. So I think we're still trying to decide what the best operating model around AI specifically is. And I think that's all infused in this in the struggle of who's going to go first. I'm going to wait for those great case studies that make me kind of rally around the investment. We're still in that gray area of how we're going to get this infused in our business.
Elizabeth: So what do you do first? I've heard you say that organizations shouldn't shoot for the moon or think that AI is going to transform the business overnight as some sort of magic silver bullet. What does a business leader need to think about doing first and where should he set her expectations for the near term?
Scott: I think she should demystify AI. It's not magic in the sense of what it's doing. I think we have to accept the fact that it's here. We've had the quote unquote AI winters. I think we're over that now. We now have the data. We now have the access to understand more and more about businesses, more and more about customers. I think you should make some big bets on the business in the sense of the areas that this will actually be accretive and beneficial to everything we're doing. And we should we should chip away at small pieces. And you see opportunities within the operations side of the house, so finance and H.R. Things that have a lot of manual effort, we can use A.I. and automation to really help us there. At the same time we have to get the word out to our employees and our customers that this is a good thing in general. It's not that jobs are just going away, it's that jobs are changing and it's changing for the better.
So we have to start that journey now and we have to show them the power of what this could do for our for our company, for our society, whatever it is. Whatever the mission of this leader's organization, we should rally around that and figure out ways to do, both at the business level and the community level, small experiments to show the value to understand and demystify and just get started.
Elizabeth: Well, Scott, thank you very much. I think you've given us useful frameworks and ways of getting started with AI. I appreciate the conversation.
Scott: Thank you. Appreciate the time.
Elizabeth: That's it for this episode of Business Lab. I'm your host Elizabeth Bramson-Boudreau. I'm the CEO and publisher of MIT Technology Review. We were founded in 1899 at the Massachusetts Institute of Technology. You can find us in print, on the web, at dozens of live events each year, and now in audio form. For more information about us please check out our Web site at technologyreview.com.
This show is available wherever you get your podcasts. If you enjoyed this episode we hope you'll take a moment to rate and review us at Apple podcasts.
Business Lab is a production of MIT Technology Review. The producer for this episode is Wade Roush with editorial help from Mindy Blodgett. Thank you to our sponsor PwC, a global network with one goal to solve important problems and build trust in society. Thanks for listening. We'll be back soon with our next episode.
Geoffrey Hinton tells us why he’s now scared of the tech he helped build
“I have suddenly switched my views on whether these things are going to be more intelligent than us.”
ChatGPT is going to change education, not destroy it
The narrative around cheating students doesn’t tell the whole story. Meet the teachers who think generative AI could actually make learning better.
Deep learning pioneer Geoffrey Hinton has quit Google
Hinton will be speaking at EmTech Digital on Wednesday.
We are hurtling toward a glitchy, spammy, scammy, AI-powered internet
Large language models are full of security vulnerabilities, yet they’re being embedded into tech products on a vast scale.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.