No company has made bigger investments in cutting-edge A.I. than Google. But those advances have not come without challenges. We’ll hear from one of the company’s key executives on A.I.’s potential impact and how the technology can be made more accessible for all.
Dr. Andrew Moore, Vice President and General Manager, Cloud A.I. and Industry Solutions, Google Cloud; Former Dean, School of Computer Science, Carnegie Mellon University
Interviewer: Jeremy Kahn, Senior Writer, FORTUNE; Co-chair, Fortune Brainstorm A.I.

Fortune Brainstorm A.I. 2021: A.I. for all

Conferences

Alright. Hi everybody and welcome to the first ever fortune brainstorm AI summit. All right, thanks so much for joining us here in boston. Uh This has been, we were supposed to be here right at the beginning of the pandemic and we had a little bit of a curve ball you might have heard about that but we're so excited that this was able to come together and we really thank you for joining us and and making this a great event. What is sure to be a great event and also to all of the people who are joining us Online. We have a big group of virtual attendees this year which is, we're doing a conference in 2021 so that's what it's all about. Um So what could be better than gathering a group of highly intelligent humans that's all of you to talk about and share insights and expertise on the increasingly robust simulation of human intelligence that we call artificial intelligence as you know. Now more than ever, artificial intelligence is being woven into our daily lives and becoming central to business processes and strategies, not just on the tech side of the organization either as I've heard a lot of conversations already, but throughout a company's entire ecosystem and that's why we're here to delve into that. Over the course of these next two days, we're going to go in depth on how artificial intelligence is unfolding and how ai is shaping all of us both, the business world and society overall. We're going to look at how it's transforming companies and industries and disrupting traditional models. We'll delve into how it's being used for R and D. Precision medicine, transformation, transportation, its transformation and every aspect talent recruitment, supply chain operations and manufacturing methods and we'll talk about the ethical boundaries and legal challenges of ai. These conversations will be led by members of Fortune's editorial team and by my fellow brainstorm AI co chair. So I'd like to take just a quick moment to acknowledge them. Fortune senior writer Jeremy Kahn who will be hearing from in a moment. Fortune senior editor VErne Kopytoff, Fortune technology writer Jonathan van Ian and Deputy ceo of Smart I dr rana El Halabi who's joining us right here in the front row boston native hosting us here are not boston native boston resident, sorry, I'd also like to take a moment right now to thank our fantastic event partners for their work in bringing us all together and it's really been a journey and we really appreciate the collaboration and partnership we've had, I'll start with our presenting partner accenture and I'd like to acknowledge Sanjeev Arora who joins us here at brainstorm Ai he's the senior Managing Director and global lead of applied Intelligence at accenture. We'd also like to thank our event partners intel and Salesforce dot org. We greatly appreciate their work and making these conversations possible for all of us. Okay before we get started with the actual interviews and discussions a little more housekeeping to go through what's said on stage up here, we're in the breakout sessions is on the record. But what set over the table during a meal or you know, just huddling out there in the networking space that is off the record unless it's agreed upon by all parties involved. We'd also like to mention the ever present covid safety protocols which are here for all of our safety. Be sure to keep your mask on except when eating or drinking. We'll try to model good behavior ourselves and avoid shaking hands, which I've already violated that I'm so excited to be in person meeting people. I've shaken hands with several people, don't tell anybody, but I am sanitizing my hands constantly. Speaking of which utilize the hand sanitizer available throughout the facility. And please do not replace any items that you've touched that may be used by others such as cups or utensils. In keeping with that. We want to mention that the microphones that we're using for the Q and A will be switched out and sanitized before each use. So you don't have to have any uncomfortable feelings about using the microphone. We want you to interact and use the microphone and ask questions. We're all here to have a interactive conversation. And for those of you joining us virtually. Uh and I mentioned, we have a great group joining us virtually as well. If you'd like to ask a question, please drop it in the chat function on the right side of your screen and we'll do our best to get our thoughts to the speakers during the course of the conversation. So with that let's get started. Google is almost synonymous at this point with artificial intelligence and the company has made massive advancements and investments in the field across the industry. We're seeing spiraling costs for developing cutting age, uh, cutting edge AI and by some estimates, the computing power required to train Big I models is doubling every 3.5 months. That's a pretty eye popping step requiring new technologies and hardware. One of google's top executives is here to share his view of the challenges, the opportunities and the future impact of the technology and how AI can be made accessible to all. Please welcome dr Andrew moore, Vice president and general Manager of cloud AI and industry solutions at google cloud in conversation with fortunes, jeremy Kahn. Oh Andrew. Thank you so much for being here with us. Um, and while while we're talking, please guys start thinking some questions because we're gonna try to have some time for Q and A. And that goes for those of you who are listening virtually as well. Um Andrew, we had a little discussion before the session and one of the things you told me is you do not want A I to be reserved for a select priesthood. Can you talk to me a little bit of what that means and what is google doing to really kind of democratize this technology. Very good. Yes. So, uh, there was a time 10 years ago when a PhD student could do something remarkable with computer vision or machine learning and natural language. But as we've all seen, and I hope it's a major theme of this conference. Ai is not primarily about the technology, it's about change and how you're gonna change the way something in the world works in a way that's good for the human condition. The people who can do those changes are not the ones who have been to MIT or Carnegie Mellon for seven years for a PhD. They are the folks out there actually living the business. And so I think it's fair to say that us internet companies were racing now, not just to have the most advanced technology, but have the most accessible, Right? Yeah. Andrew Google has invested a tremendous amount in cutting edge ai for its own business, but also to build out these tools that can offer to its customers, including its cloud service customers. But there are a lot of companies that are wary of this. They're worried about getting locked into one cloud provider for instance, and they're also a little bit concerned about tools that they have not trained themselves on their own data. I'm curious to google, how are you addressing those concerns? And do you think that's a legitimate worry for these companies have. I think it is by far the most common thing now for companies to go with a multi cloud approach uh and speaking both for myself as a academic practitioner and for google, I think that's incredibly healthy. So uh we don't like to see it when there's a system where every part of the stack depends on uh every other part of the stack which requires this kind of lock in, what What I have found. And this has been a remarkable thing for me, just over the last 18 months, is how difficult it is to ask a customer over a customer internally to suddenly produce a petabyte of data to train a machine learning model for that customer. So uh we and google because of this, we've been working so hard on the question of how to reduce data requirements for users of ai and if you don't mind, I want to give an example of this yet. So uh visual inspection, Very important thing in manufacturing is a non-0 sum improvements to have fewer defects in parts and finished products. Uh But if every time you want to train up an ai to spot floors, you have to spend months sort of showing it positive and negative examples, that really reduces the speed of uh usage of this kind of thing. So uh some of my more uh cantankerous engineers got fed up with this uh and really pushed hard on the question of what is a generic notion of a fault, is that faulty nous as a general concept and just trained up a eyes to across the board, get an idea of whether something is faulty or not without having been trained in a particular discipline. And to my surprise and horror because they really were cantankerous about it, they were right, we actually can do this kind of thing. And so now, uh we are not in a position where the first thing we have to talk to with the customer is persuade them to give us like a million examples of a good product and 1000 examples of a bad product. It's more like 10 of each and that's been a huge Ben. Well, that's interesting. I mean, we've seen this with computer vision, as you say, obviously that's happening with language where you have these sort of foundational models. Um, but and and google is using these, it's got Bert, it's got mum the planes, the search results, but as you know, there's been a tremendous concern about, well, what if there's bias kind of baked into these foundational models? What if in learning that embedding of what is a flaw look like there's some bias that you're not aware of and then you're going to use that to train, you know, uh, with a few shot learning, you know, for your specific use case, what what happens if you kind of ingest that bias? This has been kind of controversial, you know, within google itself, What is google doing to kind of address that concern about bias? So this is a genuinely hard problem and I would love to be able to tell you that there's a great google approach at the moment. I would say we are extremely conservative but our approach is very people intensive. So if you take uh an implementation of an AI pipeline, maybe something for loan approvals or job applications or uh research, health research study, uh eligibility and so forth. These are all extremely dangerous ones where it's perfectly possible for the training data to inherit some of the biases in the world. Each of these cases has to have folks looking not just at the technology piece but at the change whatever change management, remember I said ai is change change management is involved in it. To ask what are the ways in which this could make things even worse or make things better for our sort of attempts to have a inclusive world. So of course, we have statistical tests to do various kind of stratified analyses to to give a warning light if there's an obvious relationship unexplainable correlation between the protected attributes and an outcome. But that's definitely not enough right now. And I will push back very hard if someone says there's going to be an automated engine for this. One of the other big concern that's been raised about these models. Some of them are very, very large. Um is the environmental impact of these models again, what do you how do you view that concern? And do you think, you know, there's a solution for that? Yeah, that one really uh is a sad one, especially if some of our models are being used to reduce energy in places like our data centers. It kind of doesn't work if you actually consume that much energy to learn the model. So, this one, it turns out is real, it's absolutely a problem. And if, say, for example, we were able to give the best 20 0 ai researchers in the world the power to build the kinds of models you see in google or open ai bills in language. Uh that would be an absolute crisis. There would be the need for many more large power stations to support it. So what can we do? The first thing is lower power compute. So people are working at the hardware level, the I come from a more mathematics software background and she's still scared of getting involved with hardware. Uh but at the mathematical background level there's so much you can do and focus with google and around the world are working on it. For example, there's a technique we've been using with self driving car company is called neural architecture search, which as it's trying to train for really big machine learning model to recognize surprising obstructions on the street. Part of its training is you're only allowed a certain number of what of energy in your predictions. So don't just optimize for accuracy, optimize for small, sparse models. So I do expect you to see big changes there and I would not be happy if everyone tried to uh learn these big models because it is selfish and bad for the environment. I've got plenty more questions that I'd like to post Andrew, but I want to make sure we go to the audience while we have time. Uh there's someone back there at the question. I want to go, we'll go to you first. Um if you could please state your name when you, when you stand up to ask the question, I'm tom davenport's um I know you've gone back and forth between cmu and google for a while. I wonder if you could talk a bit about the current state of high end ai education and where you think that might be going? Great question, but I want to change a little bit. So high end air education is going very well in the sense that there are plenty of qualified applicants and there's a boom there, the professors who can be their mentors while they're learning how to be wild and crazy researchers as opposed to industry researchers. I'm really, really worried that those professors are dissolving into internet companies at the moment. And part of the problem there isn't actually to do with funding for the professors a lot of it is to do with the cost of funding. So a typical professor mr of me, crazy Of all their free time for research at the moment, they're getting close to 50% of it is trying to chase grant money and only 50% of it is actually spent doing research, so they always feel like the grass is greener. On the other side, interesting, I want to go to anyone on this side of the room, have any questions over here because that gives them a chance to clean the mics. Um Any other questions about this gentleman here has a question, wait for the mic please, and then if you can stand up and state your name, okay. Uh and um uh I'm from Gapping. Um Andrew speaking of academia obviously one of the big challenges in this space is talent always has been, can you talk a little bit about that? How do we, how do we solve the challenge challenge, particularly in partnership with academia? Great question on Thailand one. Uh This is going to sound surprising, but I am not particularly interested in injecting vast amounts of extra money into the big academic institutions where I really want to invest in two year colleges places where you can train folks to use the tools which good clouds, I think of one particular good cloud that I really like have made their systems. Mhm Not easy, but straight forward to build something which once you've built it, you know, it can safely go into production. So it's those kinds of folks using these technologies, which I think are the most important. So in my own sort of education lobbying right now, I see our biggest weak spot and biggest opportunity is two year colleges in the United States. We've got another question from this gentleman here. Yeah, hired roof, bookish the lawless, deadly black attacker. So google has lots of data sets you have about the boast of the world. Are you going to do anything about open sourcing some of those, not just to academia, but to other businesses, especially those who don't complete with you? Yes, it's a great question. So I want it's an opportunity to talk about something really, really important. So, so many companies come to me asking to work with google taking advantage of some of the reported data that we have about users or the state of the economy or other information and we in google cloud, we have a strong role. We never ingest data from the other parts of google's businesses for these purposes. So I'm actually not a good person to come to if you're saying like, please, can you use the results of google searches to predict where I should put my next distribution center or something outside of that. There's still huge amounts of data and a lot of it is sitting there publicly and it's hard to access because there are no shared data models. So there's an important google actually, I think I think this could be one of the most important things that google does over the next few years. A huge project called data commons dot org. Uh It's it's owned by a whole conglomeration of folks, but some of the engineers or in google, they are busy getting hold of public datasets, not republishing them particularly, but making sure they all mapped to common data models so that you can start to do things where you might ask a question about like the economic growth of one part of the country. And it turns out that linking that to uh Ocean Ocean National Oceanographic Institutes, whether models actually helps and you can do the linking right Andrew. We're almost out of time. But before we go, I was hoping you could give the audience just a preview of what you might see next, what they might see next in terms of cutting edge enterprise ai coming out of google. Thank you. So, uh one of the things I'm most excited about is there's we still even when we're able to produce the world's most accurate core center uh Question answering systems or uh excellent retail search results which delight the user and so forth. These are in my mind and in our 10 year roadmaps, minds, all components, what's coming together coming next. So you will see these apparently unrelated things to do a document, understanding speech retail prediction uh question answering for users coming together in places where if a customer is asking about the availability of a particular product, there's the speech part, but there's also access to inventory systems, access to predictions about inventory and these really important models as to what makes people happy. Great. Andrew. We're out of time. Thank you so much for joining us. Uh, it's been a great discussion. Thank you so much. Thank you.