Anna Patterson

Transcription

SUSAN WOJCICKI: And now I'd like to invite Anna Patterson to come up on stage. Anna Patterson has worked on some of the biggest, most ambitious projects I've seen at Google, and she's going to tell you a little bit about them. Anna is Vice President of Engineering and Artificial Intelligence. What an amazing role and opportunity. So let's please welcome Anna onto the stage. Thank you. [APPLAUSE] ANNA PATTERSON: So I'm here to tell a little bit of hacker stories from the trench, and I think with a main theme of perseverance. I saw that the ENIAC programmers, how they got a circuit diagram in order to program. My first programming job out of school was for the F-15E, and we would test our software the best you can on something that wasn't flying. And then we'd invariably need to put in patches. And just like the ENIAC, you had to toggle in the program. So it was very scary since people's lives were depending on it. But also I was thinking about how source control has evolved. Of course, my first programs in high school were on punch cards. But at home, I was really fortunate that my dad brought home a Commodore Pet computer, and he encouraged us to experiment and write a program on it. And I wrote a really, really horrible, terrible video game, and it only had 8K of memory. So I stored the program state in the corners so they were flashing weirdly. And, you know, I guess it was a standard trick, I found out much later. But I was from the suburbs of Saint Louis, so I wasn't really part of a burgeoning hacker community. I eventually got my PhD in computer science at the University of Illinois, Champaign. And then I became a research faculty member for John McCarthy, the founder of AI. And I see several people here who know me really well, and so you'll appreciate when I say that John McCarthy made me do something how hard that must have been. So he made me look into this area called phenomenal data mining, where you're given facts and relations, and you're given a big data set. Big was very small in those days. And then when you learn new correlations, you can explain these new correlations in terms of underlying phenomenon. So that's where it got its name and that technique is still in use. But everybody was leaving the department to start companies, so we were pretty much like the last ones out when the lights were turning off. And I co-founded, in 1998, a web mining company. Because the web was so big at 5 million pages, we thought, how are people going to be able to get around this thing? And so we had an idea to make a automatic taxonomy and kind of lead people to what questions were answerable on the web and what ones weren't. Because in those days, trust me, there weren't really recipes on the web, but the taxonomy around Linux was deep. So we got angel funded to do this just in time for the dot-com crash. And so I tried to figure out what my other passion was, and I said, mobile is just about to take off. So I moved into the mobile area in 1999, just in time for the telco crash. So after killing two industries, I was actually-- I had a one-year-old, and I was four months pregnant. And of course, I was unemployed and I was actually living in Pavni's guest room, who's in the front row. And so the doctor said, you know, you need to be on bed rest. And I said, have you ever heard of programming? And he hadn't. And I said, it's a lot like bed rest. So I volunteered at the Internet Archive, and they have snapshots of the web over time. So as such, especially in 2001, it was much bigger than the web. It had 30 billion web pages, and the live web at the time maybe had 5 billion. So I did a history-based search engine that launched at like, 12 billion pages, and you could look at how terms change over time. So you could dial back to 1997 and search for a term like "Clinton," and you'd see all these co-occurring terms in a graph and how they changed over time, and it was very presidential. Dialing it up to 1999, it wasn't quite as presidential. And then moving forward to today, everything is about Hillary Clinton, if you search for Clinton, when you look at the popularity of the things on the web. So it was a really interesting project, and I was doing it as a volunteer. And four days after it launched, various companies wanted to talk about acquiring it. So I go from volunteer, stay-at-home mom, to the world of corp dev. So that was really fun. So that's how I wound up at Google. And at Google, I really wanted to unlock some global and local information. One of our examples was about these small towns in Germany with their local Oktoberfests and how that information was really important to them. And we weren't quite big enough to get everything to everyone. And so at Google, we evolve our system so just when one system launches, we would start a new one. So when one such system just launched, I came up with the idea for TeraGoogle. And speaking of thinking big, probably 95% of people said, why would we ever need a search system for a trillion pages? And of course, the founders were like, is that all? So at the time, it was very cost efficient. When it launched, it was about 50 times cheaper than our main line index. And at the same time, some of my web ranking team went off, and they shipped one of our early deep-learning projects, which was around ads broadening, like once you bit on one keyword, what other keywords do you-- might be interesting. So after having my fourth child, I did what everybody does when they have four kids, and they're on maternity leave. I decided to start a company. So I started a company-- Cuil. And we really wanted to bring like a really high-res, rich experience to people with high bandwidth, and a really low bandwidth experience to people who had feature phones. And so what we did around the low bandwidth experience is we summarized things so that a fact that appeared on tons of web pages was on top in kind of paragraph format with a bunch of references down to much rarer facts that had much fewer references. So after Cuil, I returned to Google where I became one of the first 100 engineers in Android, because this time, I was pretty sure the mobile revolution was going to happen. But it wasn't a sure thing. I mean, at that time, Android had sold 40 million phones. And of course, today it was announced that Android now has 900 million. I was part of the team that helped launch Google Play on Play tablets. And very recently, like two weeks ago, we shipped a native search architecture onto the phone to index all the local data so that when you're offline, you can have access to it, and some data never leaves your phone. So a lot of people say, wait, Google's so good at search. Why didn't you just compile the thing and put it on the phone? Well, the phone is kind of like my early Commodore Pet computer, not quite as bad, but compared to a data center, it is. And each of the programs can only have 20 meg resident in memory and the executable is 500 meg as hard constraints. So trust me when I say that Google search system, its executable is bigger than 500 meg. So when we look at this system, because of the success of Android as a compute cluster, it's actually over 500 exahertz of computing power, which is way more powerful than the human brain. So last autumn, I moved over to our new AI area, where recently Ray Kurzweil joined us. And he says that the AIs that humans create are going to leave the Earth at light speed and populate the universe. So we're not just working on the most important project on planet Earth, but in the universe. So I do want to encourage the Techmakers out there, women and men, to really reach for the stars and use technology to help shape the world. Technology is a very powerful platform in order to help people and enable change. Thank you. [APPLAUSE]