Hey, hey, morning people! If you’re back with us from yesterday, welcome back! Or, if you just found us, we’re glad you’re here! It’s Day 2 at SMX West and I’m seated up in the front row getting ready for the keynote to begin. The awesome Matt McGee is seated to my left with his bottle of water. I’m thinking about snatching it. I couldn’t find the water this morning and had to grab a coffee instead. My body runs on your water. So does yours. [Apparently, the water was next to the hot breakfast options. Who knew conferences had hot breakfast options? Ah, SMX, you're so sexy.]
This morning’s keynote features Dr. Peter. Norvig who will offer some insight and demo’s into the world of Google Research. I’m not sure how many of these demos I’ll be able to blog for ya, but I’ll sure as hell do my best. Heck? Should I have said ‘heck’?
We’re getting a bio on Peter. He’s a rocket scientist. A humorist. And has the world record for the longest palindrome. Who even know you could snatch a world record that way? Peter’s also a ballet dancer, which I’m sure he appreciates Chris Sherman bringing up here.
- Person Finder: App engine application. Took them an hour to get it out. They built it to help find people in Chile after the earthquake.
- Power Meter: You can monitor how much power you’re using each day. Can help cut consumption and save money.
- Earth Engine: Works over Earth data. Shows things like areas of deforestation. It’s an overlay for Google Earth.
- Trike and Snowmobile: They have a tricycle and a snowmobile that provide imagery for Google Maps
- User photos in Streetview: Let users upload and geo-locate their own photos so they come up in Maps. Makes it more useful, interactive.
- Image Swirl: Image recognition software that lets you find related images.
- Web-Scale Image Annotation: Takes a set of images and a set of the queries that triggers them and matches them up. See, this why Google employs rocket scientists. And why I could never work there.
- Image Rotation Captcha: Instead of having to type in a word, they show you a rotated picture and you have to set it right side up. Which is AWESOME cause I can never type captchas. I’m a robot.
- Goggles: You take a picture and Google tells you what it is.
- Discontinuous Video Scene Carving
- Sharing Cluster Data: Creates lists of data.
- App Inventor for Android: Teaches people how to program on Android phones.
- Speech recognition
- Punctuation/Capitalization in Transcribed Speech
- Translating phone: Translate text, pages and documents from your phone.
- Low resource MT: Yiddish
- Sound Understanding: In development. You should be able to go to YouTube and look for things that sound like a rooster or a car backing up. [This is awesome!]
- Google Squared
- Attribution Extraction
- Browser Size: Optimize sites based on their browser size. Puts an overlap over your Web page to show you.
They try to observe by watching
How do you approach research projects?
They have this idea of 20 percent time. They have a market place around that where he can start a project and get excited about it. And then he can recruit buddies to help him or he can put his idea on the shelf and help someone else with their project. They then can become official projects. They make it easy for people to make experiments. When they evaluate ideas, they’re not looking at PowerPoint slides, they’re looking at demos of what the product will actually do. The demo often runs at full-scale.
Many of what you showed us have practical application, how do you decide what to do?
They’re pushing very hard towards doing something useful. The good theoretical things will be driven by that.
What are the biggest things to ever come out of the 20 percent projects?
Depends on who you ask? Because it’s so informal, different people have different opinions. One story is that someone did Gmail and AdSense because he was frustrated that it was easier to search the Web than it was to search email. At the time, not many people were interested in that. They seem to have come around.
Speech recognition was built the same way.
He’s currently looking at Education search. That’s his 20 percent time project. It’s for people who want to search like they’re in a class. It’s about finding answers through the course of a semester.
How much are Larry and Sergey involved in what’s going on?
They’re very involved. They see themselves as having two goals:
- Setting the long term direction
- Evaluating as many projects as they can.
They’re very hands on.
What’s the most over hyped project?
He thinks they’re doing a good job focusing on what matters. There’s a lot of attention on mobile and he thinks that’s appropriate. He’s pretty happy with the way things are.
Google has moved very quickly to incorporate real-time search. You don’t have the same signals as with traditional search, how did you figure out what the new signals were so that you could incorporate it quickly without hurting the legacy database?
He revises his previous answers and says that PageRank is the most over hyped thing he’s seen. Hee. It’s not as big of a factor as people pretend it is. They look at all the available data. He doesn’t think real-time and social search works that much differently than regular search. They’re looking at everything they can get their hands on.
We heard Caffeine would be in place by now, but it’s not? Why? Are you working on that?
We have it in one data center and we’re testing it. It’s doing very well and we’ll be rolling it out soon. Programming stuff isn’t done in Research. It could be, but historically those people are on the engineering team, not the research team.
What are used as ranking signals that people may not realize are signals?
He talks about citations in Local Search and how Google can ‘manufacture’ links where there aren’t any. Books don’t have links but they have bibliographies. Google creates links that way. They’re trying to match things based on synonyms, even if you didn’t type in the same words.
What are the hard problems today that will take a hard time to crack?
Vision for still images and video images. It’s challenging in terms of the computational issues – there’s just so much more data. It’s also messier trying to parse a picture up into objects and understand what’s doing what. They don’t come with words associated. Words are very nice. Objects are harder.
Any solutions for email overload?
They have some tools and prioritization stuff they’ll be rolling out. Another thing is asking whether or not email is really the right tools. He thinks email is bad because he’s on all these legacy email lists that he doesn’t need to be on. Maybe just slashing all that down and starting over will help. Sometimes starting over on things like Wave or Twitter is a good first step.
His teams are starting to use Wave more. He still uses email. He thinks they’re at a confusing point because there are so many tools now.
Anyone working on an embedded chip you could put in your head to search Google.?
He says no. But we all know he’s totally lying. Expect it next year. In beta.
How do you grow up to be a search engineer? Where do people go?
He has good ties with the academic community. In other industries you can get trained in school and step in pretty quickly. There are some new books out that really move us into the world of Web search. That may get better. We’re making this idea of cloud computing and distributed computing more popular.
What’s next in search?
It’s not just a list of ten links anymore. There are pictures and video and structured content. A results page will look more like a newspaper page. Mobile is driving things hard because its such a small screen.