Liveblog: SMX West 2010 Keynote – An Insider’s Look At Google Research
It’s day two of SMX West 2010 in a rainy Santa Clara, California, and we’ll be liveblogging today’s keynote session. Search Engine Land’s Danny Sullivan and Chris Sherman will be chatting on the big stage with Peter Norvig, Google’s Director of Research. We’re expecting some sneak peeks at future Google products and technology, so come on back at about 9:00 am PT for more.
Before we start, one note: Google’s Peter Norvig has an Apple laptop with this sticker: “My other computer is a data center.” Brilliant!
9:03 am – Looks like we’re running a little bit late. We’re going to announce the winner of Marin Software’s “Biggest Search Geek” contest before the keynote starts at 9:15 am.
9:10 am – Okay, we’re announcing the contest winner now. And his name is Tim Ossmo. Congrats, Tim.
9:15 am – And now we’re underway with introductions and such. Chris Sherman is telling us that Peter Norvig holds the world record for creating a 17,826-word long palindrome. Whoa.
Peter is going to show 21 projects from Google Research over the next 15 minutes, so I’ll do my best and hope the fingers don’t fall off.
Person Finder: the app Google created to help Chilean earthquake victims locate friends and family.
Power Meter: device you can plug into your house to monitor how much electricity you use each day.
Earth Engine: a maps-like tool that shows changes to earth (like deforestation) over a period of time.
Trike and Street View: the Google Maps projects we all know and love.
User Photos in Street View: just announced this a few days ago.
Image Swirl: a tool to show similar images that’s available in Google Image Search.
Web-Scale Image Annotation: tool that matches images to the queries that trigger those images.
Image Rotation Captcha: rather than typing in words, user has to rotate an image back to its “straight” position.
Goggles: the phone app that lets you search by taking photos.
Discontinuous Video Scene-Carving: lets you squish video down, like the cropping tool you’d find in Photoshop and similar image tools.
Sharing Cluster Data: data lists that Google likes to share with the education community.
App Inventor for Android: an introductory program environment that Google uses to teach people how to program for Android OS. Uses a visual programming language.
Speech Recognition: another phone/mobile app.
Punctuation/Capitalization in Transcribed Speech: puts the punctuation in Google Voice transcripts, for example.
Translating Phone: talking about the possibility, they have the two pieces needed to put this together.
Low Resource MT – Yiddish: translation tool for difficult languages.
Sounds Understanding: this is in development. “I want something that sounds like a car crash,” and Google will return those results.
Google Squared: the database search display tool that Google announced last year.
Clustering: tool that determines clusters of related words based on concepts.
Attribute Extraction: Goes beyond concepts to recognize related attributes for terms, i.e., “fuel” is related to car terms.
Browser Size: the tool that Google released last year to put an overlay on any web page that shows what portion of users can see your web page.
Wow. So that was 21 tools in about 10 minutes. We’re gonna do some Q&A now.
CS: How does Google approach the research process, esp. with the 20% time we’ve all heard of.
PN: Anyone can start a research project as part of their 20% time. They recruit co-workers. Projects get reviewed to decide which ones to pursue more seriously with staffing and resources. They review actual demos to decide this.
CS: asks how they decide which projects to pursue.
PN: We’re focused on doing things that are useful.
Danny: What are some of the biggest things that have come out of the 20% time?
PN: Depends who you ask. Gmail is one, but the creator (Paul Buchheit) says that wasn’t really from 20% time because it quickly became his main project.
CS: How involved are Larry and Sergey anymore?
PN: They’re very involved. They see themselves having two roles. One, setting the long-term plan for the company. But two, also getting hands-on to make sure projects are proceeding and have what they need.
DS: Do they have 20% projects?
PN: No, their jobs are 100% time. Actually, they do have some stuff they work on from home.
DS: What about you?
PN: I’m looking into things related to education search, which is different from other types of search.
DS: What technology things do people talk about a lot and overhype that you think are way off-base?
PN: I don’t know. There’s a lot of emphasis on mobile, and I think that’s appropriate. “I’m pretty happy with the way things are.”
CS: Asks about social search, and how Google determines relevant signals without a huge link graph.
PN: Let me rephrase my answer to Danny. One thing I do think is overhyped is PageRank. It’s important, but it’s just one of many things. We never felt that PageRank was such a big factor because we’ve always looked at all the available data. Paraphrasing: He says the lack of PageRank and legacy links is not a big deal when dealing with social search.
He’s talking about how Google switched from the monthly dance to doing daily updates, and then to hourly updates. “Hourly isn’t good enough,” Larry said. “Now we’ve realized his vision.”
DS: Is it time for Google to come up with a better name than PageRank to stop people from over-focusing on it.
PN: I do think we need some better branding.
CS: Asks about the Caffeine infrastructure update.
PN: It’s in one datacanter now, and we’ll be rolling out to the other datacenters soon, but I don’t have a specific timeframe on that.
DS: Asks about signals and ranking. Are there signals people don’t realize are important. Mentions citations in local search. Are there other surprises that have worked well outside of links?
PN: Yes, we’ve sort of manufactured links with that – mentions of a business can be a signal.
DS: Asks about moving beyond textual matching to the matching of concepts and objects.
PN: The clustering tool I showed is one example of that, and Google Squared is another. To get the right answers we need to do more than basic textual matching.
CS: What kind of problems will take a while to solve?
PN: Mentions visual and voice recognition. Still images and video images are still challenging. There’s so much more data in a video than a text file. It’s also messier to understand what all is going on in the video.
DS: Asks about solving email overload.
PN: Says he had an intern working on that last summer, and they do have some tools they’ll be rolling out to help with that. But also, “is email always the right tool?” Sometimes starting over is a good first step. Says he still uses email more than Google Wave.
CS: Asks about his teams – do they use Wave?
PN: Yes, they’re trying to figure out what it can do for them. “I think we’re at a confusing point.” Internally, do employees use Wave, email, Google Docs, something else for collaboration?
DS: Goes back to video. “Who’s winning – the new information flowing in, or our ability to keep up with it.”
PN: We’ll keep getting better at keeping up with things. We need tools to help us figure out what’s useful information and what’s not.
CS: Sergey has mentioned having an embedded chip in your head to search Google. Anyone on your team working on that?
PN: (chuckles) No, we’re not doing that.
DS: Asks about the idea of a training center to help people learn what Google needs its new employees to know.
PN: Says there is some issue with people coming from academia and not being fully prepared for what they need to do in company setting.
CS: Asks about how Google trains people.
PN: We have an internal course that everyone takes, and then we put them on a starter project.
CS: Asks if Google moves employees around like MSFT does.
PN: We do. We try to keep projects short so people can stay fresh. We also move them around a bit. We have a common infrastructure, so it’s easier to move from one place to another — you use all the same tools.
DS: What do you see as the next big thing in search?
PN: I think you see the page becoming more interesting and varied. Not just 10 blue links. A result page may look more like a newspaper page than a list of links. We’ll have to do a better job adapting to new ways of searching — like mobile. Talks about how the screen is so much smaller, so the right result needs to be at the top.
And with that we’re wrapping up. Thanks for tuning in.