Keeping Google AI in Touch with Reality
May 25, 2008
A remarkable post by Anand Rajaraman in Datawocky deserves your attention–close attention at that. You can read the article here. I anticipate that Techmeme will follow the discussion thread throughout the day. The Techmeme link is here.
The core argument in Mr. Rajaraman’s posting pivots on a question, “Are Machine Learning Models Prone to Catastrophic Failure?” To answer this question, Mr. Rajaraman spoke with Dr. Peter Norvig, a Google wizard and machine intelligence expert.
Google uses humans to ride herd on some algorithms. Mr. Rajaraman does a very good job of explaining that some processes just hum along; others require Googley monitoring. The answer to the question, then, is, “It depend”.
The most interesting part of this posting is what’s unsaid. For example, Mr. Rajaraman is an influential figure in the Silicon Valley world. He has had personal and business relationships with a number of Google professionals, including some of Google’s lowest-profile, highest IQ individuals. In April 2008, Mr Rajaraman posted an equally important article about Google. This essay “The Story behind Google’s Crawler Upgrade” struck me as a genuine scoop. You can read it here.
Mr Rajaraman, who is involved with some Stanford University activities, is the next best thing to having lunch in the Google cafeteria. In fact, his posts are more insightful than the PR generated by Google itself or the “run the game plan” presentation by worker bee Googlers.
Google’s reliance of different types of computational intelligence is itself useful information. The fact that Google wizards monitor certain functions and make adjustments underscores to what expensive actions Google uses to keep its system in tune. Google could let its “borg” run unattended. That’s not happening, and Google is willing to invest to ensure that its software machines–one of which is cleverly named the janitor–working the way it is supposed to operate.
Google’s willingness to invest in “fancy math” is equalled by its willingness to invest in Google engineers to keep the race car in tune. The approach stands in sharp contrast to companies who systems are built via acquisition so engineers can only use these, not adjust them in meaningful ways. Other companies are content to run certain functions as “black boxes” because the funding is not provided to get under the hood.
A couple of other points about Mr. Rajaraman:
- He is on the computer science faculty at Stanford, an institution with many ties to Google
- He founded the venture capital firm Cambrian Ventures
- He helped start Kosmix and Jungless where some Googlers worked. Kosmix is a next-generation search engine. Junglee, acquired by Amazon, was a data management system with applicability to online retail.
- He’s a graduate of the Indian Institute of Technology and has a Ph.D. from Stanford
- According to the Cluuz.com relationship map, he has a tie to Aster Data which I wrote about here. I will leave it to you to explore the other Google relationships that Mr. Rajaraman appears to have.
If you want to keep your finger on the pulse of Google, Datawocky is a must read. Sometimes I feel, without any hard evidence, that Google uses Mr. Rajaraman to put certain facts about the world’s largest application platform in circulation.
Stephen Arnold, May 25, 2008
Comments
One Response to “Keeping Google AI in Touch with Reality”
-
- Subscribe to Beyond Search
Feature archive
News archive
-
[…] scoop about Google’s reluctance to rely exclusively on autonomous software. My post is here. The Datawocky piece is here. (I’ve heard that some Googlers call the Google infrastructure […]