Google High Schools It Again
May 4, 2022
Dr. Satrajit Chatterjee may have been kicked out of the Google High School Science Club. The shame!
The alleged truth appears in “Another Firing Among Google’s AI Brain Trust and More Discord.” The venerable New York Times, owner of Wordle, states:
Less than two years after Google dismissed two researchers who criticized the biases built into artificial intelligence systems, the company has fired a researcher who questioned a paper it published on the abilities of a specialized type of artificial intelligence used in making computer chips.
Googzilla has been eager to make clear that its approach to smart software is the one best way. Sure, some may disagree — Just don’t complain too loudly or work at Google are a couple of tips. I have pointed out that the nifty approach used by the online ad company can demonstrate “drift”. I call this tendency the “close enough for horseshoes” approach. I mean that if something is good enough for ad matching, then the same system will work for other Googley things. Do you want to stand in front of a Waymo or let the Google output your health index? Sure you will. You just don’t know it yet.
Dr. Chatterjee was concerned that the information presented in the delightful, easy-to-read article “A Graph Placement Methodology for Fast Chip Design” added some flair to the write up. (Like most real science, this allegedly accurate paper is behind a paywall; however, you may be able to view it. Good luck!) Here’s the summary of the Googlers’ assertions about its smart software platform:
In this work, we propose an RL-based approach to chip floorplanning that enables domain adaptation. The RL agent becomes better and faster at floorplanning optimization as it places a greater number of chip netlists. We show that our method can generate chip floorplans that are comparable or superior to human experts in under six hours, whereas humans take months to produce acceptable floorplans for modern accelerators. Our method has been used in production to design the next generation of Google TPU.
Got that. The translation is that the Google can design chips without too many humans: Cheaper, better, faster and don’t push back with “Pick two”.
Dr. Chatterjee did and he allegedly has an opportunity to find his future elsewhere; for example, working at Dr. Timnit Gebru’s organization, a home for some Xooglers who crossed mental swords with Dr. Jeffrey Dean and his acolytes.
The New York Times, in Gray Lady fashion, asserted:
Dr. Chatterjee’s dismissal was the latest example of discord in and around Google Brain, an A.I. research group considered to be a key to the company’s future. After spending billions of dollars to hire top researchers and create new kinds of computer automation, Google has struggled with a wide variety of complaints about how it builds, uses and portrays those technologies. Tension among Google’s A.I. researchers reflects much larger struggles across the tech industry, which faces myriad questions over new A.I. technologies and the thorny social issues that have entangled these technologies and the people who build them.
Googlers can fiddle with mobiles in meetings, wear torn T shirts, and sleep on the floor under their workstation mini-desk. Just don’t disagree with the One True Way.
What if Drs. Chatterjee and Gebru are a little bit correct in their assertions that the GOOG’s smart software is dorky in many ways. What if these little dorkies add up to delivering 60 confidence in the outputs.
As I said, “close enough for horseshoes” works when burning through ad inventory. However, applied to other domains that horseshoe might land in the crowd and knock a two year old for a loop. Not good unless someone in the Google High School Science Club can address a cranial fracture.
Stephen E Arnold, May 4, 2022
Comments
-
- Subscribe to Beyond Search
Feature archive
News archive
-
