Google Smart Software: Is This an Important Statement?
July 15, 2022
I read “Human-Centered Mechanism Design with Democratic AI” continues Google’s PR campaign to win the smart software war. Like the recent interview with a DeepMind executive on the exciting Lex Fridman podcast and YouTube programs, the message is clear: The Google’s smart software is not “alive”. (Interesting PR speak begins about 1 hours and 20 minutes into the two hour plus interview. The subtext is in my opinion, “Nope, no problem. We have smart software lassoed with our systems and methods.” Okay, I think I understand framing, filtering, and messaging designed to permit online advertising to be better, faster, and maybe cheaper.
This most recent contribution uses language many previous Googley papers do not; for example, “human” and “democratic.” The article includes graphics which I must confess I found a bit difficult to figure out. Here’s an illustrative image which baffled me:
The Google and its assorted legal eagles created this image from the data mostly referenced in the cited article. Yes, Google and attendant legal eagles, you are the ultimate authorities for this image from the cited article in Nature.
Those involved with business intelligence will marvel at Google’s use of different types of visualizations to make absolutely crystal clear the researchers’ insights, findings, and data.
Great work.
I did note one passage on page nine of the Nature article:
Here is the operative language used to explain some of the democratic methods:
Defined
Objective
We wished to maximize
Estimated
Auto-differentiating
We chose
Net net: Researchers at the Google determine and then steer the system. Human-centered design meshes with the Snorkel and synthetic data methods I presume. And bias? Where there are humans, there may be bias. How human-centered were the management decisions about certain staff in the Google smart software units?
Stephen E Arnold, July 15, 2022
Comments
-
- Subscribe to Beyond Search
Feature archive
News archive
-
