Gebru-Gibberish Gives Google Gastroenteritis
February 24, 2021
At the outset, I want to address Google’s Gebru-gibberish
Definition: Gebru-gibberish refers to official statements from Alphabet Google about the personnel issues related to the departure of two female experts in artificial intelligence working on the ethics of smart software. Gebru-gibberish is similar to statements made by those in fear of their survival.
Gastroenteritis: Watch the ads on Fox News or CNN for video explanations: Adult diapers, incontinence, etc.
Psychological impact: Fear, paranoia, flight reaction, irrational aggressiveness. Feelings of embarrassment, failure, serious injury, and lots of time in the WC.
The details of the viral problem causing discomfort among the world’s most elite online advertising organization relates to the management of Dr. Timnit Gebru. To add to the need to keep certain facilities nearby, the estimable Alphabet Google outfit apparently dismissed Dr. Margaret Mitchell. The output from the world’s most sophisticated ad sales company was Gebru-gibberish. Now those words have characterized the shallowness of the Alphabet Google thing’s approach to smart software.
In order to appreciate the problem, take a look at “Underspecification Presents Challenges for Credibility in Modern Machine Learning.” Here’s the author listing and affiliation for the people who contributed to the paper available without cost on ArXiv.org:
The image is hard to read. Let me point out that the authors include more than 30 Googlers (who may become Xooglers in between dashes to the WC).
The paper is referenced in a chatty Medium write up called “Is Google’s AI Research about to Implode?” The essay raises an interesting possibility. The write up contains an interesting point, one that suggests that Google’s smart software may have some limitations:
Underspecification presents significant challenges for the credibility of modern machine learning.
Why the apparently illogical behavior with regard to Drs. Gebru and Mitchell?
My view is that the Gebru-gibberish released from Googzilla is directly correlated with the accuracy of the information presented in the “underspecification” paper. Sure, the method works in some cases, just as the 1998 Autonomy black box worked in some cases. However, to keep the accuracy high, significant time and effort must be invested. Otherwise, smart software evidences the charming characteristic of “drift”; that is, what was relevant before new content was processed is perceived as irrelevant or just incorrect in subsequent interactions.
What does this mean?
Small, narrow domains work okay. Larger content domains work less okay.
Heron Systems, using a variation of the Google DeepMind approach, was able to “kill” a human in a simulated dog flight. However, the domain was small and there were some “rules.” Perfect for smart software. The human top gun was dead fast. Larger domains like dealing with swarms of thousands of militarized and hardened unmanned aerial vehicles and a simultaneous series of targeted cyber attacks using sleeper software favored by some nation states means that smart software will be ineffective.
What will Google do?
As I have pointed out in previous blog posts, the high school science club management method employed by Backrub has become the standard operating procedure at today’s Alphabet Google.
Thus, the question, “Is Google’s AI research about to implode?” is a good one. The answer is, “No.” Google has money; it has staff who tow the line; and it has its charade of an honest, fair, and smart online advertising system.
Let me suggest a slight change to the question; to wit: “Is Google at a tipping point?” The answer to this question is, “Yes.”
Gibru-gibberish is similar to the information and other outputs of Icarus, who flew too close to the sun and flamed out in a memorable way.
Stephen E Arnold, February 24, 2021
Comments
-
- Subscribe to Beyond Search
Feature archive
News archive
-
