I read “Uncovering Unknown Unknowns in Machine Learning.” The title reminded me of Donald Rumsfeld who served as US Secretary of Defense fro9m 1975 to 1977. He is the author of Known and Unknown: A Memoir. He allegedly coined the quip:

There are known knowns, things we know that we…” “There are known knowns, things we know that we know; and there are known unknowns, things that we know we don’t know. But there are also unknown unknowns, things we do not know we don’t know.”

The Google blog post states:

The goal of the challenge is to raise the bar in ML evaluation sets and to find as many examples as possible that are confusing or otherwise problematic for algorithms to process.

Yep, Google wants others to help it deal with unknown unknowns. I won’t chop the logic of Mr. Rumsfeld’s alleged quip. I absolutely won’t compare knowing unknown unknowns to Google’s attempt to “solve death.”

I will make three observations:

  1. I think that the Google wants to use this type of Fancy Dan initiative to get in front of the tempest swirling around Timnit Gebru incendiary devices
  2. The desire to associate Google with something inherent in trying to make smart software do more than sell ads threads through the CATS4ML announcement. I think of the announcement as a variant of “stuff happens” and not even the Google can figure it out
  3. The initiative may be a building block in Google’s Spring 2021 game plan which will allow more ad revenue to flow whilst neutralizing difficult decisions about staff who raise uncomfortable research topics.

My three points are known unknowns. These are less troublesome than Google’s bold efforts to solve death and identify unknown unknowns. Nope, I don’t want to do the phenomenological existentialism of smart software. Not for me. Too old.

Stephen E Arnold, February 16, 2021