Zach Vorhies

Google Leaks: Machine Learning and Diversity Summit

This is a slidedeck found on Google's servers discussing machine learning and bias. This deck has the famous "People are Programmed" slide. A lot of the discussion involves the the "Implicit Association Test" for racial bias, which was discontinued at Google around 2017 due to the criticism of the testing design emerged from the Wall Street Journal.

Part of the full Google Disclosure, which you can see here.
PDF link.

What this document shows

This Google slide deck on machine learning and bias was found on Google's servers. It contains the well-known "People are Programmed" slide, and much of it discusses the Implicit Association Test for racial bias, which Google discontinued around 2017 after criticism of the test's design appeared in the Wall Street Journal.

Key quotes from the document

Quoted verbatim from the "RMI Diversity Summit" slide deck:

  • Portrayal matters. In CS, representation in mainstream media mattered. Nowadays Google = mainstream media.
  • Training data are harvested. Algorithms are programmed. Media classified, filtered, aggregated or generated. People are programmed.
  • So many unintended consequences! The system can be evil even when no single part is…
  • Wondering about your own programming? You can interrogate it… I.e.: If you are a good and decent person, you probably won't like what you learn.
  • Pretty much any product using crowdsourcing, distribution estimation, clustering, ranking, collaborative filtering, sequence synthesis, or any other kind of MI needs to think through its latent variables.

Sources and further reading

About the Google Leaks disclosure

This page is part of the Google Leaks: 950+ pages of internal Google documents that Zach Vorhies, a Senior Software Engineer at Google and YouTube for 8.5 years, took with him when he resigned from Google in June 2019. He delivered them to the antitrust division of the U.S. Department of Justice in August 2019 and released them to the public through Project Veritas. The documents describe, in Google's own words, how Google defined "fake news", built its "Machine Learning Fairness" system and kept blacklists across Google Search, YouTube and Google News. He tells the full story in his book Google Leaks: An Expose of Big Tech Censorship (Skyhorse Publishing).

Read the full archive on the Google Leaks page, the dated sequence of events in the timeline, and who Zach Vorhies is on the about page.

Document pages

Machine Learning and Diversity Summit, page 1 Machine Learning and Diversity Summit, page 2 Machine Learning and Diversity Summit, page 3 Machine Learning and Diversity Summit, page 4 Machine Learning and Diversity Summit, page 5 Machine Learning and Diversity Summit, page 6 Machine Learning and Diversity Summit, page 7 Machine Learning and Diversity Summit, page 8 Machine Learning and Diversity Summit, page 9 Machine Learning and Diversity Summit, page 10 Machine Learning and Diversity Summit, page 11 Machine Learning and Diversity Summit, page 12 Machine Learning and Diversity Summit, page 13 Machine Learning and Diversity Summit, page 14 Machine Learning and Diversity Summit, page 15 Machine Learning and Diversity Summit, page 16 Machine Learning and Diversity Summit, page 17 Machine Learning and Diversity Summit, page 18 Machine Learning and Diversity Summit, page 19 Machine Learning and Diversity Summit, page 20 Machine Learning and Diversity Summit, page 21 Machine Learning and Diversity Summit, page 22 Machine Learning and Diversity Summit, page 23 Machine Learning and Diversity Summit, page 24 Machine Learning and Diversity Summit, page 25 Machine Learning and Diversity Summit, page 26 Machine Learning and Diversity Summit, page 27 Machine Learning and Diversity Summit, page 28 Machine Learning and Diversity Summit, page 29 Machine Learning and Diversity Summit, page 30 Machine Learning and Diversity Summit, page 31 Machine Learning and Diversity Summit, page 32 Machine Learning and Diversity Summit, page 33 Machine Learning and Diversity Summit, page 34 Machine Learning and Diversity Summit, page 35 Machine Learning and Diversity Summit, page 36 Machine Learning and Diversity Summit, page 37 Machine Learning and Diversity Summit, page 38 Machine Learning and Diversity Summit, page 39 Machine Learning and Diversity Summit, page 40 Machine Learning and Diversity Summit, page 41 Machine Learning and Diversity Summit, page 42 Machine Learning and Diversity Summit, page 43 Machine Learning and Diversity Summit, page 44 Machine Learning and Diversity Summit, page 45 Machine Learning and Diversity Summit, page 46 Machine Learning and Diversity Summit, page 47 Machine Learning and Diversity Summit, page 48 Machine Learning and Diversity Summit, page 49 Machine Learning and Diversity Summit, page 50 Machine Learning and Diversity Summit, page 51 Machine Learning and Diversity Summit, page 52 Machine Learning and Diversity Summit, page 53 Machine Learning and Diversity Summit, page 54 Machine Learning and Diversity Summit, page 55 Machine Learning and Diversity Summit, page 56 Machine Learning and Diversity Summit, page 57 Machine Learning and Diversity Summit, page 58 Machine Learning and Diversity Summit, page 59 Machine Learning and Diversity Summit, page 60 Machine Learning and Diversity Summit, page 61 Machine Learning and Diversity Summit, page 62 Machine Learning and Diversity Summit, page 63 Machine Learning and Diversity Summit, page 64 Machine Learning and Diversity Summit, page 65 Machine Learning and Diversity Summit, page 66 Machine Learning and Diversity Summit, page 67 Machine Learning and Diversity Summit, page 68 Machine Learning and Diversity Summit, page 69 Machine Learning and Diversity Summit, page 70 Machine Learning and Diversity Summit, page 71 Machine Learning and Diversity Summit, page 72 Machine Learning and Diversity Summit, page 73 Machine Learning and Diversity Summit, page 74 Machine Learning and Diversity Summit, page 75 Machine Learning and Diversity Summit, page 76