You can add location information to your Tweets, such as your city or precise location, from the web and via third-party applications. You always have the option to delete your Tweet location history. Learn more
What a stark example of algorithmic bias. Use Google Translate, type “he is beautiful” or “she is an engineer”, translate into a gender-neutral language like Hungarian or Farsi, and then translate back into English. It changes into: “she is beautiful” and “he is an engineer”https://twitter.com/GaryMarcus/status/1375110505388417025 …
Note that Google doesn’t deliberately choose this. The algorithm is forced to guess gender, because one language has no gender and the other does, and the guess is made based on all the data Google can find on the internet. Biased data = biased algorithms
how can we change this biased behaviour generated based on the data accumulated in years?
This assumes they can’t return two weighted answers. In this case there is no reason they can’t specify an optional he/she or she/he for the display view. For API level they need to choose but could return two full responses. Poor assumptions on API design continues the bias.
The most effective way to get an algorithm to show its flaws is to give it inputs it can’t handle. Remember when Neil Gaiman searched Amazon for “Girl Scout cookies” (which they don’t sell)? Our industry needs to do better.
And statistically speaking this bias is justified :) the issue is this is not stats class
Would a human translate it any differently? Or would they guess similarly? Are there techniques good human translators use that we should teach Google Translate to use? (Like maybe the singular they one I used above?)
Respectfully, I think bias is the wrong word. The algorithm is merely following orders: guess, based on the data you’ve seen previously. Someone didn’t teach it adequately about languages without gender. I feel like it’s slander against algorithms. :)
The line between algorithm and data is blurry, because modern machine-learning algorithms are often the exact same code just operating on different data. “Algorithmic bias” has become shorthand for “algorithm+training data bias.”
This surfaces information that’s already there, giving agency to change. Transparency is good and allows us to find better solutions. Also:https://twitter.com/marjeaksli/status/1376266523849940992 …
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.
