Discussion

• @twittereng @twittereng on x

  In our latest blog post, we're sharing the findings from our image cropping algorithm analysis and exploring ways to create a more equitable experience on Twitter. https://blog.twitter.com/...

• @carnage4life Dare Obasanjo on x

  Twitter's image cropping algorithm was designed to favor what most people would look at first. Turns out it favored women & white people so they got rid of it. Great example of how even the definition of an algorithm can be biased given human behavior. https://t.co/rNrG258CNF

• @twittereng @twittereng on x

  As part of our commitment to transparency, we've also published our analysis on ArXiv and are sharing our source code so you can reproduce and better our analysis. Paper: https://arxiv.org/... Code: https://github.com/...

• @twittersafety @twittersafety on x

  We heard the feedback—we didn't get image cropping right the first time. So our ML Ethics, Transparency, and Accountability (META) team dove deep and applied what they learned. We're sharing these analysis results publicly today. https://twitter.com/...

• @lizkelley Liz Kelley on x

  sometimes the best algorithm is no algorithm. thanks to everyone who shared feedback - you made us better. https://twitter.com/...

• @thomasforth Tom Forth on x

  via many people, thank you all. Twitter have written up their work on biases in their cropping algorithm, what they're going to do, what code they used to analyse the problem, and lots more. It feels like an example of how to do this stuff right. https://blog.twitter.com/...

• @reacocard @reacocard on x

  “One of our conclusions is that not everything on Twitter is a good candidate for an algorithm, and in this case, how to crop an image is a decision best made by people.” https://blog.twitter.com/... https://twitter.com/...

• @mathowie Matt Haughey on x

  (small hill I will die on forever: I still hate the word “learnings” just use “lessons” people!) https://blog.twitter.com/...

• @xor Parker Higgins on x

  “One of our conclusions is that not everything on Twitter is a good candidate for an algorithm” <- this is a remarkable sentence and it shouldn't be https://blog.twitter.com/...

• @tylerglaiel Tyler Glaiel on x

  omg just let us choose how our images get cropped like what happens when you upload a profile picture or header that isn't the right size. just put a little button for it next to the image before you post https://blog.twitter.com/...

• @jackclarksf Jack Clark on x

  Really excellent post from @Twitter about how it investigated potential for biases in its deployed auto-cropping algorithms, including a discussion of some results (it has some slight biases) and some actions Twitter is taking. A+ for acknowledging problem https://blog.twitter.co…

• @kharijohnson Khari Johnson on x

  An analysis by Twitter out today finds that Twitter's automatic image cropping algorithm is more likely to choose to display the photos of women over men and white faces over Black faces https://www.wired.com/...

• @rahaeli @rahaeli on x

  Remember how I said Twitter couldn't figure out how to make its image cropping algorithm less racist so they just got rid of cropping and some of y'all insisted there was an innocent explanation? Twitter: “We couldn't make our algorithm less racist, so we got rid of cropping.” ht…

• @trentonjkennedy Trenton Kennedy on x

  “In one of the first instances of total algorithmic transparency from a social platform, Twitter has removed its image cropping algorithm and created new controls for users...” https://www.protocol.com/...

• @dlowd Daniel Lowd on x

  “... This release reduces our dependency on ML for a function that we agree is best performed by people using our products. ...” I love this — sometimes the best ML is less ML! https://twitter.com/...

• @tweetinjules Julie Inman Grant on x

  Great to see this commitment from @TwitterSafety in providing useful evidence of efficacy & impact of their safety features. Also commend them for incorporating user input. These are hallmarks of #SafetybyDesign & @eSafetyOffice will continue to surface up good safety practice! h…

• @laughsovich Tomo Lazovich on x

  If you were wondering about why Twitter ditched image cropping, here's one of the reasons. I'm so proud of my teammates on META who did this research! It's an important reminder that sometimes the ‘solution’ to a problem of ML bias is “don't use the algorithm”. https://twitter.co…

• @ruchowdh Rumman Chowdhury on x

  I promised we'd share our homework on our image cropping decisions - here it is. We did a bias analysis of our saliency algorithm - read about what we found, dig into our code, and join us for a spaces event this afternoon. https://twitter.com/...

• @_brohrer_ Brandon Rohrer on x

  “One of our conclusions is that not everything on Twitter is a good candidate for an algorithm” A beautiful piece of work from @ruchowdh and team. https://twitter.com/...