Twitter challenged its users to determine bad practices in its artificial intelligence, and it paid off.

The competition, which took place from July 30 to August 7, 2021, was intended in particular to identify the racist, homophobic or grossophobic choices of its algorithms.

Thus, the winner of the first prize notably targeted the discriminating operation of the social network's automatic photo cropping tool,

Wired

reports

.

Bogdan Kulynych, a cybersecurity student in Switzerland, used deepfake technology to create a lot of different faces.

He then found that the feature that chooses which part of the image appears in the News Feed showed a preference for white, young, and thin women.

In reality, this conclusion is only a reflection of the prejudice already existing among certain users of the social network, according to one of the jurors of the competition.

And now, for the winners!

Read on for all the details of the entries that took home prizes.

👇

- Twitter Engineering (@TwitterEng) August 9, 2021

A tool that discriminates against emojis with dark skin color

The actions and information provided by Internet users were indeed used to feed the tool's machine learning system, said Ariel Herbert-Voss, cybersecurity researcher.

“Basically, the more an image shows someone getting closer to a thin and young woman, the more it will be put forward,” summarized the specialist in artificial intelligence Patrick Hall, also a member of the competition jury.

In addition, the situations pointed out by participants in the Twitter contest demonstrated the efficiency and accuracy of the work of external contributors.

Several algorithm specialists have also considered that it could be useful in the future to let this type of people study artificial intelligence systems in order to identify undesirable functioning.

Bad practices could thus be detected and removed even before having generated harmful results or actions.

As part of this challenge, Twitter made the code for its self-cropping photo tool accessible.

Other participants have proven that his algorithm prefers the Latin alphabet over Arabic and discriminates against people with long hair, as well as emojis with dark skin color.

High-Tech

Twitch: A new update to better detect hateful and racist behavior

World

Google pays its employees differently if they decide to telecommute, a contested practice

  • High-Tech

  • Twitter

  • Discrimination

  • Artificial intelligence