Racial and Gender Biases in AI | Gunay Kazimzade | TEDxHUBerlin

Gunay devotes her TED Talk to her main research directions: Gender and racial biases, as well as inclusiveness, in AI. This talk was given at a TEDx event using the TED conference format but independently organized by a local community. Learn more at www.ted.com/tedx

[@__MGR_77__]

Hey all, the darker skin color creating more sesitivity is because of the nature of many ImageNets. The darker the the image, the higher the pixel intensities are in comparison to lighter images. Many of those algorithms are inspired by how the human or mamalian visual system would work, but they do not react to light, because darker pixels have a higher intensities in the additive color logic (this comes from the fact that the computers color system is optimized for printers). This means, the higher the intensities, the higher the activation rate is. A higher activation rate leads to more information, especially if the system already got trained on very "occupied" images (darker colors for facial features). This automatically makes those algorithms more sesitive to dark skinned people, that light skinned ones. I think this is also an issue with nor using sparsity constraints and n ot normalizing the input space. Another good work is done by Dr. Travis Monk, who actually used intensities as intrinsic parameter which allowes the network to use this information in anohter context, for example automatic normalization. (If you search for this you might find it quickly, because I don't have the link right ahead. Sorry.)

[@royal6355]

It's weird that this video has not many views.

[@zarinabbas6745]

Proud of you!

[@RM-xr8lq]

yes we were taught about training bias, especially found in occidental models, during upper division math courses unfortunately most people don't get far enough in secular education to understand what OP means, especially if they are unfortunate enough to be in place like US/UK...

[@ahmadhuseynli2073]

one thing to note: you said without pointing on gender of the profession, you get the gender identified in google translate. But "tibb bacisi" is already pointing on the gender, you cant expect a male nurse against of tibb bacisi.

[@RashidAliyev]

Go Gunay, Go! Great topic! Proud for you!

[@turalsuleymanzade9687]

Gunay proud of you!

[@MrJabrayil]

Very interesting talk!!!

[@nijataslanov5000]

Let's talk about AI

[@ВикторияГольдберг-п5з]

such a wonderful girl

[@MrJabrayil]

AI is just like society

[@azili78mikayilova83]

Success attends you!

[@yusifmammadov4407]

Good luck with that!

[@vugarsamadzada3540]

Perfect

[@ragnarlothbrok367]

Telling someone to do something he doesn't want, just because of his specific race or gender is racism/sexism, so, stop asking women to go in tech, we need good programmers and specialists, not more stacies going for easy money. Mark my words, but when you tell some random girls to just go in tech, then you will make all the talented girls suffer cause of this. I know women programmers who don't want to hire other women at all because they have lost so much time on interviewing some random girls who are in tech not by their own wish.

[@modera.torrent_]

Oof 🤦‍♂️ no, honey, that isnt an example of Google translate having bias. You are simply confusing the significance of gendered vs. non-gendered languages. The reason it is more prone to default gender neutral terms to male rather than female is simply because in many gender neutral languages, "he" is what is translated as the generalized pronoun. Similar to how the word "woman" contains the word "man", it is actually English that is the abnormal one to use he/his/man as something that only implies "male". Its also important not to confuse gender neutral languages as an indication of gender neutral culture in whatever country is in question. Simply because they do not use the pronoun or gender role title system in their language, doesn't necessarily mean that their culture doesn't recognize gender role, but rather they may just do so in a different manor than the Romance languages do. Gendered language was originally incorporated not as a method of discrimination, but rather as a method of etiquette, good manors and respect (towards both of the genders, neither moreso than the other)

[@shahlabakarova2234]

👏👏👏

[@nafghnagfh1226]

perfect