We birth you, you birthed me
By Thora Thøgersen
Eyebrows, yes. Big nose, yes. Short hair, yes. Result 65% man. Rosy cheeks, yes. Heavy makeup, yes. Slim chin, yes. It must be 100% woman.
These are the results of an Artificial Intelligence when looking to categorise gender in images — stereotypes that human eyes might not judge that swiftly. Soon data will not merely be aggregated from user searches and clicks online, but from our biometric data. Your face features and expression, your movement, the clothes you wear, and what you engage with on your way together with online tags, words, and names are puzzled together by an AI to paint a picture of who you are.
In We birth you, you birthed me, Thora Thøgersen unpacks multiple years of curated digital data identifications from social media as well as her personal image archives spanning this period to map out — with human eyes and hands — the multitudes of her personhood as an alternative to algorithmic categorisation.
When going through all my online data I couldn't recognize myself. even though I was the one that had posted all those pictures and googled those things of interest. For many years all the commercials AI's served me were odd and miss fitting. Occasionally there would be a shaving ad addressed to me because I have told Facebook that I identified as female. But then again I also got male shaving commercials. When looking at my "interest", the boxes Facebook had put me into, varied from liking 'Mondays', 'Blue skies', The city Łódź in Poland, and The Washington Post. None of them are of particular interest to me. Most common were cute animal videos and Wish commercials.
 Steyerl, Hito (2019) Digital Debris. Duty Free Art, Verso Books 100.
 Metz, Rachel (2019) AI software defines people as male or female. That's a problem. CNN Business. Available from https://edition.cnn.com/2019/11/21/tech/ai-gender-recognition-problem/index.html (accessed 24 May 2021).
 Laurenandcait (2019) Something terrible happened at our gender reveal party this weekend. Please share. Instagram (video). Available from https://www.instagram.com/p/B5IhVm-AFNE/?utm_source=ig_embed&utm_campaign=embed_video_watch_again (accessed 24 May 2021).
 Kulager, Frederik (2021) Now stores read your mood, gender, age and ethnicity. The man here equips Denmark with face scanners. Zetland, Danish Newspaper. Available from https://www.zetland.dk/historie/semE1G5L-mO9kYPgY-e4e05 (accessed 24 May 2021).
 Barrett, Lisa Feldman (2018) You aren't at the mercy of your emotions -- your brain creates them. TED Talk. Video lecture. Available from https://www.youtube.com/watch?v=0gks6ceq4eQ (accessed 24 May 2021).
 Remya, Emma and Wojcik, Stefan (2020) The challenges of using machine learning to identify gender in images. Pew Research Center, Internet & Technology. Available from https://www.pewresearch.org/internet/2019/09/05/the-challenges-of-using-machine-learning-to-identify-gender-in-images/ (accessed 24 May 2021).
 Unbabel (2019) Gender Bias In AI | Understanding with Unbabel. Youtube video. Available from https://www.youtube.com/watch?v=qpYyI9Tdtc4 (accessed 24 May 2021).
 Axios (2018) Biases are being baked into artificial intelligence. Youtube video. Available from https://www.youtube.com/watch?v=NaWJhlDb6sE (accessed 24 May 2021).
 Buolamwini, Joy (2017) How I'm fighting bias in algorithms. TED Talk. Video lecture. Available from https://www.youtube.com/watch?v=UG_X_7g63rY (accessed 24 May 2021).
 Hauser, Robin (2018) Can we protect AI from our biases? TED Talk. Video lecture. Available from https://www.youtube.com/watch?v=eV_tx4ngVT0 (accessed 24 May 2021).
 Tufekci, Zeynep (2016) Machine intelligence makes human morals more important. TED Talk. Video lecture. Available from https://www.youtube.com/watch?v=hSSmmlridUM (accessed 24 May 2021).
I collected the data from my account onFacebook, Pinterest, Instagram, Google, Tweeter and Danish Health databank. I tried to sort it into categories of use for me. Mostly I used pictures from Instagram and Facebook. Also, my "interest" on Pinterest and Facebook had relevance. I have from the beginning had things like location switched off and have over many rounds cleaned up in my old content.
How does AI categorises our gender? I have been using an simple AI on the website Betaface.com to identify gender, attractiveness and look at what features it detects:
Categorize the AI looks for:
- Pale skin
- Rosy cheeks
- Narrow eyes
- Big Nose
- Big Lips
- Mouth open
- Mouth size
- Hair color: Black, Brown, Blond, gray
- Beard, mustache
- Bushy eyebrows
- Arched eyebrows
- High cheekbones
- Double chin
- Chin size
- oval face
- expression: smile
- Heavy makeup
- Wearing lipstick
- Wearing hat
- Wearing necktie
The raw data looks like this: age: 43, arched eyebrows: yes(50%), attractive: no(70%), bags under eyes: no (69%), bald: no (76%), bangs: no (40%), beard: no (26%), big lips: no, big nose, black hair: no (50%), blond hair: no (76%), brown hair: yes (33%),blurry: yes (70%), bushy eyebrows: no (41%), chubby: no (6%), double chin: no (10%), expression: smile (26%), gender: male (33%), glasses: no, goatee: no (60%), gray hair: no (30%), heavy makeup: no (62%), high cheekbones: no (13%), mouth open: yes (38%), mustache: no (47%), narrow eyes: yes (33%), oval face: yes (33%), pale skin: no, pitch: -9.07, pointy nose: no (31%), race: asian (75%), receding hairline: no (61%), rosy cheeks: no (62%), sideburns: no (66%), straight hair: no (76%), wavy hair: yes (50%), wearing earrings: no (42%), wearing hat: no (81%), wearing lipstick: no (59%), wearing necklace: no (76%), wearing necktie: no (36%), yaw: -0.49, young: no (59%), chin size: extra small, color background: 776a6f (4%), color clothes middle: 48454e (3%), color clothes sides: f3c8ae (1%), color eyes: 774341 (99%), color hair: 684038 (69%), color mustache: 534549 (83%), color skin: c3756c, eyebrows corners: low, eyebrows position: average, eyebrows size: extra thin, eyes corners: average, eyes distance: far, eyes position: extra low, eyes shape: extra thin, glasses rim: no, hair beard: none, hair color type: brown light (69%), hair forehead: no, hair length: none, hair mustache: thick, hair sides: very thin, hair top: short, head shape: heart, head width: average, mouth corners: extra raised, mouth height: thin, mouth width: small, nose shape: extra straight, nose width: wide, teeth visible: no.