DATA - Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification by Joy Buolamwini and Timnit Gebru

Bias in facial recognition is topic slowly coming to the forefront of mainstream media arguably as a result of the growing awaking to data and how data is collected, used and distributed. Data is a word that can be compared to ‘Algorithm’ in the context of Gillespie’s essay called Algorithm. Gillespie expresses her thoughts on the term algorithm meaning different things to the various users of the word. No one use or meaning is more dominate, but awareness of the difference and similarities of the word is needed. Similar to this, Data is a word people hear and are familiar with on the surface but know of no concrete meanings. When a word’s meaning is unknown and speculated, policing becomes difficult. Data is what my group will explore further. How we curate online presences verse our real life presence and the concept of deleting yourself from the digital sphere. Our starting point was to read into the data protection laws for our respective countries. I also read the terms and conditions of poplar social networking sites like facebook and instagram. Reading through the long web pages I was shocked at the extent to which data we do not physically give or input is collected. For example, people are aware that you sign up to Facebook using details like your name and email. This is commonly accepted; however, the data collection does not stop there. By signing up and agreeing to the terms and conditions you allow Facebook to access a long list of things that include: 

1. The location of a photo or the date a file was created 
2. What is seen through features they provide, like the camera. 
3. Facial recognition software to analyze photos and videos they think you're in. This is the default setting and needs to be manually turned off. 
4. They can access information about your device, websites you visit, purchases you make, the ads you see, and how you use their services whether or not you have a Facebook account or are logged into Facebook. 

I wonder if widespread knowledge of terms and condition will have an affect on the popularity.

4. 
   

Joy Buolamwini is a computer scientist and digital activist who brought the bias in facial recognition to the forefront of many large institutions. She has been instrumental in the progression and diversity of a new stanard of training datasets. For her university studies she wrote a practice-based thesis titled ‘Intersectional Accuracy Disparities in Commercial Gender Classification’ that focused on machine learning algorithms not being able to accurately categorise women and men of darker skins tones.  She begins by stating ‘who is hired, fired, granted a loan, or how long an individual spends in prison, decisions that have traditionally been performed by humans are rapidly made by algorithms (O’Neil, 2017; Citron and Pasquale, 2014)’.  This immediately positions the text as significant as there is a possible threat to one's quality of life. She emphasises the point further expounding that ‘A year- long research investigation across 100 police departments revealed that African-American individuals are more likely to be stopped by law enforcement and be subjected to face recognition searches than individuals of other ethnicities (Garvie et al., 2016). False positives and un- warranted searches pose a threat to civil liberties.’ Errors can be detrimental and for this reason improvements must be made. 

Buolamwini  created a new face data set of 1270 individuals ‘annotated … with the Fitzpatrick skin classification system’ and 4 subgroups: darker females, darker males, lighter females and lighter males. Skin colour is a more precise visual in comparison to rather than race or ethnicity. Even further her ‘work introduces the first intersectional demographic and phenotypic evaluation of face-based gender classification accuracy’. She chose members of parliament from Rwanda, Senegal, South Africa, Iceland, Finland and Sweden. Countries where you will typically find the lightest or darkest individuals as well as a high population of women in parliament. The results of her study results revealed; one, all classifiers performed best for lighter individuals and males. Two, the classifiers performed worst for darker females. She concludes that further work should focus on ‘increasing phenotypic and demographic representation in face datasets and algorithmic evaluation’. 

People have studied ways to create fairer algorithms in many digital fields however the effects and ‘implications’ of false recognitions call for more relevant action to be done. ‘While the NIST gender report explored the impact of ethnicity on gender classification through the use of an ethnic proxy (country of origin), none of the 10 locations used in the study were in Africa or the Caribbean where there are significant Black populations’. Buolamwini implies diversity and representation is key at all levels. It can be argued that the oppressed will do a better job at speaking to the struggle they live than the oppressor speaking to the struggles of the oppressed. She goes on to say ‘Previous studies have shown that face recognition systems developed in Western nations and those developed in Asian nations tend to perform better on their respective populations (Phillips et al., 2011)’.  It is not on everyone's agenda to do a thorough job. It can also be argued that Joy, as a Ghanaian woman living in the diaspora, has different motivations for finding thoughtful and creative solutions to issue she is personally affected by or an advocate of.