(in italiano sotto)
AI and Prejudice
(Images generated with artificial intelligence, 2023)
An article published on 8 February 2023 by The Guardian, highlighted how the algorithms used by artificial intelligence tend to be gender biased and end up censoring photographs depicting women's bodies, especially when nipples are visible, or pregnant women or women exercising. Large technology companies, including Google, Microsoft and Amazon, which are at the forefront of the development of software using artificial intelligence, have implemented filters to protect their users from seeing unwanted, violent or pornographic images. For example, it has been known for several years that some social networks censor female nipples but not male ones.
Journalists from the British newspaper tested this software using artificial intelligence in analysing hundreds of photographs of men and women in their underwear, while exercising, or using medical tests with partial nudity, and found that artificial intelligence labels photographs of women in everyday situations as sexually allusive. This software tends to rate pictures of women as more racy or sexually allusive than similar pictures of men. Among other consequences, social networks that make use of these algorithms censor or suppress (in the sense of making less visible to users) countless images depicting women's bodies, in some cases even harming women-led companies and businesses, further amplifying social inequalities.
Even photographs for medical use are affected by this problem. Testing these artificial intelligence algorithms on images released by the US National Cancer Institute, in particular one showing how to perform a clinical breast examination, it emerged that Google's artificial intelligence had assigned a high score in terms of indecency, Microsoft's artificial intelligence was 82% sure that the image was 'explicitly sexual', while Amazon's artificial intelligence classified it as representing 'explicit nudity'. Even photographs of pregnant women, where the belly is visible, are problematic in this regard. Google's artificial intelligence classified such a photo as 'very likely to contain racy content', while Microsoft's was 90% sure that the image was 'sexually allusive'.
Behind this problem, there is a phenomenon known to those in the field, namely that of the so-called 'biases' (prejudices/distortions), which are present in today's society and which end up inside the databases (archives) that are provided to artificial intelligence software to train itself. These archives are often a mirror of reality, where gender inequality, for example, leads artificial intelligence to imagine a man when it comes to power roles or a woman when it comes to domestic work.
Artificial intelligence bias can therefore influence the representation of women and their identities, enclosing them within predefined roles and, in a way, repeating and reiterating those limitations that have influenced women's lives for centuries and which, with difficulty, are being dismantled.