This September, a research, published by Stanford psychology professor Michal Kosinski and researcher Yilun Wang, claimed to have identified the sexual orientation of people by analyzing photos of faces by means of an artificial intelligence algorithm (deep neural networks). The research has generated heated debates in the US and European media, and has also been reported in some Brazilian newspapers.
Before explaining how the research was performed, it is important to remember what is the current level of development of artificial intelligence. In a simple explanation, nowadays, what the various AI computational techniques can do is to identify patterns by analyzing a huge amount of data. That is to say, they are algorithms created to accomplish specific tasks and are still far from resembling a human intelligence.
The study analysed 35,326 images of men and women, both Caucasian, who were collected from public profiles of an American online dating website. The researchers justified that they have utilised just one ethnic group due to a lack of data representing other ethnic groups, such as Afro-Descendants.
To classify the sexual orientation of a person, the algorithm analyzes both permanent facial features (e.g. maxillary size, etc) and transient features (e.g. haircut, makeup, etc.). So the algorithm was trained with this set of images to learn what were the patterns. At the end the study claims that the artificial intelligence could accurately distinguish, only with the analysis of one photo, whether a man was gay or heterosexual, in 81% of cases, and in 74% of the cases for photos of women. The rate of success for humans analyzing the photos was 61% for photos of men, and 54% for those of women.
Following the publication, two major LGBTQ rights groups, the Human Rights Campaign (HRC) and the GLAAD (Gay & Lesbian Alliance Against Against Defamation) in the United States have heavily criticized the research. The GLAAD’s Chief Digital Office, Jim Halloran, said:
“This research isn’t science or news, but it’s a description of beauty standards on dating sites that ignores huge segments of the LGBTQ community, including people of color, transgender people, older individuals, and other LGBTQ people who don’t want to post photos on dating sites.”
And HRC Director of Public Education and Research Ashland Johnson, said:
“This is dangerously bad information that will likely be taken out of context, is based on flawed assumptions, and threatens the safety and privacy of LGBTQ and non-LGBTQ people alike. Imagine for a moment the potential consequences if this flawed research were used to support a brutal regime’s efforts to identify and/or persecute people they believed to be gay.“
In addition to the ethical and methodological issues about the research itself, it also raises the debate about what should be the limits for classifying social groups through automated decisions based on the analysis of large amounts of data, whether for research, trade, national security, public policy, among others.
Similar criticisms may apply to various online services which we use on a daily basis for “free”, in exchange for our personal data. While an academic research can be publicly criticized and debated, the use of Artificial Intelligence and Big Data by governments and companies presents additional risks as to the guarantee of transparency of their use.
It is interesting the metaphor used by the author Franklin Foer, comparing our digital age with the food industry. In the second half of the 20th century, especially in the US, the market for processed foods and ready-to-eat meals took off, presenting a future in which people would not waste time preparing meals. It was only after decades of debate that it was perceived the price to be paid for this convenience. Like the harm to health from the consumption of large amounts of sodium and sugar, and its environmental impacts, due to the production chain of these industrialized foods. Similar to the food industry, maybe only now we are freeing ourselves from an era of dazzle with the digital.
To help people who are not from technical areas involved with Artificial Intelligence and Big Data, it is useful to mention the Carl Bergstrom and Jevin West model of thinking. They recognise that many people may be reluctant, at first, to criticize the use of a statistical or AI techniques because they do not understand its functioning. However, most of the problems, mainly social and ethical ones, are related to: (1) the initial data (input) which will be used by the algorithm, as in a problem of biased data; or, (2) the problem lies in how the final results (output) were interpreted. Thus, the focus of the analysis should be, in other words, what goes in and out of the algorithm.
While developed countries present a greater number of debates, it is necessary to foment greater discussions in Brazil on the use of these technologies and what will be the impact on the protection of human rights. As new industries begin to incorporate these technologies, new ethical and social problems will arise. Thus, it will be necessary to include the various social segments affected by Artificial Intelligence on the debate.