Experts warn against discriminatory algorithms and call for greater inclusion of women in AI developments
Vienna (PK) – Every second AI system has a “gender bias”, meaning it can reinforce and perpetuate gender-specific discrimination and prejudices. This was the result of a study from 2022, i.e. before the presentation of ChatGPT, as Sabine Theresia Köszegi, Professor of Ergonomics and Organization at the Vienna University of Technology, reported today at an event in Parliament on the occasion of International Women’s Day. The reason for this is not least the dominance of men in the IT sector. AI is also increasingly being used to perpetrate sexual violence against women, for example through AI-generated deep fakes. According to research by journalist Ingrid Brodnig, relevant AI-generated images of Taylor Swift were viewed more than 45 million times before the posting was removed by X (formerly Twitter).
But from a feminist perspective, does AI also offer opportunities? And how can we raise awareness of gender aspects in AI algorithms and specifically counteract discrimination? This was also the focus of today’s event on the topic of “Women and AI – Discrimination Trap or Feminist Upswing”, to which the Second President of the National Council, Doris Bures, invited people to the House. Artificial intelligence will change the whole world, now is the time when you can have a say in how AI changes the world, Bures quoted Mira Murati, the woman behind ChatGPT, in this context.
If you feed machines with clichés and role models, inequalities will be reproduced and not corrected, warned Bures, citing as a negative example the AMS algorithm, which was hotly debated in public and which suggested tech-savvy job prospects to men, while women were steered into female professions. Perhaps AI could also succeed in ushering in “a golden age of equality”. The Second President of the National Council is convinced that feminist participation in this technology is definitely needed.
Experts for greater inclusion of women in AI developments
That was also the tenor of the panel discussion, in which the AI experts Barbara Herbst, Andreas Kraus and Carina Zehetmaier took part, moderated by APA deputy editor-in-chief Katharina Schell. Once AI has been built, it is difficult to understand and change it, said Andreas Kraus, who, as co-founder of enableYou and feminist AI, advises organizations and supervises AI projects, among other things. You have to start beforehand and pay attention to diversity in development as well as general power structures. It is also important to ask the question why and for what purpose AI is used. If you are aware of the goal, you can build AI completely differently, Kraus is convinced. For him, AI will only be of use to society as a whole if it makes more objective, non-discriminatory decisions. Kraus noted that he noticed that companies were increasingly becoming more aware of ethical issues, at least externally.
AI expert and management consultant Barbara Herbst, founder and managing director of en.AI.ble, sees increasing pressure on companies to massively increase the proportion of women in the field of software development. There are currently very few developers, this is where we have to start, she emphasized. Regulating AI is important, but changing algorithms is just as important. This requires a national effort from companies, social partnership, politics and science. It’s okay that you can’t understand AI as a whole, but only individual pieces, said Herbst, so you don’t have to feel powerless.
Carina Zehetmaier, founder of the non-profit association “Women in AI Austria”, appealed to the – predominantly female – audience to become active themselves. AI must be there for everyone, she emphasized. According to her, there are already so many female experts, especially in the AI field, that no one dares to have all-male panels on panels anymore. She also believes it is important to raise awareness and eliminate skills deficits. In any case, AI is not a tool that delivers objective results; rather, there is a certain subjectivity inherent in AI, Zehetmaier claimed. You have to be clear about this. The expert expressed fundamental praise for the European Union’s “AI Act”; it makes sense to base regulations on the application. In certain areas, algorithms must be comprehensible.
When asked by Schell what he wanted from lawmakers, Herbst mentioned “open innovation.” Not only stakeholders should be heard, but everyone. Zehetmaier advocated including networks like Women in AI in decisions. Kraus would be happy if politicians “positively framed” AI and geared funding, for example for startups, to ensure that ethical principles are observed in the development of AI.
Köszegi identifies the “vicious circle of digital inequality of opportunity”
Previously, university professor Sabine Theresia Köszegi, director of the Institute for Management Sciences at the Vienna University of Technology and head of the Department of Work Science and Organization, had drawn a rather sobering finding. She spoke of a “vicious circle of digital inequality” that begins with the fact that women bear the brunt of unpaid work such as care and child-rearing while men develop their careers. Among other things, these role models contribute to the fact that only 0.5% of girls want to study computer science – 5% of boys – and only 5% of programmers worldwide are female. Segregation in educational decisions leads to segregation in the professional world. Despite better performance in computer literacy, girls also have less confidence in their own skills.
AI leads to a further increase in inequality, Köszegi pointed to various studies. Technologies are mainly developed by men for men. These would not only not counteract the gender gap, but would actually increase it. Even before ChatGPT came onto the market, a study found that every second AI system had a gender bias. In addition, according to Köszegi, large language models like ChatGPT not only have an open bias, but also a hidden bias. In their opinion, this problem cannot be brought under control. Köszegi sees Iceland as a role model for equality: It has succeeded in greatly reducing gender inequality, and it all started with Icelandic women taking a day off.
Brodnig also sees challenges for civil society and the media
The journalist and author Ingrid Brodnig illustrated how difficult it is to fight against pornographic deep fakes on the Internet. The Internet is full of fake porn images, which can easily be found with a simple Google search. Complaints, on the other hand, would often lead to no reaction.
But the issue is much bigger, emphasized Brodnig and referred to research by the Washington Post, which used stable diffusion to create images that were brimming with role clichés. When asked to take photos of a Latina, bikini models in particular were generated in relevant poses. The tech companies tried to counteract such undesirable results with “quick fixes,” but that wasn’t a real solution, said Brodnig. After all, it is not known whether these “quick fixes” lead to other problematic results.
Brodnig nevertheless appealed to the audience to use the opportunity to complain. It is frustrating that civil society, the media and science need to take a closer look in order to bring about changes, but there is no getting around it because technology companies do not want to address grievances on their own.
As for the future, Brodnig said one could develop an AI that is feminist from the start. In any case, now is the crucial moment to intervene. In twenty years it will be too late. AI could also be used to illustrate inequalities, such as the fact that men are quoted or portrayed in the media significantly more often than women. When developing APPs, social concerns could be given greater focus instead of entertainment.
Hopefully in a few years we won’t need a law against AI hatred, commented moderator Schell on the critical findings.
Following the panel discussion, an AI-generated dance performance by a robot from HAUSorg brought the audience closer to the interface between humans, computers and machines live. A photo box with AI-generated backgrounds was also available to visitors. (end) gs
A NOTICE: Photos from this event as well as one Review of past events you find in Parliament web portal.
Questions & Contact:
Press service of the Parliamentary Directorate
Parliamentary correspondence
Tel. +43 1 40110/2272
press service@parlament.gv.at
http://www.parlament.gv.at
www.facebook.com/OeParl
www.twitter.com/oeparl