Hi. Hello. Thank you very much for inviting me. I'm honored to be able to speak here and I'm really happy that such an important event also gives feminist AI a room. And I see now the time is also running. That's great. Um I want to talk today about my work and my research and I like to start my presentations um with an AI generated image um and see what AI is actually making um of such a situation. So I um wrote the prompt create an image of the rise of AI conference 2025 and I don't feel represented as the speaker but it might be similar to Fabian. Um I'm really happy that I see more diversity in this room um in dress code and in hairstyles in many other forms than on this image.
Why am I doing what I'm doing? What am I talking today about? So um I want to take you um along to my story. I shortly will tell you why I am doing this and I want to convince you that the future of AI must should and will be feminist. Um so I will firstly be talking about why today the present of um AI is not feminist and then I will talk about how the future of AI will be feminist. So why am I doing what I'm doing kind of two years ago um I prompted this picture powerful people at a table and I think it sums it very well up why I'm doing what I'm doing.
Um because AI is here representing powerful people as a group of white privileged men. So if you look at this image a little bit more closely even the hairstyle, the gestures and the color coding of their suits is almost identical. And I think we know such groups of people but actually this is not the reality. This is not the truth. Um so you've already introduced me. I'm doing research on the intersection of AI and power and I'm mostly focusing on AI regulation and AI governance and how to transform AI, how to make AI more feminist. And I'm not doing this on my own. This is not possible. Um last year um end of last year we co-founded the feminist AI community.
We um have expert talks there. We have community meetings. All of this is online and for free. So in case you're interested to interact more with feminist AI, please feel free to join. So why is the present of of AI not feminist? Yeah, the picture again I think this shows it well. Um at my research I'm looking at power and AI and I just brought two examples where power and power structures in the society and AI correlate. So one example and this text is in German. Um I'm from Bavaria and the Bavarian Parliament has only 27% women. So this is far too less. But when I asked AI to create an image of this, firstly this is not the Bavarian Parliament.
Maybe some of um you know it it's like 1970s style or something in in reality. Um but if you look a little bit at the people here displayed um I didn't find a single woman. So here AI is even um yeah strengthening the power structures we have in our society. Um another example um is credit rating. Women get far less credit uh in our society. They have higher interest rates. They get lower credits and fewer credits. Also investments. I think I've heard there's investors here too. So please fund women. They get far less credit. We also see this with AI systems. So the Apple card was inf famousamously famous um when users noticed that some women only got 5% of the credit limit of their spouses with whom they shared credit limit credit rating and also income.
Money is important in order to do a lot of things and also to work in AI and actually um the AI use has a lot of negative impact for marginalized people. So um one example I already told you is the apple card. Another example um is from recruiting. Um this is one of my fa famous or my most most close examples to me because I wrote my master's thesis about it. Amazon um created an AI for recruiting to recruit the best people based on their former recruitments and they automatically sorted out women. And AI is of course also being used in health care and there is a lot of chances but also AI is for instance not working very well um to detect skin cancer on darker skin then there's a lot of different biases stereotypes in generative AI for instance in text generation for instance in being used um to create recommendation letters and if the gender is given by the first name um often biased um biased descriptions appear and um also an
image generation. And here the photo of this woman of color. Uh she wanted to create based on her own photos um an application photo. I also did um a small study on um bias and text to image generation systems. Okay. Now I see something different. Thank you. Okay. Um systems. There was a football something here on my screen. Um and I looked at three contexts at power, success and beauty. Um so these are three exemplary photos that were generated. And here I looked at the gender um representation and also at the representation of people of color. And um overall we saw that only 21% of um generated people were women. Um it was 12% for powerful people, 20% for successful people but almost 60% for beautiful people.
And now if you look at Delhi um in the context of power it was only 6% women. If you look at midjourney in the context of beauty it was 92% women. And we see a similar picture if we look at uh people of color um also here we saw that there were much much less people of color being depicted than white people over all the prompts and all the systems. So overall um we found that there's gender bias. um there were far less women created than men and also women were often sexualized. Then there was racial bias. It was far more white people being generated than people of color. And there was some also some class bias.
So most people generated looked firstly very skinny and very beautiful in a way which is very classic beautiful and um they were also looking very um very prestigious. Um what the makeup and everything they were wearing looked very much privileged. I think all of those things were undeliberate. Um AI has been used there and um by coincidence and by the data that was um underlying um there was discrimination and bias. But actually AI is also being used um in a way to actually manipulate and oppress. And if you look at the oppression um for instance in um Iran um Iran Iran um there we see that AI is being used to um recognize women that are not wearing heat shop or not wearing it properly um according to the rules um then we see for instance a lot of um deep fake pornography being generated mostly targeting women over 90% of the victims are women mostly those um deep deep fake pornography videos are being generated either as a revenge act or to um blackmail
women. And then and I think we've also touched this before in this event of course this also has political implications. AI is being used um in platforms but also to create content which is misleading to um yeah change public opinion. In my research or in our research, I'm looking at the intersection of power and AI and we've created a framework and there we look at how power currently patriarchal racist and many other forms is actually influencing the context of AI development and application. We see that this context is then also petrically formed and this has an impact on the AI systems which are there and if they are then being used they recreate um yeah the systems the structures which are there um I have some examples how this happens so firstly on the context in the political context we still live in a patrical systems which is male dominated um the male is the norm and we also live in a neoc colonial system where for instance in this in the realm of AI
um for instance for CHBT and the data has been cleaned um by Kenyan workers in the global south for very low payment they had to clean toxic data for AI systems we are using in the global north and then we've also heard this yesterday from Fabian we also live in a capitalist society so very often profit first is one of the main motivations within a company which also has influence on the AI systems looking At the organizational context, we see um of course not every company has all of those attributes, but in general we see there's a lot of power imbalance, a lot of um rivality going on and often there's private uh profit first um meaning in the company.
Then we have different um different aspects that are influencing AI such as the people developing AI systems but also the people that have influence on AI. Those could be leaders, those could be investors, those are people that spend money within a company that decide why and how are we using AI. And there we see little diversity and we see a lot of um power imbalances. Looking at the data, this of course also is a problem because those AI system are being trained on data and we see that for instance we have gender gaps, we have um missing data, we have missing representations and we have stereotypes in this data. Then design can be a problem.
In the focus group that we've been doing and Marco is here also he's a co-author um with me. Um we have seen that um often there are value and resource conflicts meaning that um that developers told us they knew the data wasn't perfect. They would have liked to change the data to make it better but they didn't get more time. So this profit first mentality within the organization didn't permit them to actually work more on the data. And finally and I think this is the most important aspect the purpose of AI. So currently the most systems that I am seeing we are using and developing AI to automate existing processes and structures. And this actually means if we are recreating unequitable systems and processes that we of course recreating unequity and not making it more equitable.
one short look um into the people um that um work in AI that that research in AI here in the aspect of gender we have um yeah gender gap here we have far less women working in this field than men especially in Germany and if you look um at the uh AI gender gap within industries um I would say not surprisingly in nonprofits women are mostly engaging in u AI work this matters because um the people. The the creators behind those systems also influence the outcomes as has been shown by a study last year that the different large language models show the political leaning of the creators behind them. But women aren't only um in the creation process less uh active or less involved.
They also use AI less. This was a big meta analysis um looking at very many different studies and they found that there was a big gender gap um in looking at generative AI um in the US and globally women were um using generative AI less than men and this also matters because we are retraining those systems to an extent and those systems should fit much better to the user that they are um but if it's mostly men it will fit better to men also in the future. So overall they saw that women were 20% less likely to use generative AI than men were. I also have one example of data and I still really like this example because I like to show um yeah how strange and how biased data can be.
This was um an example from before I used generative AI to to prompt images for my presentations. there. Then I looked I searched for images on Unsplash and um here I looked for mench so human um at Unsplash. So I'm sure we are seeing some diversity here. There's some diversity there. But you might have recognized that there were no women and also no children. And I don't really want to talk about why in this instance it's like that. But just to say that we have data sets that might not be as as easy to see that there are bias. But we have data sets that are very um that have lot a lot of gaps and if we train an AI now with those images, it will function better for men than for women and children.
Now what do we learn from this? Um we could now say okay we shouldn't use AI but I think all in this room agree that we want to use AI and also that is actually here to stay and we cannot yeah now decide we are all not going to use AI. So we actually have to think um where to use it and how to use it in a responsible way. Um so now I want to um yeah tell you why u the future of AI must be feminist and I think this is what um because we have now a lot of systems of AI systems that are discriminating that are biased in many in many fields for instance they're making false diagnosis in medicine they are discriminating in recruiting um or they are producing biased outputs in generative AI and many more and I think currently um especially also and the political developments that are taking place, we cannot not act on those threats.
So looking on what is happening currently in our world and which is also empowered by AI. For instance, the rise of the far right which is also connected to the use of AI. We see a lot of manipulation and fake news often powered by AI. We see that democracy is under attack which is also being done with AI. we see rising inequity and um there was a saying AI makes the rich richer and the poor poorer. So we also see that AI is being involved here. We have war and conflicts also within Europe and also here AI is being used. And finally and I don't think we can forget this um the climate crisis is also very strongly connected to AI use which is consuming a lot of resources.
So I think all of us um we have the responsibility to think about how to use and AI use and design AI to make the world more equitable. I think this is our responsibility to use this atscale nature of AI to actually do this and to also protect and promote our European values. So I think this is complex. It's not easy to do because we have to change the the underlying system that is changing AI and feminism can be such an approach to do this. Now when I'm talking about feminism um I'm talking about um a framework and approach which is not only about women and I think this is very important to understand. This is a critical and intersectional perspective which is not only about women.
So firstly it's critical. This means it's looking at underlying power structures is trying to understand oppression and the reasons of oppression. And then it's intersectional and this is very important. It's looking at the intersection of discrimination. For instance, the intersection of racism and sexism. It's not about black and white. And in that meaning, feminist AI is not just one simple technical system. It's also an approach to get there. And it's not only technical. It's also activist. it's scientific and it's many other things um to promote and to build systems that change power and not only formal power but also informal power in our societies. So coming back to our framework now it would be best if we just change power because this would have an impact on AI.
I see some people smiling. I think this is not the option we we should take. I mean we should keep on doing this but we cannot change power quickly and currently I think we see more changes back than uh into feminist power. So we have to change the context of AI development and application because this in turn will also change AI systems make them more feminist and this will create also feminist outcomes in AI system and then this is a circle this will also have an impact on the EI application um that we are seeing. So how can we do this? I think firstly we have to work on the political context. Now we have and I'm really um I'm really happy about this.
We have an AI regulation within Europe. I think this is an important step and um I think in this ah okay and looks a little bit different than it was supposed to do. But um um but I think we have to integrate within AI strategy also on a political level that we are concentrating also on equitable AI that AI is being used to promote equity and it should also be included um in AI regulation that we that this is inclusivity and this is sustainability important. We need more research in the area of feminist AI. And one thing is very important here. We need more funding. And I think when when um yeah, when giving out tax money um taxpayers money, we should really then um look that projects, look at the quality of data.
They should look at intersectional perspectives and they should also look who is actually um profiting from from this project and who isn't. We also have to change the organizational context. And here we are looking into how to make organizations more feminist, how to change power structures, how to make them more self-organized and how to promote feminist leadership there. I think what we also need um in organizations certain frameworks that empower feminist AI. I think one thing every organization has to ask itself is what AI values are what AI goals are we following? Will we use AI to actually transform or to reproduce the status quo? AI values. Do does the AI align with the values of our organization or doesn't it do it?
AI strategy. Will AI become essential to our value um creation? And I think most companies here will say yes. And then we should also think about to implement a high ranking person looking and being responsible for AI. And finally, and I think this is also important even though it often sounds a little bit boring, AI governance. I think it's important to ask ourselves what guardrails will put into place and do we want to do the bare minimum for instance compliance with the AI Act or do we want to go beyond looking um at the different impacts on AI. So people, data, design and purpose, we've seen that there's a problem that the people often lack diversity.
And this is something we cannot change very quickly. But there are certain ways where we can still include diversity within the process. So for instance, there is this concept of including committees of diversity within processes. Those can be um those can be made up from people from all of the organization. It can have different genders, different ages, but especially also people from outside the organization, people that will be affected by those AI systems. We've seen that data is often um including biases, stereotypes and here we of course we can um clean the data, we can change the data, we can maybe use synthetic data, but one idea is also to use um biased data for positive discrimination.
Finally, um I've been talking about the design and that there have been a lot of resource and value conflicts here and um I think one very important thing is here that we have to have an AI governance in place that is making such prioritization less um less difficult. So that there's a certain there are certain things within those processes that have just have to be done and they just cannot be um yeah left out of the picture if you don't have much time. And then I think this is the most important thing again is the purpose. I think we should be using AI to think about how can we change processes and structures that are existing.
How can we make those more equitable with AI? And I've been asking um a couple of months ago um my LinkedIn community what feminist use cases they would be wishing for. And I've brought I've brought some ideas and I really like the first one. There was the idea um there could be an AI um looking at the meetings and like um yeah being there within the meeting and then analyzing the the speaking there analyzing sexist behavior and if men talk more than 50% of the time then there would be a red traffic light going on. I think that's a really nice idea. And another idea would be um for instance to analyze um bias within the educational system um with AI.
For instance, I I cannot remember that I've read um a um um female um author's book in my school time. I think it was mostly white German men, I think. Um but this is not only wishful thinking. There's actually um different NI initiatives um that are already going on this way and Franchelli is here. She is um CEO of Bentos AI which I really like. Um and there's a lot of other initiatives, organizations, NGO working on AI that are promoting feminist values. For instance, there is this um this project uh from Bavaria. They are creating an AI system that is trying to um to pre uh to to pre-diagnose um heart attacks in women because we know that doctors often don't recognize the signs within women's bodies and um then those heart attacks aren't being recognized.
So this um this project is there to to recognize those health issues. Then fair solution for instance this is a program where um different organizational informations are um are analyzed which is actually intersectional. They're looking at gender, at country of origin, at age and um disability. And they're looking for instance at gender or different um at equal pay and uh and promotion scenarios on different things and um showing the organization where there's gaps. And maybe one other thing is Rori. This is an AI um being used in Africa, built in Africa, for Africa um bringing math education to children especially also to girls. So what can a feminist AI do? What can it change? So for instance, we can use it uh to promote feminist values following a feminist goal.
So like bringing math education to girls. We can use it to make processes and structures more inclusive like creating more simple language with AI, making it more barrier free. We can use it to make structural bias and barriers visible like with the AI that still has to be built. So if anyone is interested um doing the speech um analyzation about the sexist patterns and then for instance also to make products fit yeah um big diverse population here I often think um of um health trackers. I have an watch here as well. And when they came on the market firstly, none of them was actually tracking menstruation. And this is a phenomenon which is for a lot of people something which is happening during their lives and is still a little bit surprising that this was not on the list.
So if we develop and this is what I want you to take along. If we develop and use AI um in a feminist way with a feminist approach from today, we can actually change our world at scale because AI is being used everywhere. It's changing our world everywhere. If we do this with a feminist goal, this can really change the world. And we cannot do this alone. So, let's do this together. Thank you. If no offense to anybody else who spoke before her, but I took pictures of all your slides. I mean, it's not that I didn't know it, but it's And I also asked myself, uh, do I find it so interesting because I'm a woman?
No, I think it's just like in general, it's it's unbelievable in what kind of world we are living and that it is so obvious in your face, men. Oh, okay. Great. Hi, we have a question from question from Tamara microphone. Okay. Um I was um I studied communication sciences and once I had a really nice talk with my professor and I told her, you know, I realized that um why I love AI is because it's also communication in some way because it's making things transparent. And so even though AI is um part of the problem problem here, it is also part of the solution because it made it visible what we're tackling currently right in our world.
So um I I think it's not really a question but and it's I just wanted to share it because you again reminded me of this thought and um I thought it might be interesting for you too. Thank you Tamara. I think this image of AI as being a mirror of our society I think is very true. I think it's not really a mirror of reality. It's a very skewed mirror but I think it is a mirror. Yeah. Helga for me the purpose of the feminist is caring caring and sharing and this is different um compared with competing. So I think the transformation is in a caring society and that's what Tanya Singer a very famous psychologist on Max Blanc um institute what she says we need a caring soci so society because only caring is sustainable and I think it's something which is missing in the AI edge yeah please reach them in the whole check yeah I have another question and thanks for your uh great presentation so when we think about this all
autonomous cars and stuff like that that over proportionally kill women on the streets in Silicon Valley and stuff like that. Do you think um what is your stance on why are companies shying away of changing it? Do you think they are scared of having themselves considered feminist? Because it's still a term that people are really like, "Oh my god, I'm not a feminist." Right? So what do you think? Why is it that nothing's happening in on big scale? So I think things are happening um but I think they often just don't have to. And I think this is why regulation is really important. Um, I think a lot of people say, "Oh, it's difficult for companies."
And it probably is, but I think we've seen that it's not working without regulation. So, I'm really happy that we have regulation in the EU. Um, I think the the term feminism, yeah, of course, this is a very difficult word, especially for a lot of men, but also for women. Um, but I think for me, it's still the very right term because I think it talks it talks about caring. It talks about equity, and I think it's just the right term. But um I think some people even if it wouldn't be called feminist I wouldn't convince if they don't if they don't mind um um killing women or maybe having those bi biased decision they wouldn't care either but I think we've seen also good um instances for instance last year at Gemini had a big um yeah was big in the news um about um they have they have changed the AI systems to very much show more diversity on the pictures and I think this was really good but then there were people
prompting um people from German var so soldiers from the second world war and those then were marginalized people. They were black people, they were indigenous people, women, Asian people and of course this was a little bit of a problem and they had a lot of um image problems then but I think like seeing this it was also good to see they thought okay we have to change we want to promote diversity. So I think we are seeing things but it's slow and I think in the current environment in the US it's worse. Um can I have um from the tech team the last slide back where we saw the four pillars or maybe you can do that four pillars with the I think it was about actionable things to do because um I had earlier a question like what could you um tell AI developers and organizations what to do like now implement be more inclusive is there anything you would like to add to to that if I if I'm a developer now what
should I do I think developers aren't also the people that really have the power in the system. So I think they can definitely question this. They should also be aware. They should educate um and learn something about gender bias or other forms of biases. But I think they are not the people really deciding upon AI systems. So they may be the ones who are deciding. I think these are people in power. These are people that are taking making decisions and I think they um have responsibility and they should really also take this responsibility and I think often they are not very much aware that their decisions also impact AI very much. So like just imagine the question where are we using AI and why this is a really big question which is having such a big impact on AI systems but it's not technical at all and I think a lot of people in decision power don't realize this and I think just thinking about this is like the most important thing I would want
them to do. Thank you. I would leave that as your last words. That are perfect. Thank you so much.
