Eva Gengler is a business information scientist and author and examines AI from a feminist perspective. But what constitutes feminist AI anyway? In the ARIC interview, Eva Gengler explains why AI is not neutral, what myths are circulating and how technology could be made fairer.
ARIC Hamburg: Who are you and what do you do in your work?
Dr. Eva Gengler: I am Eva Gengler, PhD, and in my work I deal with artificial intelligence from a feminist perspective and from a socio-technical perspective. I understand AI not only as something technical, but as something that is very much connected to society. Society has a strong influence on AI, but AI in turn also has an influence on society. Over the last four years, I have been doing my doctorate on this interplay. I have looked at where AI can ultimately lead to oppression, why this is the case, what this has to do with power structures and how it can be changed. My perspective is intersectional feminism; it’s not just about women.
What is your personal background?
I have a doctorate in business informatics and co-founded our feminist AI community in 2024 and have been involved in an association that campaigns for women’s rights for many years.
I came to AI through my Master’s thesis. I wrote it on the ethical implications of AI in recruiting. I very quickly realized that women, for example, are sorted out in recruiting when data is used to show that the majority of hires to date have been men. Basically, it shows that bias and prejudices that prevail in society can be reinforced by AI. I have also led several EU-funded projects and during this work it became increasingly clear to me that it is actually about feminism.
What exactly does feminist AI mean to you?
It does not mean that we reinforce existing injustices that exist in our society with AI, which is indeed what many AI systems are currently doing. For me, feminist AI is an approach that is about understanding from an intersectional feminist perspective: Who has power? Who has privilege? How can we create technological systems that ultimately make our world fairer and distribute power more equitably? It’s about tackling social problems with the help of artificial intelligence.
“we actually see in a wide variety of application areas that AI is not neutral“
Can you give a few more examples of cases where AI tends to support injustice?
We see it in the generation of images, where the most diverse stereotypes are depicted, where, for example, mainly white people, perfect people, with very beautiful skin, very slim people are depicted, women are very strongly sexualized.
Of course, this is also the case with texts, for example when ChatGPT is used for letters of recommendation. But AI is also used in police work and in military work. And in police work, for example, AI often discriminates against people with darker skin color, i.e. racist connotations are depicted. Likewise in advertising, different people are shown different apps. We also see unfair tendencies on social media, for example in terms of: what is played out, who has reach? So we can actually see in a wide range of application areas that AI is not neutral and reinforces injustice, because AI has already become very much a part of our lives and is already having a very strong impact on our entire lives. And it’s not just generative AI that plays a role, AI is also involved in a wide range of processes and influences a wide range of decisions.
I have also had my experiences with image generation. For example, I tried to generate a realistic-looking older woman. It was really a lot of work and she still doesn’t look average, but has model facial features.
Yes, slim and with long hair, where you think to yourself: Hey, I don’t even have that much hair at my age anymore.
And always totally feminine, as if there weren’t other expressions of womanhood…
But where does the responsibility actually lie? Is this an issue that, in your opinion, lies more with the users of AI or more with the developers?
At the moment, a lot of responsibility actually lies with the user, but I would say that’s a problem. Yes, if I use AI now and create a text or an image with it, then of course I am responsible to a certain extent because I prompted it. If I continue to use it, I have to look at what’s in it and what it shows.
But too much responsibility actually lies with the users. Providers could also give us a message: Hey, maybe there’s a bias in there, why don’t you take another look? Why don’t you check the result? Are you really happy with it?
There could be very active hints that make me question myself as a user. But I don’t think that’s the intention. Much more could be done: Transparency about which data was used? What happened in the development process? Who was involved? Who was not involved? Where did the data come from? Was it somehow processed in the Global South? Did this happen under precarious conditions or under fair conditions?
So there’s so much in there that a user has no influence on. Those who really have leverage are ultimately the providers. To specify this again, I would say that the responsibility here also lies at the top and not in development teams. These are the people who say which world views play a role. I had just given a presentation where we also talked about Grok and the undressing of women and children. Grok received this Spicy functionality last year in August. People explicitly released it. People said: We want to develop this, it should be part of the system.
I don’t think that just any developer at the bottom of the food chain would come up with such an idea, but that the problem is much higher up. And in my opinion, this responsibility is borne by the people in management who either actively want this or do nothing about it. And that is something that many people in management have not understood at all.
I’m not thinking about the providers, but about the companies that buy and use AI. Management is absolutely crucial here too. It considers where budgets go, how much budget do we have for what, which AI might we want to use? Who do we work with, how much time do we spend on testing, and so on. In my opinion, the greatest responsibility here also lies very, very clearly with management and only significantly downstream with users.
What advice could be given to management departments on how to avoid a faux pas and make truly meaningful feminist use of AI?
We first have to think about why we are using AI, i.e. what is the goal? This is a really important question at the very beginning of the process. Once we are a bit further along, the question arises as to what values we have in the organization and are they actually reflected by the AI or are they perhaps actually being corrupted by it?
I am also thinking of something like ecological sustainability. AIs consume a lot of resources. But many providers don’t make it public what exactly that means. When I have spoken to companies in my current research and work, the topic of ecological sustainability was virtually non-existent.
Another point is discrimination versus diversity. For example, if we now say, hey, we actually want to reduce our ecological footprint, we want more women in management and so on. But we are now using AI systems that actually have the opposite result in terms of their world view or the way they work. That is a problem.
In my opinion, I need a person at a fairly high hierarchical level who is really responsible for AI, who is involved in decisions, who makes the topic a management issue.
And in order to implement this, it is very important that we talk about AI governance. This means that we consider what processes, what structures we create in a company, what quality standards do you have to meet in AI, what quality checks do I have?
There are quality checks and compliance checks in a supply chain anyway, or when systems are purchased. AI should also be viewed from a critical perspective here. We actually also need this because of the EU’s AI regulation.
About the interviewee
Eva Gengler holds a doctorate in economics and social sciences. As a business information scientist, she researches and works intersectionally and feministically at the interface of power and artificial intelligence. At FAU Erlangen-Nuremberg, she completed her doctorate in the Business and Human Rights program on AI from a feminist socio-technical perspective. As a speaker and co-founder of the feminist AI community and enableYou, she is committed to fair and responsible technology design. She is the author of “Feminist AI”, which was published in March 2026.
Eva’s book was very popular with the ARIC team. It is a very well-founded presentation of feminist perspectives on artificial intelligence that covers many aspects and is also easy to read.
You also wrote about AI reality and myths in your book. Which myths concern you?
- The first myth: AI will take over the world. At the moment, I’m not afraid of that at all. It comes from science fiction films. There are people and institutions that warn against it, but I think this myth is really very unlikely at the moment and a distraction from the real problems. I would now be much more afraid of the institutions and people behind AI and what they use or misuse AI for. I do believe that powerful people have more power with the help of AI, but it’s not the AI itself, it’s people.
- Secondly, AI will take all our jobs. I think that there is some truth to the idea that jobs will change, that fewer people will work in some areas and that boring activities will disappear. But I just saw it again today on Instagram in a study that ninety percent of companies say that AI doesn’t actually make them more productive. But what I would also like to say, from a feminist perspective, is that women and men will see different effects, because many women will end up with more assistant jobs, perhaps not management tasks, but more preparatory tasks, research tasks and things like that. These are areas that are easier to automate in some cases. That’s why I expect different effects on women and men, and that’s what initial studies show.
On the other hand, many women work in care work, either at home or paid. And this is something that AI is less likely to automate. - Thirdly, AI will replace human creativity. I also worked with an illustrator here, Jan Hendrik Ax, who said from his perspective that he finds it difficult when publishers or institutions simply say, okay, we’ll do everything with AI now, because it’s actually not nice and it’s theft of intellectual property because the AI has been trained with data from artists without their consent.
- Then: AI is like magic. Because I think many people end up thinking it is. But – unfortunately – in the end it’s not magic, it’s statistics.
This is mystification itself in its purest form.
Exactly. And in the end it is also an overestimation of AI.
You feel like you have people who attribute divine powers to AI. Or just devilish…
Exactly. It’s one or the other. Either it can do everything and is divine, or it is demonized in the end because it is inherently evil. In the end, of course, neither is right.
Next myth: AI is intelligent and thinks like a human. In the end, AI does not think or reflect.
Last point: AI is objective and error-free. I think that’s an explicitly difficult myth because a lot of people think that if they use AI now, it shows the truth. They use systems like ChatGPT instead of Ecosia and Google or whatever search engine. Of course, search engines don’t necessarily show the truth either, but at least they show different perspectives and their origins. But AI is anything but objective. It reflects patterns from data and of course it also makes mistakes.
The consequences of these myths were then based precisely on the fact that AI discriminates, disadvantages, that it absolutely reinforces neo-liberal capitalist tendencies and that it ultimately contributes to the intensification of the climate crisis.
Do you know of any examples where you can say: Yes, that was solved well or maybe it really is a feminist AI.
I would probably rather speak of more feminist AI. A truly feminist AI is a very ambitious goal. A really cool example was MissJourney, instead of Midjourney, with which you could generate images. This tool came from an agency that has a lot to do with images and campaigns. And it has sa gt: Okay, we are just realizing that there are a lot of problems with the representation of women with AI in image generation. We’ve been fighting for years to ensure that women are portrayed differently. And now it feels like AI destroys everything again within a click.
That’s why you developed Miss Journey, which only created images of women and these were relatively diverse. Women with headscarves, women with different skin colors. They were still very beautiful women. The system wasn’t perfect, but the idea was to show, okay, here, for example, there are no women who are pilots, who are somehow CEOs and in many different work contexts. So we have fewer female pilots than male pilots, but we have even fewer images of them. Images have an influence on our thinking and that’s why we need more images of women.
Then there is Aymur AI, which is an AI that is being developed in Latin America in relation to gender-based violence.
This AI is actually from an NGO and was funded by a Canadian ministry, among others. They had actually already funded projects on feminist AI in 2021 – I think that’s really cool. And various initiatives have emerged from that. Then, for example, there is also AI in recruiting that does matching. Rather than just selecting the ‘best’ based on previous decisions, we ask women where they actually are, where they want to go, where companies are and then bring them together. Why is that important? If we look at current recruiting, we know from various studies – not exactly, but roughly – that women apply less often than men if they don’t fulfill the entire list of requirements. Because women tend to think that they have to fulfill all the requirements and men tend to think, I’m already great and I’ll just do it. And that means that we actually lose many women in the step before an application is sent to a company. Women don’t even apply in the first place. And then, for example, there are also AI systems that are being developed to match jobs and people. This can lead to women being matched to jobs that they would not actually have applied for themselves.
So there are many different ideas or feminist causes where AI is used to change such injustices in society. These are often smaller initiatives, such as from smaller companies, they come from science, are scientific projects or are really based in the NGO sector. There are also certain communities and initiatives in larger companies. But many of these initiatives that we see actually come from an activist context.
“What often characterizes feminist AI is that it tends to be context-specific solutions“
Why do the positive examples tend to come from an activist context?
In many projects, there is no explicit mention of feminist AI. For example, the recruiting example I gave is software from Chemistry and it was used by herCAREER, the largest leading trade fair for female careers in the umbrella region. They talk about good AI. I would still say that the purpose is clearly feminist.
Why are larger companies sometimes less interested in feminist AI? I think maybe they don’t really think about this possibility. What often characterizes feminist AI is that it tends to be context-specific solutions and not so strongly scaled. When I think of large language models, the goal behind them is that they should be able to work for everyone in some way. Every person should be able to use it. I don’t think that’s entirely true, because it probably works better for US American people than people in Africa, because it has simply been fed much more with their world view and data and ultimately works better in English. But the idea is that everyone should be able to use it. And with many feminist AI solutions, I think it’s more about addressing an injustice in a certain context.
Another example is Bentos AI. This is a small non-profit AI start-up from Berlin, founded by a Brazilian woman. They want to use AI solutions in coastal regions where there is a high level of biodiversity to support climate protection projects with local people. And the idea is not to take all their data, extract it and then use it for my business model, but rather to create small, isolated solutions that people can use themselves on site, where they can share information with each other in a certain context – for example, to protect mangrove forests. People who are also in such contexts can then use this system. But there are various smaller, I would say, system solutions for different contexts. They can also communicate with each other to some extent, but these systems are actually small and not scaled locally. And that’s why they naturally use fewer resources and the data is more secure. And I think this is also very typical of feminist AI solutions and perhaps not the biggest business model at first glance.
Perhaps as a final point, what I think is often overlooked by large companies is that they often think that they are already making very good decisions. And we often don’t, because we humans have biases and mental distortions. But if we approach the design and use of AI with this knowledge, then we could use AI in a much more feminist way. Of course, this is also good for companies and their business, because if we reproduce existing problems and existing human errors with AI, it’s not good for anyone. That’s why I think that perhaps this awareness is not yet there enough and the pain, the issue with We want to change something is probably just higher in more activist, smaller contexts.
“I don’t think it would be easy to turn ChatGPT into a feminist system”
Mind game: You and I can take over ChatGPT. What do we do differently to make it a bit more feminist and fairer?
I don’t think it would be easy to turn ChatGPT into a feminist system – simply because this system is based on so many patriarchal, neo-liberal capitalist and neo-colonial thoughts, goals and practices. But we could certainly change things.
The problem is that there is already a lot of data in there that is not feminist. That’s why I would firstly argue that we need new data. We need more feminist data and we certainly need to think about how we can collect more representative, fairer, more equitable, more diverse and more feminist data. Perhaps we could then continue to use the existing system and fine-tune it with such data.
We could also talk about system prompts. For example, ChatGPT should usually answer nicely, act affirmatively and so on. We could get involved here and, for example, say:
A system like this should also tickle users in their own opinions and say: Hey, is it really true? Think around the corner again! Have you thought about this and that?
What such an AI system actually does is to represent a majority opinion, a statistical probability. But of course you could also go and say that you also try to show certain fringe opinions. By that I definitely don’t mean right-wing extremist content. But if we now go into smaller groups or consider various other contexts for which there is simply less data, then it doesn’t actually occur as an answer. And we could change that using system prompts.
For me, the question would really be how we manage to show such opinions and represent an alternative to the “norm” to some extent.
I would find it very important that we provide information to users: What we say is not the truth, but it is a probability. Use it responsibly, maybe question it, click on the links, see if they work, see if they lead to what the system says. So I would say that supporting users to use the system responsibly would be very central for me and actually that from the very beginning, so to show a bit more, what’s inside, what data is inside and so on.
The last point would be ecological sustainability. We need to record what the system consumes and what it has consumed during training and make this transparent. For example, you could also show a water bottle or another visualization on the edge and depending on how long I prompt, it gets emptier and emptier. So just show, hey, this is doing something, think about what you’re using it for.
I’ve noticed that people always talk about ChatGPT and then they say ‘He said’ and ‘Chatty said’, it’s somehow always a male entity.
As you just said, it’s always so nice and approving. That’s basically what you expect from such a nice woman. That she always smiles and says: “That’s great.” And I once read studies about how service bots are always read as female.
Yes, even something like Alexa. These are the systems that are supposed to assist us, that have female names and female voices. I also referred to this briefly in my book. Alexa used to have a function for dirty talk, for example. It no longer exists, but you can watch it on YouTube. She says some really weird things.
But the question is still: Why does this function exist? Then people have said some inappropriate things or sexualized insults, then something like thank you for the feedback or something. It might seem funny at first. But actually it’s not at all. We should really think about when we gender and humanize an AI, which is fundamentally a problem, as you say, Chatty… some of them really do have names and a whole relationship with these fictional characters. Difficult. Especially if we do that in a sexualized way, then I think that can have serious consequences for human relationships and we should definitely not normalize that.
“we live in democracies and we have a political character“
Is there another topic related to AI that you would like to place?
I believe that politicians have a great responsibility here. To a certain extent, this is also being realized through new regulations and so on. At the moment, however, we don’t know exactly how far the regulations in the EU might be softened again due to the “efforts” – the pressure – from the USA. Politicians need to look at what AI means for our democracy and society and need to be aware of this themselves. And they should have an interest in protecting human rights, women’s rights and the rights of other groups. I see far too little of this in politics. And here, too, there is usually a lack of intersectional perspectives, i.e. not just in relation to women and men, but how people with different structures of disadvantage are affected, so including something like that in regulations doesn’t actually happen. That’s what I really expect from politicians.
In the end, it’s not just about the economy, we live in democracies and we have a political character. We have regulations. We set incentives, for example in relation to subsidies. What must be fulfilled for funding? If I want to put a project out to tender, I could specify that you have to pay attention to data, you have to pay attention to intersectionality, you have to look at risks. So these are points where politicians can set positive incentives. They could also say that there is explicit funding for AI systems that try to help solve social problems. Politicians can create opportunities on both sides, through rules and fixed things in regulation and more positive financial incentives and facilitations. This is also an important responsibility that we must not forget.
Interview: Sabrina Pohlmann
With our interviews, we want to introduce you to different perspectives and players in the field of AI. The positions of our interview partners do not necessarily reflect the positions of the ARIC.
Further interviews:
