Yeah, thank you very much for the nice warm words and thank you also for my support here. Thank you for inviting me. Um, it has already been such a great conference with so many nice people and feels very much like being at home already. I've heard um some mentionings of being female just now in the talk before and I also saw some images that I think would have fit in my um yeah slides too. uh I brought an image to start off the session um and I promptly created image of the leading AI conference in Germany Berlin of course I meant the rise of AI and this was the result and I'm going to zoom in a bit more so that you can see and we can we could say okay maybe this is Fabian maybe you can't see it so well but actually all of those people in the front or all of those audience are men um and I think it was little bit similar there just now and now now it has
changed a bit um but we are seeing that AI can have bias and that it's representing um certain stereotypes and that what um I was looking at in my research u my dissertation I looked at the um intersection between power and AI how current power structures shaping AI systems and also how we can change that and um I've also written a book on this um called feminist shakai so in case you're interested on this um you can read more on not female AI but feminist AI. What um I will be talking um about today is firstly um whom does AI currently shape. So what um where does AI have effect on people? Then we will be talking about AI and power and who's actually shaping AI systems currently and then how can we shape AI and change it towards more empowerment.
So whom does AI shape? who's affected by our systems. I brought um a few yeah ideas um probably that's not new for you but AI is being used in very many different industries and scenarios very many different use cases. So for instance it's being used in surveillance with face recognition. It's being used in medical um diagnosis. It's being used to match people when they want to have a date. It's being used to u do navigation production and many things more. In my research, I concentrated on AI in recruiting. I concentrated on um AI on social media. So on LinkedIn, how does it show visibility and are there gender gaps? I looked at um bias in generative AI systems.
And finally, we also looked at AI being used in news platforms to shape what we are seeing and what we're not seeing. AI can help us and we've just I think heard a lot of things where AI companies might also scale and become big and I think a lot of people of us are using generative AI in order to maybe write email. So for instance for me it's sometimes helpful if I'm a little bit angry um and the email is not as nice as it maybe could be then it's for me helpful to use AI but it could also for instance um make language more inclusive or give it more context specific examples. So it can of course help us assisted um us in our jobs but it can also have problems and um oh firstly this so this can help us.
So firstly it can save us time we can be more efficient with generative AI then um it can also improve the quality of our work and finally and I find this um especially important that it can actually especially empower juniors in their jobs that don't have so much expert knowledge yet. So this can all be very positive. we can yeah make maybe our working lives more easy but it can also have problems and one of them is that women actually face different changes in their working lives than men do. Um the jobs that women um are doing are often very much differently affected and more negatively affected than the jobs that stereotypically more men do.
So we could ask is generative AI a risk for women. On the other hand, we are seeing um for instance LinkedIn data showing that AI could also bridge longstanding gender disparities and also looking on LinkedIn, we see that the female talent pool has been growing immensely in the last couple of years. So we can also say maybe generative AI is a new chance for women. And I looked at this very specifically and what I found very important in my research I looked at power and AI. I think we really have to ask who has power when shaping those systems and also what power structures are being rebuilt with AI. And I brought you one example.
So looking at the ICT or the IT professionals in the EU, we see that in the last couple of years they have been rising. But we also see that it's still only like one in five people is a woman. So this percentage hasn't really changed even if the um overall um yeah number of people have been rising. And this is a problem and you might have heard about this um example. Um it was the first that I came across with um showing that AI is not neutral. AI was being used by Amazon in recruiting to look for the best candidates. They had trained the AI system in recruiting on their past um recruiting. So they have trained it on their own workforce from the last 10 years.
So they have recruited mostly men and this AI system has learned to really automatically sort out women. And now I want you to think about what data you're using, what data might be in AI systems that you're using, and what data might be in your company that you're retraining and fine-tuning AI systems with. And what would this also do? So what would it reproduce? So coming back to the many different industries where AI is being used, um I brought some examples where it can have detrimental effects for marginalized people. So for instance in the way of credit rating um studies have shown that for instance women get less credit than men do or in the instance of mortage algorithms they often reinforce racial bias in lending.
We've seen something similar in recruiting. Um I showed you this example already but we also saw racial bias here in Tik Tok on social media. Um they have been um they have actually um yeah said um that they had actually changed the algorithm to um reduce the visibility of queer people and people with disabilities um saying they wanted to protect those people. And if you look at diagnosis um at AI and medicine, we see that very often we have less data on women and actually that AI systems are not functioning as well for female bodies than they are doing for male bodies. And looking at facial recognition, which is largely used in many different surveillance instances, um we see that for instance, women of color are especially um strongly mclassified in those systems.
And looking at generative AI, there's also many many examples where they show bias. So for instance, JGBT was used to produce um recommendation letters and showed very strong stereotypical descriptions of women. AI was used to create images of people of of women of color and very much sexualize them in their depiction. And then um you might be aware of Grock um which is the AI that um is behind where Elon Musk is behind this AI system. It's also a large language model and so you can produce text and images. And when we look at the um results it's giving it's often very um political incorrect. It's often anti-Semitic and for instance a user asked what would be the person the solution um from the 21st from the 20th century um a political person who might um actually solve the problem of the um hate against white people and Brock said this would be Hitler no question on the other hand um you might also have come across this um end of last year and beginning
of this year um Grock was widely used on X to undress women and children. So users and this was actually male users were using the system directly on X in order to change images that were there already and to undress um people. And um I think in this instance I also have to show this um example um where allegedly also deep fakes have been used against Colleen Fernandez by her own um ex-husband um in order to um yeah to create instances which she didn't agree to. Um yeah so AI can be used in a way that it's marginalizing people marginalized people and I think this is a question of power and this is an image I created I think already three years ago but if you saw the image um the first image I showed you maybe not so much has changed so we see that actually people that have power are still often white they are privileged and they are very often male and this is what AI replicates so who shapes AI systems
Um I'm now focusing on the example on of Grock um in order to show you why we see those results, what's happening behind there. It's very important um that we as humans shape those systems. We give priorities. We shape them with our design. So actually what we see is that the deep fakes being created on X have been very good for business. It was a business opportunity and actually um this um the that so many users use those systems was very good for XAI. So the the business behind Grock on the other hand we are seeing that Grock um often makes very controversial I would say statements because one of the core functions when we're looking how it's actually working internally is that it's firstly researching what is mask saying on a certain topic and then it's replicating this and adding certain other aspects and finally there's also system prompts within Grock how it should behave this is it's like that in all large language models and for instance JBT is often very much supportive
saying, "Wow, that's a great idea. You should go on with this." Grock is a little bit different. So, Grock um is actually told to say political incorrect statements if it finds sources for this and it's also told to be very um skeptic toward traditional media. On the other hand, we are seeing of course that data is very important. So, we are training those systems with data. And looking back at Grock, we are seeing that the system is being trained on very many available data on the internet and a lot of this is also toxic. There's pornography, there's a violent content and so on. But what we're also seeing is that Grock is being trained on data directly from X.
So when you write a post there, this might be training data for Grock. And we don't we know that there's no facteing anymore on on X. So actually this could be any content. Furthermore, Musk actually urged his fans to um post was Twitter now X to post controversial content there in order to train his spot. And what followed were a lot of misogynist, racist and many other um claims that that were also just false. Also looking more deeply into the data, we are seeing that of course again humans shape them. And what we're seeing um the light blue is the female portion on X and the dark blue is the um is the male portion on X.
So around 35% of um users on X are female and of course this has also an influence who is writing there what perspective is being seen. So looking at this from a broader perspective what is happening. So we see we have certain power structures in our society and they influence what is happening um in the creation of AI systems. If we look at Grock we are seeing that the people that shape this it's very much shaped actually by Elon Musk but we also see that a lot of those people are actually male and privileged. So looking at the gender gap in AI, we see much more men working there than women. And this is also true not only for technicals um people but also for those that really make decisions in the management.
And furthermore, we also see a gender gap in the usage of AI systems. So generative AI is far more used by male users than by females. And of course also they determine um the next couple of answers. they have give impact they give feedback to those systems. So also the users of course are very important and also here we have a gender gap and this also will mean that um as women we might not learn certain skills that we will actually need in the future. Looking at the data um with Grock we saw that it's very much shaped by X is very much shaped by male users. Um looking at data generally we see for instance a gender data gap.
We have less um visibility. We have less information on female bodies for instance in medicine. But we also have a gender gap um in so so many other um fields. And this is also intersectional. So we also have gaps. We have much more information on white people often much more information on um academically educated people. So for instance the people posting on Wikipedia, posting on Reddit, those are much more privileged people because they have the time to do um unpaid labor there. Looking at the design um in the instance of rock, it's actually very very much shaped by this one person but um it's also shaped by other priorities. um people in those systems they set the priorities that um yeah the um the AI is supposed to do and the most important thing I would say is the purpose.
So why is AI built? Why is it used? And what we're seeing here in the instance of Grock, it's to manipulate people, I would say, or to manifest a certain world view that actually is behind um behind Musk. But on the other hand, it's also very much profit-oriented business case. In general, companies building AI but also buying it, using it, they want to automate their existing systems. So very often it's they have a process. So for instance I was saying the um Amazon using AI and recruiting. So they had a recruiting process now they want to do the same process but more efficient. Um but they don't really rethink the process but they just want to automate it.
It's not getting better with this. What is also important is the context where all of this is happening. And looking um at the political and social context um maybe difficult to read but um the first word is supposed to be patriarchy. So we live in a world which is still very much male-dominated where often male is the norm where a lot of men make more decisions than females. We live in a system which is also very very much driven by neoliberal capitalism. So capitalist values profit is often the first goal and we live in a neoc colonialist system and I think this is also so important to really um look at this in the field of AI.
AI systems are often very much developed for instance in the US they are being designed there. They're being trained there. But the data that is often also coming from the global north is often reworked is labeled is um changed and shaped in the global south in with very many very detrimental conditions very vulnerable working conditions there and those people profit less and often also the systems don't really function for them well in their context. So what we are also shaping here is a new colonialist um system for my book. Um one illustrator um kind of yeah put this into an image um with all the data workers um on the ground kind of carrying the system then all of us kind of being um yeah using this sharing our data uh being addicted to that and then it's going on and um up and it's getting more and more um homogeneous.
And here on the top uh he showed um Musk, Ezus and Musk, Bzos and Zuckerberg um as kind of the people that that rule this system. So we see that the people shaping those systems um from a companywise are often quite homogeneous but also um there's impact from countries and from political real um settings. So for instance, the US has really changed and deleted data since um m um since Trump is back in power. Um a lot of different data sets on climate change, on disinformation, um on diversity, equity, inclusion, on racism, on police violence, a lot of those data sets have been either removed or have been changed and so they won't be accessible anymore for future AI trainings.
On the other hand, they have um yeah published um an AI action plan last year saying of course all of this happens on the basis of the freedom of speech and by doing this they want to eliminate the references on diversity, equity, inclusion, disinformation and climate change from the AI risk management framework. And furthermore, they only want to um in the future work together with large language metal model providers that make sure that their systems are bias free and objective. And of course, the question always here is what is this? So, but we also see that political actors can take a very strong um um a very strong influence on AI systems. Now, this was quite negative, I would say.
Um but AI is here to stay and now we have to question how can we change it and how can we actually use it to more empowerment. So how how to do that? Um I'm have been using the word feminism a couple of times already and of course it's not an uncontroversial um term. So actually one in five people in Germany believes that feminism is disrupting um the social harmony and order. um yeah which which actually might be right um if the current social harmony in order is not equitable which it isn't but I think it was meant in a different way. So what do I understand when I'm talking about feminism? I understand this as a very critical perspective on our world which is intersectional.
So it's looking at privilege, it's looking at discrimination but not only on one factor. It's not only about women and especially not only about white women. It's critical. So it means it looks at oppression. It wants to understand power and why do we live in such a society? What is the root cause? And then it's intersectional. And this means it's looking at overlaps. So for instance, it's coming um from actually black feminism where black feminists have seen okay I'm discriminated by gender and by race and this is very much different and stronger than for instance only being a white woman or a black man. And we all have multiple characteristics. We have an age which is actually the most discriminated aspect in Germany.
We have a gender, we have a background, a social background, but maybe also a socio socioeconomic background. We have a certain um appearance and so on. And all of those aspects add to the privileges that we have in our life and also the opportunities but also the discrimination. So feminist AI is not one system and it's neither only a technical approach but it's actually activist it's researchbased and the goal is to change AI systems in order to create more equity and also also to change with AI um existing power structures in our society. So how can we get there? we have to do a lot of different things because it's much easier to recreate existing structures with AI than to really change them.
But I brought a few um ideas what we can do and I think the most important thing again is the purpose and I brought here an example. It's called Mist Journey. I don't know if people in the room have heard about this. You might have heard about the app Midjourney where you can create images with Yes. Um and this journey was actually a campaign for one year that wanted to show we have a bias in AI systems and we have far too less images on women for instance being CEOs, being politicians, being polic but we have less images and um they said okay if for instance young girls don't see those im images they might not want to be in this position and also AI systems will again be trained on images.
So we need a higher representation there. So the purpose of the system was to create more images on women and also very diverse images on women. So the purpose was to change an existing problem. I would say talking about people um I think diversity of course is also important. So we have to think about how can we create more diverse people in the whole process. there's a lot of talk about um including um participants um that are affected by our systems more diverse people which is not easy. I think one um way of doing this in a more straightforward way is um having a kind of diversity panel or an ethical um committee. Um also big companies like IBM for instance have something like that.
And when you're making decisions about AI and your um workforce your team might not be as diverse as you it should be to make those decisions you can integrate this um committee in your decision making. For instance, you could place people from different gender, people different age, but also people that are affected by AI systems. Looking at the data, of course, that's um one of the main reasons why AI is biased. And I think one step that's I would say kind of easy that we can do. A lot of us are now fine-tuning AI systems for our own use, for our own companies. And one important thing is to make transparent for other users and maybe for our customers what data is in there.
So then we can already question better what results it might have. So really make transparent what's in there. And then um on the design so what priorities do we set? Um in our focus groups in our research we saw often that um the problem was that um a lot of decision making was based on very personal individual decision making. So maybe a person in in a management position that no we don't have time we want to go um live even though data is not good but if we have an AI governance in place that really says we have to do this this and that before a product can go live it's less based on individual uh decision- making now in order to get there um we have to talk about AI goals so why are we using those systems do we want to use them to transform existing processes or do we want to do it in order to just automate what was there then usually it's not becoming better.
What values do we have as a company? Um what values do we set ourselves and what values might be an AI system that we are using? Maybe Grock is not the right AI system for our company. AI strategy um will AI be very important for our um for our work in the future? Then maybe we should also have a person on a high level that is really making decisions upon AI and then all of this can be implemented by having AI governance that is structuring really all of this and making priorities that have to be done when working with this and that's not all just in my head or in wishful dream but there's actually a lot of different companies NOS's different institutions that really build already systems that solve social and political problems.
One of them is my protectify. They are building an AI bot um called Maya which people can use mostly women that face domestic violence. So they can get information, they can get help and support there. Koko is an AI system that is being built to um create very diverse images um by different forms of prompting and by different training. So hatskai is an AI system which is being built um in Bavaria also supported by the Bavarian ministry. Um and there the the purpose is to um early detect female heart attack which is often by human doctors um very late. So they they detect it often very late. Fair solution is an AI system which is looking at data within companies and um trying to find gaps there.
So on gender, on people with um disabilities for instance and also on age and uh and social background is an AI system which is being implemented and used in Africa being built to give children and especially girls smart education at home. You can use it via WhatsApp. And then of course I also have to say Bentos AI fans is also here um which is an AI system which um is um should learn or should really support people that work on the basis in merit team ocean conservation projects and then also connect them and work with together with them with their data and don't just extract it and use it somewhere else. How can we get there?
So all of us can I think in our daily lives do something about this. So for instance, we might be involved in procuring AI systems. We might make decisions which AI system to buy and which not. And we can do basic diversity checks. We can for instance, I hear this very often. I would not recommend to do this, but I hear companies use CHBT to um to sort out um um applications of people. Um, if you do that, I would not recommend to do this, but you could um at least put CVS there which are the same except maybe the name which might sound female male or German not German and see what happens. If you do a call for tenders, include diversity and intersectional perspectives here.
If you're involved in AI projects, um, look, if you have diversity, equity, inclusion people in your company, include them. This is something which is not happening at all in AI projects. The real responsibility lies with leadership and management. Those are the people that set the standard. They set the baseline and they um actually kind of decide why are we using AI systems. So those are the people that really have the responsibility. Most of us are also using AI systems and also here we can do something. We can question which AI systems we want to use. We can question can we use this AI system to change something? maybe to make for instance language more inclusive and not just to replicate patriarchal speaking and of course there's prompting also very important um on this we did um a guideline two years ago on how to do AI prompting in a more fair way so what can you do what should you do you should use AI in a more feminist way because then you would also
change the next um yeah generation of AI systems you should do this from a feminist perspective in order to highlight light bias to show toxic content and to really think rethink processes. We should we all need the positive cases. So share positive AI cases, make them visible because there's always also other solutions that you can use. Very importantly showcase the work of others. This is something not only in AI but it's everywhere important and especially showcase women um such as women as speakers. I think that's one very important thing. And then you can and should join communities to work together. And one community we have um is feminist AI. There's also people from our community here.
So in case you want to talk about this more, you can do this in our community. So what can a feminist AI do? It can make our world more equitable in very many different ways. So for instance, it can offer math education to marginalized girls. It can make processes more inclusive and um bring down barriers um in order when closing health gaps for instance. It can make structural bias visible for instance when looking at company data and making gaps visible. And it can make services more fit for a large um audience. For instance, health trackers that also can track menstruation. In this case, it's not only fair or nice or good, but it's also good for business.
Um, I'm doing research on this. I collect examples. I'm doing talks on that. And I also assist companies in moving in this direction. Because when we are thinking and using AI in a feminist way, we can really change our world because it's used everywhere. We can't do this on our own. And mass mobilization in the past already has changed our world. So let's do this together. Join the community and with this thank you very much for your attention and maybe talk to you later. Thank you so much. If uh let me know this is probably not the easiest to wave the flag of this. Like you said a lot of people have negative connotations to this.
So thank you for doing it. Thank you for getting up here and uh showing us all also the examples and what we can do. So now is the time to grab some of that birthday cake I was talking about after the break if you want because we