Affects Ex-Machina: Unboxing Social Data Algorithms

1:31PM Apr 26, 2021

Speakers:

Caroline Sanders

Ariana Dongus

Claudio Agosti

Keywords:

data

facebook

understand

content

image

harassment

people

day

media

form

system

social networks

algorithm

research

experiment

hate speech

platform

policy

data sets

year

3d printed models of illegal ammunitions to train neural networks in order to recognize those ammunitions in video footage from Syrian war as presented yesterday panel, archives for collective resistance, tech, sky and Husky we want to explore new forms of activism and civic participation at the level of increasingly complex technological abstractions. I guess it is common sense today to say that data is the new currency of corporations, and the data medium of state control. But, of course, the situation is more complicated. For instance, yesterday in our keynote on console temporalities Jackie Wong mentioned this hypothesis of global High Tech Data panopticon that Nick Bostrom envisions that is facilitated by the alibi to protect the vulnerable planet, since technological process has made it easier than ever, to cause destruction on Cantor's scale. As much as we challenge AI as a medium of political control, we should also investigate its economic impact, its effects on labor markets, and measurement of labor. There's a new universe of precarious labor behind the avatar of AI. So my own research I have gay, I study new techniques of biometric control in refugee camps in Jordan and Iraq, where this new matrix of computation integrates with the economy of war and humanitarianism, and produces new computational measurements of labour time. our panel today examines how machine learning and askew algorithms analyze and manipulate individual effects into political sentiments, eventually amplifying class, gender, and racial bias. So these new forms of data capture the data emotional, behavioral, or biometric negative form of governments, that is our wismec, rendering new forms and levels of control and surveillance of lives possible. By that they are exposing mostly those people in our societies that are already marginalized by the dominant powers. So it's time to find ways to reclaim not only our data, but also our ethics and transform them into a new country infrastructures, if you wish, as collective forms of social resistance beyond fragmented filter bubbles. So how is it possible to reverse engineer the expanding capturing of emotional data, which forms of resistance are possible against social data algorithms for upcoming European elections in May this year, by the way, also in India, to data surveillance in India, the speakers will discuss and present new projects and strategies of algorithmic activism and data sovereignty. The people we invited work to make biases visible to show the inner logics of machines and it's designed to develop tools that serve as counter infrastructures against today's seemingly monolithic and privatized global networks of control and vision. So let me introduce the speakers now they forgive the inputs in the same order. Karen Sanders is a machine learning designer, researcher and artist. She's the founder of convocation design and research, and has worked with Amnesty International into IBM Watson and the Wikimedia Foundation. She's a research fellow at the Harvard University's Kennedy Kennedy School of Government and policy. Her work has been featured at MoMA ps1, the Euston center for contemporary art among many others.

Claudia gase is a self taught hacker, who since the late 90s, has gradually become a techno political activists. In the last decade, he worked on whistleblower protection with global leaks, advocated against corporate surveillance and from Alex for algorithms exposed campaign, of which Facebook tracking exports or tracking expose is one of the tools he will talk about today. He's a research associate at UVA, the executive team, and vice president of the Harvard Center for Digital human rights. Nine times out of time is a researcher based in Bangalore working on feminist politics of data and speech. Core a core part of her work is building tools and resources that help situate stakes and digital rights issues for the larger human rights movements. In her talk, she will discuss the role of ethic and Facebook user, the Indian context. On shorter note, I want to say unfortunately, I won't allow can be part of this panel today because of some health issues. He writes in Berlin. But unfortunately, he cannot join us, but we wishing him a good recovery. And now, welcome to Carolyn.

Yeah. Hi, everyone. I'm Caroline Sanders, and I am a machine learning designer, artist and Research Fellow with our university as well as a senior fellow with the Mozilla Foundation. I spent the past seven years studying online harassment networks and systems, as is I'm particularly interested in how language is so contextual, as well as conversations, digital spaces, but how language and conversation can be weaponized as far as that's understood and how it fits inside social networks. And that funding can be from a policy standpoint, how policy relates to language, or technical standpoint, or algorithms to understand or rather look at words or from design standpoint, the things, the ways in which we're able to interact on social networks are things we can post to any kind of image or content. That's fascinating because I think language is both objective and subjective is literal and figurative language is really complex. But language inside of any kind of technical system is a form of data. And I want to highlight an important project in the space when it comes to thinking about data data from an ideological standpoint or data from artistic practice. And I'm gonna highlight neowise work the library's missing datasets, which perfectly captures I think, all the nuances of data, or even the nuances of a lack of data on all rights, missing data sets or blind spots, and otherwise days after it faces. The library busy days that started in 2016 is an ongoing physical repository of these things that have been excluded in society are so much else is collected. And unsurprisingly, this lack of data typically correlate to issues affecting those that are most vulnerable. In this context. The word missing is inherently normative implies both a lack and an audit. Something that does not exist but should, that should be somewhere is not in its expected place. Established system is disrupted by distinct absence.

Just because some type of data doesn't

exist doesn't mean that it's missing. The idea of using data sets is inextricably tied to a more expansive climate of inevitable and routine data collection. And why does this matter? That actually more reveals more than that we get attention to it's in it's a nice things that we find cultural and colloquial pants on Wednesday, two important slots that we left blank reveal our hidden social biases and differences. News project with particular names that you see highlights different kinds of missing data sets. And at times, we can think of where the important to be seen by a system how system relates to you. If you're seeing any of this, there are many other points inside of Mimi's collection of missing data, data where we wouldn't want things captured. What would it be like to have a list of all the mosques in one city? I was very easy to find a collect. You can think of all problems in that. No, this is my I've been thinking a lot about how do we look at data beyond the cold the literal or careful? data can be messy, qualitative, complex, and quite small. conversations are data points in conversations or emotional spaces, especially in digital systems. Our conversations are caught and saved as data as emotional data. As an artist and designer explored the in betweens of conversation, how does emotion is saved in digital realms? How you take something that is inherently qualitative, like online harassment, and distill it into a quantitative data set? How do you take emotional trauma and perceive it into an Excel spreadsheet? And is this possible? Kevin Davis, open jersey, I like talking about subjective data sets, which she describes as data sets create a worldview, we have the ability to shake this worldview through working with us adaptive data, this kind of subjective data that allows us to explore the nuances of data the in between so what it is that can be my shouldn't be. I think this is a great area of data set. So sure, as you do this project I'm working on with the artists and technologists are often when I'm creating an emotional data set and a self portrait algorithm. I'm using this further now as an example a metaphor for the deeper points of this presentation. I want to refer back to the first image of this presentation. This looks just like a random regular highway. It's so mundane and seizures. And to me so completely American and then I was worried about this is this is the entry point for Water, Hurricane Katrina, what are the most devastating natural disasters in the United States? Well, there's a graph for Mississippi in 2005. It completely leveled it completely leveled the towns and the surrounding towns. And to the point where it was mostly left were slabs of concrete houses turned sideways with our system, see this image? What kind of data does this image represent? So I might get through some out of the box, Ai, computer vision demos. And let's look at some of these experiments. What is clarified sort of see, it's a struggle, transportation stack, no person, water airport. So it says some of us are right, and some of them are not right at all. And this is from Microsoft. Again, outdoor beach sky sitting Harbor, there's a train apparently, in here, that's used to me. Also a jet that's I don't see that either m, or D Grade II, which is a view of a city street at nighttime

display.

I thought this was really sunny in the middle of the day. Now he's really happy with me whatsoever, again, with me. And then he's also really, really captured actually, with the image actually is, this isn't, you know, an image of a city taken at night. more deeply, I wonder how we could understand that trauma when systems can't even acknowledge it. I grew up in New Orleans, Louisiana, a town affected by Hurricane Katrina and my entire family for the past two generations is from Gulfport from the surrounding area. And that's the area where Katrina made landfall. They've been photographing my family since 2005. What do you see on the screen is a small slice of these images of this dataset, which is also an archive. And this one, I have a few 1000 images, which is probably enough to train the computer vision model on enough to create some kind of model some kind of system around six main analysis. But there are more emotions here than just the big vines, analysis, which are fear, distress, sadness, joy, anger, because the emotions in this dataset are just a portion of anger, or small potential joy, what I see are his pain in many different shapes of it. But I also see beauty. I see complexities and different flavors of feelings that are extremely hard to put into words, I see something that is a bit hard to quantify, a little impossible to touch on lies. While this may seem unrelated to harassment, I think it's adjacent, it's very metaphorical. a hurricane is just a weather system. Its temperature, its air pressure, its location, its weather data analyzed by chemical system. But the effects of a hurricane are inherently traumatic, and they're extremely humid as online harassment, online harassment, core social media data, its words, its time, its location, but the effects of it are also extremely human. These are earnings within two boxes, they feel impossible to fit into spreadsheets and hard perhaps going into data, but it's full of data. And that's why I'm interested in the emotions that are hard to classify the things that we've sort of fit into what we could call a gray area. And that's a lot of what network explores me see this gray area perfectly in social networks, particularly in their policies for defining harassment, hate speech is that were concrete, what about the different levels of harassment that are less literal, less concrete, more abstract, these are the things that are very hard to find out policy. harassment is often extremely contextual. It can be continuing events are not isolated, and at times if you're just using a specific word, or a specific kind of threat. So let me give you an example. Do some, you know, favorite tweet, jobs harassment? Maybe not in that one isolated event? What if they favor all of your tweets every day? For a week? What about for a year? What if they favorite and favorite the same tweet every day for a year? What if you've asked them to stop? What I've described are many different forms of harassment voting system to be able to recognize this kind of behavior, a person could personally could feel uncomfortable, I'm sure it violated in this in this face. It's hard for technical Sassoon river policy to define what I've described as harassment. And what I've described is also harassment in the data point. It's about interactions, frequency, time location, it's also about the context. In social networks, all this data is caught and captured. Like my hurricane project, though I'm interested in is hard to classify data, this extremely subjective data. This kind of gray area is the area that I work in. It's the first to define a policy but it exists in so many profound ways. And we see this with new emerging terms from Neo fascist and alt right movement. Could we say that magga is hate speech or harassment? Well, what if we take a step back look at the idea of ideology that that word represents

some of the gray area of my researches and put into practice in April 2018 years in right To work with answer and national as a consultant on Twitter, so today we're doing one of the largest of its kind. With a few with a few 1000 volunteers, we have labeled nearly 600,000 tweets on the resume. It used to be that female politicians and journalists in the United Kingdom, what we learned is that black women were disproportionately targeted, that women, regardless of political affiliation, to conservative or liberal were equally targeted with varying levels of harassment and abuse. I was running to Azerbaijan, particularly to help design conceptualize what kind of system we would need to build to label this data. How do you explain these labels to our volunteers? I'm going to structure this especially I was trying to try to label the messiness of human interactions in a way that a system could read. This is sort of like what starts to look like, well, things are designed for this label system where the categories of labels and we started off with problematic adhesive or nine after the publication of our research, career legal policy, trust and safety lead visionary guide to a bridge with this term problematic citing concerns of censorship. And I want to take a moment to offer a rebuttal to this problematic can be a good term that encourages disruptive for different kinds of content, problematic also label and look at the gray areas around harassment. Social media companies do separate different kinds of harassing behavior to different forms or levels of harassment. Facebook does in particular classifying behaviors into different categories that have different citations, reparations and outcomes. Bullying is their term for less severe harassment, and harassment, same term for more severe the use, Instagram does this as well.

And the reason I bring this up is

it's important from a standpoint, especially when it's user led research to think about content and we relate to it beyond have it been thought of as a content takedown sugar recipe only given the length of content moderation, or content associated must be removed. harassment is outside of that harassment is extremely, extremely contextual. I work through some of these ideas around this space of programmatic and the way it relates to the context here, especially through diagrams because behavior is so nuanced. And the sonogram is an example of the kinds of sketches I've made to work through doxxing and releasing public documents, such as real name phone number scrapers, etc. are sorry, dog's name, which is a really simple like documents relating to neighbors and credit card numbers. Is there were antagonizing policies as a part of most companies during service in essence, companies are recognizing that these adopting is actually a form of aggressiveness but are their forms of positive boxing. I can't agree. And I've tried to, I've looked through this. What I find fascinating though, is a space in which we can acknowledge what kinds of behaviors is enemies and all the concepts of how boxing can be misused. So for understanding different kinds of harassment data, such as GamerGate, which is a harassment campaign designed to push women and marginalized groups out of gaming communities, to train analyze large scale catastrophes data is emotionally charged and say social networks data is extremely personal. It's about people in their everyday lives daily. It's just drive to so many different kinds of metaphors, for example, data as oil. And generally I find these networks to be extremely confusing, but there are some technologies here or adds context to this oil metaphor, which I'd like to share. He says, We're oil is composed and compressed bodies of long ago dead micro organisms. personal data is made from the compressed fragments of our personal lives. It's a dense howling state of our human experience. And from this definition, I think data is activated, it's referring to the inherently personal our data inside social networks is a reflection of ourselves that mirror the many varying shades of ourselves the previous lives. We looked at online, the previous likes of ourselves, how data is caught and captured and copy has extreme political and emotional notification is especially exciting social networks.

Thank you.

Hello. When

he says is our source project that I've already outputted which is Facebook tracking score, I think the tool is intended to make you understand I always works behind the scenes. The thesis is that they decide what matter for for you, and what does not climate change them up going to communicate them between a non expert, why they matter. They act in the way that we don't know yet in our society. So how we can make this story compelling, better. Better, because they get some of the advocacy and activism in the USA. And working together is our goal to understand how this political issue can be painted or pencil. But before talking about how Watson is better, take a step back and try to view together how the Facebook algorithm work, we focus on Facebook, because it is the biggest competitive platform we have to deal with. But this approach can be applied that you may have seen, this has been our first

shipment

anything to Ghana, where someone created a contract, it could be you or a doctor, or vectorizer. Someone you already saw this visual main time, you can select the summon feature file, again, it will be public or friends. And that will give it to the left box of facebook, facebook, behind the scenes, some metadata is something you will see you have control of some of them on the on the top of the table, you can see which data you believe to have control like this, because if you write some hate speech, that's a behind the scene that Facebook can play. If you are suddenly in the Facebook system, which maybe will lead to something I don't think so because they are blocking your sample. Awesome, awesome. Awesome. You are completely unaware. For example, if you get some bad news based on the IP address of a place. This is happened to me when I was in San Mateo County. When I was when I was in it the computer was using it was a big, like, they were not a trace of a previous navigation was just repeating accounts. So for this kind of experiment. And the only suggestion that I was getting from a Turkish people. Because I mean, that's it was showing it was mostly using it as Wi Fi

emotions.

We don't know exactly how it works. But we're sure that this is happening. We always because Facebook raise surgeons and that's a great, outrageous experiment that is of massive scale emotional contagion through social networks. Basically an experiment that made our people back Okay, 600,000 people, the split into two groups. The first group will obviously appear in either news feed pasta with positive emotions. And that study if they behave differently, well, these are outrageous, because psycological cannot be done on our Samsung. And he's an amazing business. Maybe this kind of experiment or Afghanistan phase because because these are not shared by anybody on Twitter. She was a friend. Once this friend that went to bed saying offer an offer that you seek to get through the next weekend. And nobody saw this pasta hit either after the surgery they found out later. And this is an example on how can I suffer some feedback if you were deciding

to move into Hanson

Laughter Okay, well, that person messaged, someone logged into Facebook and Facebook, look at the profile that happened behind the scene to understand what should be said to this person. So again, there's a good reason to control that and you have to follow the settings you have the data. outside of your convention, we don't even know which they are. But they ancestry for Facebook, because they keep a profile of people as a

user, so

no data collected. And then finally, Facebook, this is what you're going to see. This is an example I took this screenshot is that a Facebook data is a female up because in the next time we will log in will be different. If you refresh the page samples from your Ma, CDs, or maybe we'll never be again. In this account either wonderful to be fancy. I thought okay, tonight Bayesian analysis wasn't putting it this yesterday and since since 24 hours, and this will settle a little bit for us. So Facebook bigger this, this is what we call content, prioritization, or customization, of content. But at the end of the day, this is a form of our deciding what will appeal to you and what will not what will appear constantly on top of water. Because you already know that you want that. For me, so we can more easily relate to if we compare it with the news with the newspaper, The old media that we were thinking, luckily, we are getting because because of the internet, so we can find that information we want, we are free to publish our own content content that has been beneficial professional interest makers, we saw that was exciting and liberating, because that was the right thing. But it is this we are suffering of is a platform Hmm. Water, you can see an athlete with a form of the news media, we can understand where we are in the New York sector. So if I'm looking at unused media that is a boxer with the knowledge about the past, it is clear that I can understand that better. This question is quality, exactly as it is on the source. And so I can take between the most extreme left the media and the most extreme right media and understand the different position, we understand each other because we can see the other and put ourselves in context. But because the human after nine, if you want to compare what you're getting on the ways of sharing your mobile phone to someone else

is a bit difficult.

And anyway, imagining that we can prove that because this form of our superior graphically, we are society literally ready to do what is conveyed on average is the right tablet can exist, the right type of reason, consistently fails to operate. The same way also to my listener can claim to be objective, whether the result was a big round of us our recession, so we will experiment with a copy of water that people see on the timeline. We do a copy of what Facebook is we we don't actually don't even get to

the goals.

And that's we can often visitation because it will be the case. We try not to compare between profile. But if I compare my profile of course it will be so different because we came from different trees on the water. So I want to come back to the comparison need to have somebody or I want six providing this experiment. We're following the same pages, the Web TV page that we're accessing in the same time of the day, struggling to capture the film out of Boston exposed to a lot of information, their co founder and they just like different pages of content And each one of three times media, and we're about to use the news media as a subgroup. So that bottom number one was liking it. Number four, and number five is on to understand how that works, and where to see what Enter, and what goes out what enter, you can see that if you look at the page of the news media, this is what they produce this amount of content. But this is not a professional. This is because they were behaving differently. There is already a group of just more people in Africa on Facebook, and that mean that that is that the input on the records.

Now,

you have to go back, this is a robot who here she or it works on Amazon. So the best thing to do is if metamod the filter bubble, or if Facebook is a fair place in business, which Facebook is a faith based business is that our 60s, so in the same person things, these media because they will be treated fairly, and they use are following the patient, we will be neutrally silent. If they are constantly fed, this is only the first one that we only see a blue bar before we see a yellow bar. And this is we see that green bar now, we're not gonna last as well as your answer. Nobody, I can't do this better yourself. Ready? What it is, is that nobody won. Because no question system users Saturday and she's on someone else. Let's see what this means is data in the box. Number one, selecting content from Rwanda, we see the process that is through your bot is getting more. Yes, the second and the third that there were other bots, they were not liking anything of these three media. And the fourth one is the one I think the yellow button was why it was so the one thinking out of the green. The fifth one is the one in the green, which by the way, is the one that was making the most effort to do this one, it was producing more content, but at the end of the day seems to be the least the most mistreated, again in the 65 that will be competing, that is a way to do some accountability. checking how the input is deleted, I understand that what is happening is not the device, we can never understand all the metadata that Facebook is using, or the logic behind. Understand that they have made this in mind. The point is that we're looking at the name but I mean, you should decide that what you get because the perception and the selection and the creation of content as a huge influence on that on the message. And who decided my reality. In theory, I want to say my only I cannot water matter and maker indeed well capable to run. But it is monitored. It's a bit it's a bit different. I'm changing the goal again, horrible things from focusing on helping you find the relevant content to help you have more meaningful social interaction when I use it on Sundays, because it's a Facebook. Anyway, he said we live in the debate and Facebook is happy to play in that they will change the algorithm to make it better conversation but only I can know what is a good conversation for me because I changed my time I want to explore different things or I can just get a taste for my family. But it is pointless to enable on your search because again, it will be really in our ivory tower saying Facebook is claiming to be the same as that with Bella Vista and the loser on our making it unfair to the government. So in this contest, because these That's also an issue. And we should understand that our society and how can we study later it and why this has an impact on us. And that's why I love to watch people that has a Facebook account, and that we are making progress in implementing our platform is our browser extension that you can start on Chrome or Firefox. And this website, I mean, you can Swamis, your friend, thank you please. What happened and you're easily connected, not that it was shattered the way that they want not to wash, I think when you get a link on top that kind of leads to your day. And you can see, personally, this is a copy of what you get water, we are processing and because we send it to our server, and these are any mean, it's not.

Because you have access, but then you can start to test with your friend who's getting different content. And this is an example on how to different the boss, the boss of the experimental before, we're receiving information on our SSL packet, and the theory, mousetrap anything. I mean, the Venn diagram is quite clear to show that they're true, then we're getting information, some of them are more samples. And another profile is getting the user minority of small amount of information on this specific subject. You can download the CSV of the day that collected and we are working on it to let you pay your friends and your group of ally, colleagues, students to test because we cannot expect that we get to you how we should get involved is only to peer review of people understand the complex issue, understand what I'm missing, and the discussion that maybe you want to believe but not the best from you. When the sun confess that you should only see these marquee comparisons and understand the matrix matrix will judge the concept that we lack. This example of the column are one of the sixth in the room, they are the eighth. And then we have the art of photo boxes that were appearing on the timeline. So you'll see that on the first column, you got to Boston for the 52nd percentile and the fault of the Technical Center in the same place, the second profile photo that says I'm waiting, understand how someone else is getting any better information. That is we got it. That's pretty outstanding that you made You're the one getting a picture of the far right.

And when we brought

in addition, because of the publications being made out of this collection. Anyway, this is facing some challenges. Guys, mostly one is legal and technological. And some further down is a book published by an investor that talks about 40 of female finisher, rich fish for structure. They get pregnant by Spotify, because they were testing in a similar way. By the way guys on Facebook, Spotify also send the IMA to the equivalent of the Minister of Education saying about your research researcher and doing something bad Not only this illusion that was on the side of this. Because if we accept that the feminine condition for me, the two of us understand what's going on. We cannot expect us to sit on the bench

and the others

because I mean compared to our team and that is composed by less than 10 people that our face will be damaged. And that is a story from propublica. They were running with a fancy tool that was what was going on with Isaac.

processing.

This has been our stellar survival challenge. Our goal is that everybody should change it when they want ever customize it tested and implemented your own policy in this way. We cannot ask and everybody will be able to live operating system can damage your values.

Yes. Yeah.

Yeah, yeah, sure. The next

few weeks, West County, not only less than better, was the country to be concerned about on highway two mobilization, we see that our attention hasn't ended. So because we feel like we can be more smarter, is the right time to run a campaign campaign. To citizen I'd like you to inspect the work session and to participate in this political experiment. I like the goals. And it also because it explain the second mutation that we evolved in ourself to protect your data in the present, because they all again, is a starting phenomena, I believe, or propaganda, not individual behavior. A unity is a campaign and bingo site. To do this, and publish. This is part of the aleksa project, it was funding from the European Research Council. And that's a capital asset to ramp up money don't have to do business on on data, beside the ethical levels on data on test results, and how can we use that to understand phenomena but not at the fiber guide? Okay, those are four. Let's see which of them they are. So to scientist, they are the most interesting, because normally, they want to test how the algorithm has an impact on society. But we want to be fooled by others to enable every Facebook user on the family session of the day, because this was a position I made by this fellow activist is not the best way to communicate, maybe you can figure out something better. Because they believe to be disintermediated. But actually, they're not. So those are Google Sites. Even Athena is already online for Google services, software. And I suppose that will be the academic front end of the project, Facebook, so Facebook site and start to be the two years ago, and I will became the technical reference. And that's is the time etc. Because we can claim that we can claim that we can use Facebook, and he did that he can do for you. We can also believe Facebook is there because we are going to experiment. I mean that you are above average age is over 50. And the generation that you face included will keep voting and be funded through Facebook. And that is why civilization we can't imagine that Facebook Google officially our soon. That is on that. Thank you.

Okay, it's all in German and attention. Hi, everyone. Thanks for coming to the talk. And thanks to transmedia and the organizers of this panel for having me. So if we could talk a bit about the spread, okay for content in Facebook in India, and vocabulary that Facebook has to deal with the same content and what kinds of stuff actually falls through the cracks of Facebook so heavily. Now there's a large there's a really high volume of hateful content on Facebook, a lot of incendiary posts that circulate on social media in India. And this creates an atmosphere of intimidation and violence so many communities because the content is the sample big cast is sexist in test the consequences. Now, this happens in all kinds of different social media platforms and there are quite a few which are popular like WhatsApp share, chat, etc. But for the purposes of this presentation, I'm going to focus on Facebook and my understanding of Canada and Hindi language, Facebook groups. Now, there's a wide prevalence of Hindu supremacist content, which is rife with use of signs and symbols that assume meanings that might not be evident at first glance. So jumping straight to an example, I try and unpack this site, let's say I'm exposed to this one is from a public group, and it's in Hindi. And I'm going to go into a lot of context of these they will be useful to understand this. So the image is of shows a hand holding a dagger as well as a religious object. That is one usually around the neck. But here, it's on the wrist. And the text within the image basically says mala in one hand, and dagger, on the other hand, that's who is the caretaker of hindutva either going to face literally meaning belongingness to Hinduism, and represents a sort of nationalistic and often supremacist ideology that is, the last couple of years being currency. So this post arguably talks a violent kind of position for forwarding the purpose of hindutva. And if you look at the text that accompanies the image, it actually goes into mythological characters and incidents of violence and fast and almost a call to action. So what this is, with incidents that I'm referring to, is an really long, ongoing conflict in part of North India, or you hear about the building of his temple. And so there's been like for decades, there's been a demand for a new temple for godhra because it is believed to be his birthplace. And you could argue that once upon a time, there was such a temple. As part of these demands in the Europe 90 to the 16th century, most was raised them was demolished by internationalists leading to rights across the country with more than 2000 people who came back to the page. Now the text accompanying this image says that heroes will be asked for proof, in this case, the existence of a temple under the day that Hindus remain wrong, in which case we're noting in this case, or noting someone who comes in place. And the data used to come around, which is another factor in Hindu mythology, who's famous for having a terrifying temper. Even an offspring of a bird, which and Jessica, in this case, is a good ruler in 16th century and also also King. And Joshua means glory to long run. So basically, the last text that we see in separate here is Joshua, it's also what has been commented by the different users. Now this is clearly a post that at least indirectly incites violence upon certain communities.

So my point in like explaining all these different things with meaning was to show how something innocuous seeming in different contexts like a phrase gesture down, or the hashtag, or color saffron assume different meanings that might not be easy for, like even people to grasp if you're not steeped in the same kind of media and political climate from where it comes from. And that brings me to another second more implicit example where you see to go super, rather disturbing trend, where children's images are used along with some specific symbolism that we encountered in the previous site. So in first image, there's a second flag which is attracted to Rashtriya swayamsevak Sangh, which is the RSS, which is an umbrella group for penguin nationalists. It is also own, which you might have encountered in like spiritual or religious settings, but with a totally different type of provocation is image. In the second image, I see a lot of the red flag images, which are commonly appropriated because it's the closest and saffron flag, and basically the show loyalties and because of internationalism. Now, I think we've theorized about the use of children's bodies as a vehicle through which equal message mixed messages spread, perhaps appealing to some, I don't know insane, non violence like people and making these messages maybe more palatable in some way. But basically, this is also happening in the background of concrete events like it's happening in the background of people being lynched for eating beef at 26th anniversary rally of the same demolition, which led to large Communities fleeing from the back and sitting down for a couple of days during the rally for fear of violence. Yeah, so in in these images, there's a coming together for a lot of the symbolic elements. So for example, the color saffron and the mark of loyalty and forehead and like the flag all of that. And my point in showing this post is that the images, really rich in context and often communicated through elements that are steeped in shifting and even contested meanings. Okay, so then that brings us to Facebook. So what is the vocabulary that Facebook has to understand content that might be problematic, right? So Facebook's categories like hate speech, terrorism, etc. So these are categories that Facebook has for you to be able to report to particular post. But the baby kitchen previous segment actually not make the threshold to qualify as hate speech. And bad imagery posts rarely do because they're saying symbols, hashtags, whatnot that encourage a particular kind of imaginary, but might not actually be calling for violence, like in first image. Search posts often become wider, and the categories that Facebook has is not meaningfully capture the impact of this kind of larger ambient media climate that is created. Although it is very much based within contract events in like a particular kind of politics public sphere. What I'm trying to suggest is that, if any platform benefits from virality, and if that viral content is harming particular communities, the platform should arguably be accountable for what codecs algorithm incentivizes. So it's not really about like whether one individual post qualifies as

has its

own Facebook's categories of problematics. And their inability to address the issue of the harmful media climate. Let me think of photographs words that masters tools do not dismantle the Masters house. So yes, platforms like the regtech sorry. So platforms like Facebook also construct the idea of privacy or safety as an individualized idea of control what you share, as opposed to understand what gets shared in a more like collective sense. What is it said for time supposes incentivize that differ depending on different situations, language, etc. But essentially, the idea in the coming months. The idea in the coming months as the general elections in India approaches is to also use Facebook tracking, I suppose that Korea was speaking about to understand the algorithm a bit better and yeah.

So

thank you, the three of you for your very compelling inputs and presentation, I have a lot of questions already that maybe we could start with the question of categories and contextualization. So

because

what also is a problem is how Facebook is using categories and how they also fail. So it's like categories that are pre set up that we just saw my entire presentation, but also content moderation as human beings, so maybe it will be nice to the both of you this question. So maybe you could explain that from your perspective and your research. How how, like, what are the boundaries? What are the many devices?

Sure, so I'm going to be looking at Facebook as a particular use case and I think propublica did a lot of really great reporting this specifically I caught the moderation as well as a documentary called the cleaners. There are different kinds of ways in which you have to moderators have to understand their content and they have to react to the policy very quick way so from Pro publica has research. There's this kind of like mathematical equation moderators have to go through is what's being said? And also, what is what's being like, what is being written? Who's that directed to? And where does that class of person fall. In this case, it was something where the group of white men were more protected group than per se black children based off this strange mathematical operation they come up with, I think the deeper problem is that other than that actually being extremely problematic equation and system is that constant motors have only a couple of seconds to make any kind of quick decision as to what they're looking at. But more importantly, the tools that they use, lack so much context, so they're looking at things that are incredibly isolated. So in in something I was trying to make a point anyway, in my presentation, what happens if you are receiving continuing harassment? So if someone is liking liking something, boop, you're gonna have a year? Well, you could get 365 different content moderators looking at that. So they wouldn't be able to see the longer term contract, what they're seeing is everything, every individual say level. And I think that's true. If you look at hate speech inside of our weeds, for example, they're seeing individual pieces of content, not the entire trend of it. That's because when the content is recorded, is going to individual moderators, right, it's not going as a collective whole.

Yeah, I think also, apart from having to understand context, which itself is such a huge, seems to be such a big issue is also like getting rid of this sharer of neutrality that a lot of platforms have. And this really reminds me of recently, when jack Dorsey was the Twitter person was in India, and he was photographed holding a banner called snatch the American patriot. And wherever photograph was in Twitter, and essentially he, there was a lot of pressure from different kinds of groups. And he they issued an official apology for bullying that which many of us are here, there's nothing wrong with it, like potentially unsafe box also, I think there is this issue of even even like saying with categories are not going to next second thing any reason not not I think that separate self is not there. Because in India, I think hate speech, for example, does not recognize the basis of gender, sexual minorities, a bunch of things that does not include class, which is a huge missing piece. But yeah,

mostly takes the best heavy in a culture and change with that

theory. Because the moderator is the person who is giving this whole community and why should I be someone I select? question will be someone paid by Facebook, but outside of Facebook and also employment is we can assume that you can have better being protected by someone who shares your political or local values without be subjected to the fact that Facebook invest in a country only when they are profitable. And only after outrage, like the leads in as of the last year, they start to hire new people who are looking at that language is older, wiser, that will be that slashing the price. But the decision to be sustained, super committed to having a sensation of the diversity of the world because otherwise it will be only based on the first languages.

Okay, and what I would be interested in for Carolina's any workshop yesterday, we talked about how to build a feminist data set. So I'm really curious, like, what are the inherent problems or difficulties you have any sort of flat open a black box open data set? And then yeah, that could be like a very good way of approaching, say, a feminist, they're just not

sure, I think that there's a lot to unpack there. And if we sort of look back at the nightmarish resolution, for example, how do we create datasets that are highly contextual or that are, I guess, highly charged, so hate speech, for example, hate speech, shivers, culture and culture and diverse place to place and so partially in my workshops, we also talked about hate speech symbols specifically from the American Alright, so being able to understand the nuances of a conversation is extremely important. I think that's also I think that's something we're all touching on today. I was gonna miss Davis. I was in reaction to the research. I was doing American alright. By building your Nazi data set by analyzing different piece features ever using jurors, you can imagine looking at new Nazi data fucking sucks, like, for lack of a better phrase, but it's a lot and I was interested in thinking about what is in artistic practices or in France. This this growing fascist ideology, this sort of space and inequality, and granted from an American standpoint, right, I'm American. And a lot of the data that I look at I love the different groups that I'm in culturally absorbed end up in the American filter bubble. But a lot of them these days, I was also thinking about, what does it mean to be seen inside of the data set? There are a lot of data if we think that it's hard to find it's written works and the campaign or if it was a monkey over the phone, sort of In a similar vein of you can't be you can't see. So how do you create spaces of easy to find goodness data famous artists and practitioners or feminists text in a blank space? That's easy to find? How do you create effectively, more feminist archives? So feminists days that sort of comes from that point of trying to collect an aggregate a lot of a lot of this, how it's used as a data sets are for those that choose to engage the data set that she's using around the planet is something else.

I will be also interested in the question, what would it mean to intervene at the level of the data

and the level of the algorithm? So

maybe?

We talked about that before. Maybe you could just yeah, make the distinction between maybe media theory which is more interested in it and you said like, practically it doesn't really matter, but it will be interesting to Yeah. Feel free the others also to answer that because stone is just kind of life. So

prior to the panel, your role sort of huddling since we're missing, sadly, a co panelists and one of the questions you brought up