Behind the AI Curtain: A US Data Labeler’s Journey Since 2008

Digital Dominoes. Hello and welcome to another episode. I'm really excited again. I have a data worker, Krista Pawloski, and she's an activist with Turkopticon Krista is in Michigan in the us, which I find really interesting because I've only met data workers from Africa, so I'm really excited to find out what's the same and what's.

Different. Thanks so much for for talking today, Krista. Thanks for having me. I'm very excited to get the word out about what data workers go through in the us. What's really interesting is how long you've been doing this. I think a lot of people haven't even heard of Amazon for as long as you've been a data worker.

And at that point I ended up doing the work full-time. And through the course of all those years, the type of work has changed drastically, but it's been a wild journey and. It's often been completely different from one day to the next. So even though I've been doing it for a very long time, I don't know that my experience is much different than somebody who's only been doing it for a few months, because every day is so different.

Is that stressful to have, just not know when the day starts, what's gonna come? Yes. That's one of the hardest parts of the job because you're only gonna get paid for the. Exact moments that you're working and if there isn't any workup, it doesn't matter if you need money or not, there just isn't workup, there aren't tasks to do.

So I. You know, if I sit down and I make all my money in five hours, I have a really short, really successful day. If I sit there for 12 hours and I don't make my daily goal, then I worked longer, but I have a less productive day. It, it can be very frustrating. What influences, like whether you're gonna make money fast or slow.

'cause a lot of people don't understand the work very well at all. I mean, I was looking into this Amazon mechanical tur and I saw, uh, some lists of job and some are 1 cent per task. Is it related like to the amount of complexity or the kind of task? I mean, how, how is it, I mean, just to give a better idea, like what kind of work it is that you do that would make such a big difference.

So I've been doing this long enough that I have a lot of specialized qualifications on the platform, so I am able to be very picky. I won't take anything at pays less than $15 an hour. That's not common. Most of the people working on this platform are taking very, very low planning work. Now the jobs that you said you were seeing aren't necessarily low pay, because a lot of times if something's only paying 1 cent or 2 cent or 3 cent, it's just something you're gonna do real quick and click a button real quick and then you're onto the next one.

Unfortunately, Amazon is really good at loopholes and we are not considered workers or employees or anything like that, so a lot of times these platforms, because there are platforms outside of Amazon as well, but they'll call us participants. Clients, they will not call us workers. They never call us workers, so we don't fall into that category.

Training AI stuff and make it sound like, you know, I've got this big tech heavy job. But then, you know, people are like, where do you work? And I'm like, Amazon Mechanical Dirk, and nobody knows what that is. Or if I wanna go to another platform and work on that platform, I can't bring my qualifications with me.

So I start back at zero again. So I'll be one of those bottom of the barrel low wage earners. Wow. Yeah, and I, I've actually heard someone else say that it was a data worker in Kenya saying that, you know, she had trouble helping, she was an activist as well, you know, helping people to get out of the abusive, you know, data work that they, they have there often, and that as the problem that the people come out and don't actually technically have qualifications.

I really don't like them, so I avoid them when I can, but people don't even consider it a job. It's like, well, I've been paying all my bills somehow. Wow. And that's kind of hard to even wrap my mind around. I don't know that anybody knows how many there are because it's a job that's often called ghost work, like your ghost workers.

And I think the efforts are quite intentional to make you unknown. Wow. So what are your thoughts on that? Like on the, the hidden part and, and the fact that there's so many people doing it. It's actually having a huge impact on society in general because of tech is advancing because of it, but it's just hidden.

If you knew that your AI is driven by somebody who's not being treated well. Just like the sweatshops that make our clothing, you might not wanna purchase that brand. You might wanna look for an AI that was made based off of well-paid work. They keep us in the dark and make people think that all this stuff is just magic software, and they don't realize that there is actually.

A little man behind the curtain, you know, like the Wizard of Oz. I often compare it that way. Yeah. The, the Wizard of Oz is just, we're we're the little man behind the curtain. I found out about this, the existence of Jada Labelers around six months ago. So I'm one of these people that like, had no idea and I was so shocked.

It's just a bunch of people, like one example, the the security cameras in in European supermarkets, and there's just a bunch of people in Madagascar in a room watching cameras. Or even Amazon, right? There was the Amazon Go and there were actually workers in India that were pretending to be this, the ai.

Why? Because this is really important. So one of the biggest concerns with AI in the medical field, I'm sure you've heard the term hallucinating and AI can hallucinate when it can't answer something. Do you want to get medical advice from something that just makes up an answer when it doesn't have the right one?

I don't, but also because the way that. This labor force is treated and the small amount of information we're given when we're working on something means that even the best intentioned person is feeding bias into the machine. So if you are gonna go to the doctor, you might be dealing with an AI that was largely programmed by somebody from a different ethnic background that won't understand.

I know from at least one person is talking about having to label some images and if there was cancer there who's like, I don't know anything about cancer. And so I think this is also a common thing, right? It is. I do these jobs where I rate surgeons that use the little robot arms, the da Vinci machines.

I have no medical background. I'm rating a surgeon who has, I mean, I don't even know how many years of college it takes to be a surgeon, but I have nothing. I have no medical background and I'm rating this guy, it's completely inappropriate. They'll give you little training pictures like, this is good, this is bad.

That's another hardship when you do this work is you don't know what your data's going for. You could be working on some great project, or you could be working on some horror of humanity. I. Wow. I guess that's hard to feel fulfilled, right? Because if you don't know, and that's important, right? To feel in life, to be doing something, you know what it's gonna be for.

Is that a challenge as well? In the beginning, it was exciting because when all of this stuff, especially with AI for Circ coming up. Back then, before all the problems started surfacing, it was like, oh, this is exciting. You know, when you say, okay, Google, or Hey Alexa, that was hours and hours and hours of me recording myself saying that, and you know, hundreds of thousands of others.

It's easy to say that that's for your ring camera, but it's also used by. Police departments and security and surveillance in various ways. Talking about my phone, I activated my phone. That was great. Alexa's like, yes, how can I help you? That was perfect. That's perfect. I just recently read a book by Joy Bini.

So what you're saying. I guess this could also have to do with the fact that in training there was some base facial recognition tool that everyone was kind of building on and that maybe the tech companies weren't careful to get diverse data labeler workforce, and could that be a reason why that these kind of things happen?

Very much. I did a lot of facial recognition where we did not have any training. You'd just be shown a picture that you were supposed to guess the age and race and gender of this person and. They didn't ask any background on me, so they didn't know where the data was coming from. So I could spend the whole day programming or looking at pictures of Asian people, and they don't know that it's a white person doing it that maybe can't recognize an Asian face as well as an Asian person could.

I know that in your, your activist role, you have a lot of experience also with data workers around the world. Over the years, I don't know how long you've been aware of like the, the international like workforce in this. Have you seen it develop more like outside of the us? I'm just wondering if a lot of these, like bias that AI models are not aware of, you know, dark skinned different cultures could have to do with the fact that the, the data labeling profession at the beginning was really focused maybe in the US and it slowly expanded out.

But I don't know when, I think it was a slow and unnoticed shift at first, so I, I can't really put like a specific timeframe or date on when it happened. Okay. I mean, this would be interesting. I heard somebody say it. His name is Antonio. Kasi. He's a professor in France, in Paris. I don't remember where he said yesterday.

They think that the jobs just aren't there anymore. Is that your experience as well? I think that a lot of it is out of sight, out of mind. If you can look at your neighbor down the street and say, this person is being hurt by this thing, you're gonna fight about that thing. It's your neighbor that's being hurt.

If they're halfway across the country, it's easier to just say, oh, it's so bad for them. Hopefully they're organizing and doing something over there. I think companies are doing it so that they can be put out of sight and out of mind, and they don't have to put all of the expenses into humanitarian concerns and mental health concerns and ethical concerns.

A requester can say that they don't like your work and not pay you, and you have no recourse as a worker. Amazon won't get involved. We have had cases that we've brought to Amazon Turk. Opton has, where Amazon was like, oh yes, the requester should not have done that and still would not overturn the rejections.

Is there anything that can stop that? So that's actually how Tur Coptic Con started. It started as a way for workers to come together. It was started by Dr. Ani. She's a professor, but at the time was a, I believe, a PhD student working on her doctorate. I don't know a whole history. That was before my time with Turk Gun, but.

I mean, the requesters can do whatever they want and not only when they reject your work does it hurt you, but they get to keep your data. So yes, it absolutely would be scammers galore if we didn't talk to each other as workers. So as a third party outside the platform needs to kind of police the platform.

'cause the platform is like, I don't know what they think their role is. Yeah, pretty much. How has it been like, um, trying to get changes pushed through? Amazon likes to give us a PR person that we can talk to and it's. Basically akin to screaming into the void. We very rarely get responses from them. We have been able to have a couple meetings with them and they like listen to our ideas and they're like, oh yeah, we're gonna take this to the team, and then nothing ever changes.

Then you have the supply chain. I mean the, the morals. People say, no, I don't want people to be treated that way. They'll make the decision not to buy clothes if they know that there's a sweat shop in the supply chain. And there's the pressure from the public. So the idea is, you know, the more you talk about it, the more people will think, oh, this is the same with data workers actually.

And even if they knew that, they're like. Everywhere. You know, even throughout the US, probably every country, I assume I'm not an expert in these things. How can the average person help people in your profession? I'll call that profession. 'cause that's what it is, even though you can't always say it.

If we are vocal about our displeasure with the unethical behaviors, eventually they're going to have to listen. If enough of us speak up about, Hey, this is unethical and I don't like it, eventually they will have to listen. I think so too. And slowly this grassroots movement is, is coming, you know, that makes sense.

Yeah, that could be a whole different podcast. Excited. We can talk for another half an hour about labor laws and thank you so much for for talking 'cause I know it really isn't. Easy just for people to be aware. There's no rules, right? You're not supposed to speak up and you're doing it anyway, which is very courageous.

Thank you. Yeah. Like you said, the rules, we're not supposed to talk out about this, and if they take my account away, there's nothing I can do about it. But somebody has to start somewhere. Somebody has to start the conversation so people are aware. And I really appreciate you taking the time to let me tell my side of the story.

Yeah, thank you too. And I really hope that the awareness really spreads as fast as possible. Amazing. Thank you. Thank you. And that brings us to the end of this episode of Digital Dominoes. We hope you've enjoyed learning about another piece of the puzzle that makes up the vast and complex digital world.