首页AI 新闻
These Prisoners Are Training AI
3

These Prisoners Are Training AI

Mcmuffin
Mcmuffin
2023-09-13 16:20

In high-wage Finland, where clickworkers are rare, one company has discovered a novel labor force—prisoners.

ACROSS A STERILE white table in a windowless room, I’m introduced to a woman in her forties. She has a square jaw and blonde hair that has been pulled back from her face with a baby-blue scrunchie. “The girls call me Marmalade,” she says, inviting me to use her prison nickname. Early on a Wednesday morning, Marmalade is here, in a Finnish prison, to demonstrate a new type of prison labor.

The table is bare except for a small plastic bottle of water and an HP laptop. During three-hour shifts, for which she’s paid €1.54 ($1.67) an hour, the laptop is programmed to show Marmalade short chunks of text about real estate and then ask her yes or no questions about what she’s just read. One question asks: “is the previous paragraph referring to a real estate decision, rather than an application?”

“It’s a little boring,” Marmalade shrugs. She’s also not entirely sure of the purpose of this exercise. Maybe she is helping to create a customer service chatbot, she muses.

In fact, she is training a large language model owned by Metroc, a Finnish startup that has created a search engine designed to help construction companies find newly approved building projects. To do that, Metroc needs data labelers to help its models understand clues from news articles and municipality documents about upcoming building projects. The AI has to be able to tell the difference between a hospital project that has already commissioned an architect or a window fitter, for example, and projects that might still be hiring.

Around the world, millions of so-called “clickworkers” train artificial intelligence models, teaching machines the difference between pedestrians and palm trees, or what combination of words describe violence or sexual abuse. Usually these workers are stationed in the global south, where wages are cheap. OpenAI, for example, uses an outsourcing firm that employs clickworkers in Kenya, Uganda, and India. That arrangement works for American companies, operating in the world’s most widely spoken language, English. But there are not a lot of people in the global south who speak Finnish.

That’s why Metroc turned to prison labor. The company gets cheap, Finnish-speaking workers, while the prison system can offer inmates employment that, it says, prepares them for the digital world of work after their release. Using prisoners to train AI creates uneasy parallels with the kind of low-paid and sometimes exploitive labor that has often existed downstream in technology. But in Finland, the project has received widespread support.

“There's this global idea of what data labor is. And then there's what happens in Finland, which is very different if you look at it closely,” says Tuukka Lehtiniemi, a researcher at the University of Helsinki, who has been studying data labor in Finnish prisons.

For four months, Marmalade has lived here, in Hämeenlinna prison. The building is modern, with big windows. Colorful artwork tries to enforce a sense of cheeriness on otherwise empty corridors. If it wasn’t for the heavy gray security doors blocking every entry and exit, these rooms could easily belong to a particularly soulless school or university complex.

Finland might be famous for its open prisons—where inmates can work or study in nearby towns—but this is not one of them. Instead, Hämeenlinna is the country’s highest-security institution housing exclusively female inmates. Marmalade has been sentenced to six years. Under privacy rules set by the prison, WIRED is not able to publish Marmalade’s real name, exact age, or any other information that could be used to identify her. But in a country where prisoners serving life terms can apply to be released after 12 years, six years is a heavy sentence. And like the other 100 inmates who live here, she is not allowed to leave.

When Marmalade first arrived, she would watch the other women get up and go to work each morning: they could volunteer to clean, do laundry, or sew their own clothes. And for a six hour shift, they would receive roughly €6 ($6.50). But Marmalade couldn’t bear to take part. “I would find it very tiring,” she says. Instead she was spending long stretches of time in her cell. When a prison counselor suggested she try “AI work,” the short, three-hour shifts appealed to her, and the money was better than nothing. “Even though it’s not a lot, it’s better than staying in the cell,” she says” She’s only done three shifts so far, but already she feels a sense of achievement.

This is one of three Finnish prisons where inmates can volunteer to earn money through data labor. In each one, there are three laptops set up for inmates to take part in this AI work. There are no targets. Inmates are paid by the hour, not by their work’s speed or quality. In Hämeenlinna, around 20 inmates have tried it out, says Minna Inkinen, a prison work instructor, with cropped red hair, who sits alongside Marmalade as we talk. “Some definitely like it more than others”. When I arrive at the prison on a Wednesday morning, the sewing room is already busy. Inmates are huddled over sewing machines or conferring in pairs over mounds of fabric. But the small room where the AI work takes place is entirely empty until Marmalade arrives. There are only three inmates in total who regularly volunteer for AI shifts, Inkinen says, explaining that the other two are currently in court. “I would prefer to do it in a group,” says Marmalade, adding that she keeps the door open so she can chat with the people sewing next door, in between answering questions.

Those questions have been manually written in an office 100 kilometers south of the prison, in a slick Helsinki coworking space. Here, I meet Metroc’s tall and boyish founder and CEO, Jussi Virnala. He leads me to a stiflingly hot phone booth, past a row of indoor swings, a pool table, and a series of men in suits. It’s an exciting week, he explains, with a grin. The company has just announced a €2 million ($2.1 million) funding round which he plans to use to expand across the Nordics. The investors he spoke with were intrigued by the company’s connection to Finland’s prisons, he says. “Everyone was just interested in and excited about what an innovative way to do it,” says Virnala. “I think it’s been really valuable product-wise.”

It was Virnala’s idea to turn to the prisons for labor. The company needed native Finnish speakers to help improve its large language model’s understanding of the construction-specific language. But in a high-wage economy like Finland, finding those data laborers was difficult. The Finnish welfare system’s generous unemployment benefits leaves little incentive for Finns to sign up to low-wage clickwork platforms like Amazon’s Mechanical Turk. “Mechanical Turk didn’t have many Finnish-language workers,” says Virnala. At the same time, he adds, automatic translation tools are still no good at Finnish, a language with only 5 million native speakers.

When Virnala pitched his idea to Pia Puolakka, head of the Smart Prison Project at Finland’s prison and probation agency, she was instantly interested, she says. Before the pandemic, another Finnish tech company called Vainu had been using prisoners for data labor. But Vainu abruptly pulled out after a disagreement between cofounders prompted Tuomas Rasila, who had been in charge of the project, to leave the company.

By the time Virnala approached her with his proposal in 2022, Puolakka was eager to resurrect the AI work. Her job is to try and make the relationship between Finnish prisons and the internet more closely resemble the increasingly digital outside world. So far, she has been installing laptops in individual cells so inmates can browse a restricted list of websites and apply for permission to make video calls. She considers data labor just another part of that mission.

The aim is not to replace traditional prison labor, such as making road signs or gardening. It’s about giving prisoners more variety. Data labeling can only be done in three-hour shifts. “It might be tiring to do this eight hours a day, only this type of work,” she says, adding that it would be nice if inmates did the data labeling alongside other types of prison labor. “This type of work is the future, and if we want to prepare prisoners for life outside prison, a life without crime, these types of skills might be at least as important as the traditional work types that prisons provide,” she says.

But how much data labeling offers inmates skills that are transferable to work after prison is unclear. Tuomas Rasila, the now estranged cofounder of Vainu, who managed the prison project there for a year, admits he has no evidence of this; the project wasn’t running for long enough to collect it, he says. “I think asking people, who might feel outside of society, to train the most high-tech aspect of a modern society is an empowering idea.”

However, others consider this new form of prison labor part of a problematic rush for cheap labor that underpins the AI revolution. “The narrative that we are moving towards a fully automated society that is more convenient and more efficient tends to obscure the fact that there are actual human people powering a lot of these systems,” says Amos Toh, a senior researcher focusing on artificial intelligence at Human Rights Watch.

For Toh, the accelerating search for so-called clickworkers has created a trend where companies are increasingly turning to groups of people who have few other options: refugees, populations in countries gripped by economic crisis—and now prisoners.

“This dynamic is a deeply familiar one,” says Toh. “What we are seeing here is part of a broader phenomenon where the labor behind building tech is being outsourced to workers that toil in potentially exploitative working conditions.”

Toh is also skeptical about whether data labor can help inmates build digital skills. “There are many ways in which people in prison can advance themselves, like getting certificates and taking part in advanced education,” he says. “But I'm skeptical about whether doing data labeling for a company at one euro per hour will lead to meaningful advancement.” Hämeenlinna prison does offer inmates online courses in AI, but Marmalade sits blank-faced as staff try to explain its benefits.

By the time I meet Lehtiniemi, the researcher from Helsinki University, I’m feeling torn about the merits of the prison project. Traveling straight from the prison, where women worked for €1.54 an hour, to Metroc’s offices, where the company was celebrating a €20 million funding round, felt jarring. In a café, opposite the grand, domed Helsinki cathedral, Lehtiniemi patiently listens to me describe that feeling.

But Lehtiniemi’s own interviews with inmates have given him a different view—he’s generally positive about the project. On my point about pay disparity, he argues this is not an ordinary workforce in mainstream society. These people are in prison. “Comparing the money I get as a researcher and what the prisoner gets for their prison labor, it doesn't make sense,” he says. “The only negative thing I’ve heard has been that there’s not enough of this work. Only a few people can do it,” he says, referring to the limit of three laptops per prison.

“When we think about data labor, we tend to think about Mechanical Turk, people in the global south or the rural US,” he says. But for him, this is a distinct local version of data labor, which comes with a twist that benefits society. It’s giving prisoners cognitively stimulating work—compared to other prison labor options—while also representing the Finnish language in the AI revolution.

Without this kind of initiative, Lehtiniemi worries that non-English languages are being locked out of this next generation of technology. Smart speakers still struggle to understand Finnish dialects. “Not all Finnish people speak English very well, so there's a need for these local forms of data labeling as well,” Lehtiniemi says. Metroc isn’t the only company that has been forced to get creative about finding Finnish data labor. In 2011, the national library created a game to incentivize volunteers to help digitize its archive. In 2020, broadcaster YLE teamed up with Helsinki University and the state development company VAKE to ask volunteers to donate recordings of them speaking Finnish.

There is a sense in Finland that the prison project is just the beginning. Some are worried it could set a precedent that could introduce more controversial types of data labeling, like moderating violent content, to prisons. “Even if the data being labeled in Finland is uncontroversial right now, we have to think about the precedent it sets,” says Toh. “What stops companies from outsourcing data labeling of traumatic and unsavory content to people in prison, especially if they see this as an untapped labor pool?”

It's also not clear whether labor conditions in Finland's prisons—which famously focus on rehabilitation—could be replicated in other countries with a less progressive approach to justice. In the US, 76 percent of prisoners report that prison labor is mandatory, according to civil rights group, the ACLU. “The prison system in the United States is very, very different from what we have in Finland or Nordic countries. It's a completely different idea,” says Rasila. “In Finland, there is an exclusively positive feeling around the project because everyone knows that this is very voluntary.”

AI companies are only going to need more data labor, forcing them to keep seeking out increasingly unusual labor forces to keep pace. As Metroc plots its expansion across the Nordics and into languages other than Finnish, Virnala is considering whether to expand the prison labor project to other countries. “It’s something we need to explore,” he says.

转载自wired查看原文

全部讨论

no data来都来了,坐下聊聊