Workers who rate the chatbots “are often not given enough time to evaluate longer responses,” one of the workers, Ed Stackhouse, 49, wrote in a May 15 letter to two senators leading a congressional hearing on the risks of AI. “The fact that raters are so exploited could lead to a faulty and ultimately more dangerous product.”
Appen told the workers they were fired because of “business conditions,” Stackhouse said. Appen did not respond to requests for comment.
“Appen is responsible for the working conditions of their employees, including pay, benefits, employment changes, and the tasks they’re assigned,” said Google spokesperson Courtenay Mencini. “We, of course, respect the right of these workers to join a union or participate in organizing activity, but it’s a matter between the workers and their employer, Appen.”
With the letter to Congress, the workers joined a growing chorus of voices concerned about the rapid rollout of AI tools to millions of people. AI researchers, politicians and tech advocates have raised concerns that the technology is infusing bias into tech products, enabling cybercrime, replacing some workers — and could eventually exceed human control. What was initially a fight to secure better work conditions for themselves has become about something bigger — the impact of AI on society, Stackhouse said.
An explosion of interest in AI from businesses and consumers has kicked off an arms race between Google and its archrival Microsoft to develop and sell AI tools and put the tech into existing products, from Google Search to Microsoft Word. The boom was triggered by OpenAI, a much smaller company, which released its ChatGPT chatbot in November, astonishing the world with its ability to conduct cogent conversations, pass professional exams and write computer code. Experts attribute OpenAI’s success in part to its use of human testers and trainers to refine the bot and teach it to be less offensive and more interesting than previous versions of the technology.
The fast pace and competitive nature of the AI boom is causing concern among AI ethics experts, who say the technology reflects racist and sexist biases that exist in the reams of data vacuumed up from the internet and used to train them. And the bots routinely make up false information and pass it off as real.
To combat these effects, Google and other Big Tech companies have turned to outside workers, part of the huge armies of contract workers amassed over the years to do everything from run cafeterias to write computer code. Appen contractors like Stackhouse, for example, have for years helped improve Google Search by rating its search results for helpfulness and accuracy.
Even before it pushed Bard out to the general public in March — a move the company described as an “experiment” — Google has leaned on these contract workers. The Appen contractors say Google began shifting their work to Bard in February.
“It felt very dramatic,” said Michelle Curtis, 43, another fired Appen contractor. Raters have a time limit for accomplishing assigned tasks, she said; with Bard, that could mean they had five minutes, for example, to evaluate a detailed answer about the origins of the Civil War. “There’s just no humanly possible way to do it,” Curtis said.
Curtis and Stackhouse are part of a group of Appen workers who had been trying to organize their colleagues to demand better pay and benefits, with help from the Alphabet Workers Union, a group of Google employees and contractors associated with the Communications Workers of America. Curtis is a mom from Idaho. Stackhouse, who has a disability, lives in North Carolina. Both worked from home part-time for Appen, an arrangement they described as a lifeline for supporting their families.
In 2019, Google said it would begin requiring contractors to pay employees $15 an hour; Stackhouse and Curtis said Appen never met that mark. The company agreed to a raise wages to $14.50 an hour, but declined to give employees sufficient hours to qualify for benefits, the workers said.
With the NLRB complaint, the workers are escalating the fight, drawing more attention to the reality of low-paid labor behind cutting-edge AI. The process could take months, as the NLRB works through a huge load of cases, but eventually they hope to get their jobs back, secure back-pay and maybe even get Google to sit down with them and hear their concerns about AI, Curtis and Stackhouse said.
Contractors are like “ghost workers,” Stackhouse said, unrecognized for the huge value they provide Google. “Without people doing, our jobs there would not be any AI.”
