“Ghost work” needed to build artificial intelligence systems may be harming workers, a group of eight Democratic US lawmakers wrote in a letter to top technology and AI firms.

The term refers to tasks such as data labeling and rating chatbot responses for accuracy and safety. Workers handling these tasks can receive low wages; no benefits; and they can be exposed to traumatizing content such as torture, rape and murder, according to the letter.

These workers often work on third-party platforms including crowdsourcing platforms.

“Contrary to the popular notion that AI is entirely machine-based and autonomous, AI systems in fact depend heavily on human labor — for tasks ranging from labeling training data to rating chatbot responses for accuracy and safety,” according to the letter. “Today, millions of workers perform these tasks, paid hourly or for piecework on digital labor platforms like Amazon’s Mechanical Turk, and their numbers are growing.”

The letter said the median wage of workers on Mechanical Turk has been estimated at $1.77 per hour, according to research. And workers are often under surveillance and spend uncompensated time doing such things as searching for tasks, the letter said.

It also raised concerns about quality. According to the letter, one study found that YouTube had banned some LGBTQ content because the data workers were located in a country where LGBTQ content was censored.

The letter was sent to top executives at a list of the largest tech firms: Microsoft, Amazon, Meta, Google and IBM. In addition, letters were sent to OpenAI, Inflection AI, Anthropic and Scale AI.

“Data workers” are defined in the letter as “company employees, contractors, subcontractors, independent contractors, pieceworkers, and anyone else who your company compensates directly or indirectly for performing data work on any internal platform or externally.”

Lawmakers signing the letter were Sen. Edward Markey, D-Massachusetts; Sen. Ron Wyden, D-Oregon; Sen. Elizabeth Warren, D-Massachusetts; Sen. Bernie Sanders, I-Vermont; Rep. Pramila Jayapal, D-Washington; Rep. Jamaal Bowman, D-New York; Rep. Katie Porter, D-California; and Rep. Mark Pocan, D-Wisconsin.

SIA has reached out to the tech companies for comment.

Separately, more than 60 US senators attended a closed-door briefing on AI produced by Sen. Chuck Schumer, D-New York. One topic was  whether AI should be regulated. Also on hand were tech leaders, including Elon Musk, Bill Gates, Sundar Pichai, Mark Zuckerberg and Sam Altman.

The Alphabet Workers Union, which seeks to represent directly employed and contingent workers at Google, issued a statement regarding working conditions in AI. The statement came from Ed Stackhouse, who has spent 10 years as a Google rater and described himself as a “ghost worker” tasked with rating the quality and accuracy of responses from the Google AI chatbot Bard.

Stackhouse said he is paid $14.50 per hour and he raised concerns about working conditions, including lack of time to adequately complete tasks.

“The working conditions for AI workers like me are incredibly strenuous,” he said in his statement. “Timing for AI tasks seems to be random. Raters many times are faced with two or three minutes to finish a task that takes 10 minutes to verify accuracy. If we take too long, our account can be locked out for productivity by an AI on the back end — and that means no paid work until that is addressed days or weeks later.”