AI Has a Secret: Humans Do the Work
October 10, 2025
A key component of artificial intelligence output is not artificial at all. The Guardian reveals “How Thousands of ‘Overworked, Underpaid’ Humans Train Google’s AI to Seem Smart.” From accuracy to content moderation, Google Gemini and other AI models rely on a host of humans employed by third-party contractors. Humans whose jobs get harder and harder as they are pressured to churn through the work faster and faster. Gee, what could go wrong?
Reporter Varsha Bansal relates:
“Each new model release comes with the promise of higher accuracy, which means that for each version, these AI raters are working hard to check if the model responses are safe for the user. Thousands of humans lend their intelligence to teach chatbots the right responses across domains as varied as medicine, architecture and astrophysics, correcting mistakes and steering away from harmful outputs.”
Very important work—which is why companies treat these folks as valued assets. Just kidding. We learn:
“Despite their significant contributions to these AI models, which would perhaps hallucinate if not for these quality control editors, these workers feel hidden. ‘AI isn’t magic; it’s a pyramid scheme of human labor,’ said Adio Dinika, a researcher at the Distributed AI Research Institute based in Bremen, Germany. ‘These raters are the middle rung: invisible, essential and expendable.’”
And, increasingly, rushed. The write-up continues:
“[One rater’s] timer of 30 minutes for each task shrank to 15 – which meant reading, fact-checking and rating approximately 500 words per response, sometimes more. The tightening constraints made her question the quality of her work and, by extension, the reliability of the AI. In May 2023, a contract worker for Appen submitted a letter to the US Congress that the pace imposed on him and others would make Google Bard, Gemini’s predecessor, a ‘faulty’ and ‘dangerous’ product.”
And that is how we get AI advice like using glue on pizza or adding rocks to one’s diet. After those actual suggestions went out, Google focused on quality over quantity. Briefly. But, according to workers, it was not long before they were again told to emphasize speed over accuracy. For example, last December, Google announced raters could no longer skip prompts on topics they knew little about. Think workers with no medical expertise reviewing health advice. Not great. Furthermore, guardrails around harmful content were perforated with new loopholes. Bansal quotes Rachael Sawyer, a rater employed by Gemini contractor GlobalLogic:
“It used to be that the model could not say racial slurs whatsoever. In February, that changed, and now, as long as the user uses a racial slur, the model can repeat it, but it can’t generate it. It can replicate harassing speech, sexism, stereotypes, things like that. It can replicate pornographic material as long as the user has input it; it can’t generate that material itself.”
Lovely. It is policies like this that leave many workers very uncomfortable with the software they are helping to produce. In fact, most say they avoid using LLMs and actively discourage friends and family from doing so.
On top of the disillusionment, pressure to perform full tilt, and low pay, raters also face job insecurity. We learn GlobalLogic has been rolling out layoffs since the beginning of the year. The article concludes with this quote from Sawyer:
‘I just want people to know that AI is being sold as this tech magic – that’s why there’s a little sparkle symbol next to an AI response,’ said Sawyer. ‘But it’s not. It’s built on the backs of overworked, underpaid human beings.’
We wish we could say we are surprised.
Cynthia Murrell, October 10, 2025
Comments
Got something to say?
-
- Subscribe to Beyond Search
Feature archive
News archive
-
