A Bloomberg report reveals that thousands of underpaid outside contractors are being vetted under immense pressure by Google’s new Bard chatbot. These workers reportedly have just three minutes to validate the accuracy of Bard’s generated responses.
The contractors work for firms like Appen and Accenture and get only $14 per hour. Most have received minimal training in assessing AI chatbot outputs. Their role is to rate Bard’s responses for reliability and usefulness to improve the system.
But with the enormous volume of Bard conversations requiring review, contractors describe feeling “scared, stressed, and underpaid.” One said the “culture of fear” impedes quality and teamwork. The short response review timeframe compounds the pressure.
Google defends its human evaluation methods as rigorously developed over the years to enhance factuality and reduce bias. A spokesperson said human ratings do not directly tune the AI models powering Bard. The company maintains accuracy is improved through multiple techniques.
The revelations underscore concerns about labour practices as large language models like Bard and ChatGPT go mainstream. While AI can generate human-like exchanges, experts argue human oversight remains critical to catch mistakes.
But the working conditions reported at Bard’s vetting facilities suggest even the human oversight is compromised. Burned out, anxious contractors cannot provide the level of oversight required to ensure chatbot safety and quality.
The report will raise doubts about Google’s ability to responsibly deploy AI systems like Bard. It poses difficult questions about the hidden human cost of rapidly advancing AI capabilities sought by tech giants.
Greater transparency and ethical AI practices will be needed to build public trust in chatbots. For now, Google’s chatbot comes at the expense of exhausted contractors pressured to validate its work.