Krista Pawloski, an AI worker on Amazon Mechanical Turk, recalls a moment two years ago that shaped her view on artificial intelligence ethics. While moderating tweets for racism, she encountered the term 'mooncricket' and discovered it was a racial slur against Black Americans. This made her question how many similar errors she and other workers might have missed.
After years of witnessing AI's inner workings, Pawloski no longer uses generative AI products and advises her family to avoid them. She does not allow her teenage daughter to use tools like ChatGPT and encourages others to test AI on topics they know well to see its fallibility. She often wonders if her tasks could harm people, and says many times the answer is yes.
Pawloski is not alone. A dozen AI raters told the Guardian they now urge friends and family to avoid generative AI or use it cautiously. One Google AI rater, who evaluates responses for Google Search's AI Overviews, said she uses AI sparingly and has forbidden her 10-year-old daughter from using chatbots, citing the need for critical thinking skills first.
Amazon stated that workers choose tasks at their discretion, while Google said ratings are aggregated data points that do not directly impact algorithms, and that protections are in place to surface high-quality information. Experts say that when those who make AI trustworthy trust it the least, it signals a larger issue of speed overtaking safety.
