Earlier this year, I captured a lovely full moon photo and uploaded it to a social media platform. Within seconds, my luminous moon was tagged as the property of a global stock image library, and I was advised I had breached their copyright. The Artificial Intelligence (AI) behind the automated ‘content moderation’ wasn’t working correctly that day.

Researchers Narayanan and Kapoor believe that humans tend to latch onto overblown claims about AI by development companies. We take those impressive-sounding accuracy numbers at face value while ignoring insights that might provide a more complete picture.

AI Snake Oil explores predictive AI tools – used to forecast crimes, disease outbreaks and election results – sold with the promise of accuracy, fairness, and efficiency. Apparently, they don’t work. Generative AI tools – ChatGPT, Midjourney, etc – do a reasonable job at generating content, but Chatbots are consistent ‘bullshitters’. Finally, content moderation AI tools – responsible for filtering blasphemy, identifying child exploitation content and identifying copyrighted content – can work but are more influenced by company policies than technology.

The rise of AI in almost every aspect of our lives is a hot discussion topic. AI is a technological wonder of our time, but sometimes it doesn’t work as promised. AI Snake Oil introduces notions that AI appeals to ‘broken institutions’ looking for a quick fix. The authors are ‘not okay with leaving the future of AI up to the people currently in charge’.

Reviewed by Mark Parry

ABOUT THE AUTHORS

Arvind Narayanan is professor of computer science at Princeton University and director of its Center for Information Technology Policy. He is the coauthor of Bitcoin and Cryptocurrency Technologies (Princeton) and Fairness and Machine Learning.

Sayash Kapoor is a PhD candidate in computer science at Princeton. He previously worked as a software engineer at Facebook, where he helped create AI for content moderation. To learn more, please visit AISnakeOil.com.