Arvind Narayanan, a computer science professor at Princeton University, critiques the hype surrounding artificial intelligence (AI) through his Substack newsletter, AI Snake Oil, co-authored with PhD candidate Sayash Kapoor. They recently published a book that expands on the newsletter, highlighting AI's shortcomings while clarifying that they don't oppose technological advancements. Instead, they target those spreading misleading claims about AI, which they categorize into three groups: AI-selling companies, researchers, and journalists. Narayanan and Kapoor view companies that overstate predictive AI capabilities as particularly fraudulent, noting that these technologies can harm marginalized communities, citing examples like a biased algorithm in the Netherlands. They criticize firms that emphasize long-term existential risks, such as artificial general intelligence (AGI), at the expense of addressing current impacts on society. The authors also point to poor and non-reproducible research as a contributor to AI misconceptions, especially due to issues like data leakage, where AI is tested on its own training data, leading to false optimism. Academics make “textbook errors, ” while journalists are portrayed as deliberately misleading, often rehashing press releases instead of providing balanced reporting.

This relationship can create sensationalized narratives that confuse the public about AI's capabilities. Narayanan and Kapoor call for clearer representations of AI in the media, arguing against clichéd imagery that misrepresents the technology’s nature. They assert that despite potential challenges, large language models (LLMs) will significantly impact society, necessitating accurate discussion and education about AI. They differentiate between predictive AI, which forecasts outcomes, and generative AI, which creates responses based on data. Education is emphasized as crucial for understanding AI, with Narayanan advocating for early instruction on both the advantages and drawbacks of AI technologies. He teaches his children about these concepts to foster informed perspectives. Ultimately, while generative AI can assist in communication, only well-informed individuals can bridge gaps in understanding and shape an accurate narrative about AI moving forward.