Raphaël Raux
As a PhD candidate in economics at Harvard's Kenneth C. Griffin Graduate School of Arts and Sciences, 2025 Harvard Horizons Scholar Raphaël Raux explores the nuanced interplay between human cognition and artificial intelligence (AI). His project, "Human Learning about AI," conducted in collaboration with fellow PhD student Bnaya Dreyfuss, investigates how people mistakenly project human cognitive processes onto AI systems, which leads to misunderstandings about the capabilities and limitations of these technologies.
Raux's interest in AI was sparked by examining how humans and machines perceive the world differently. He notes a common misconception: people assume that machines struggle with tasks that are difficult for humans. However, AI performs certain tasks far more efficiently, despite requiring significant computational resources. His project offers insights into these differences, emphasizing that our tendency to equate human-like behavior with machine intelligence can skew our economic decision-making processes, particularly the adoption of AI in various industries.
Through experiments with chatbots such as ChatGPT and Dewey, a chatbot specializing in parenting advice, Raux examines how the phenomena of “human projection” leads people to project human-like thinking onto AI, affecting the generalizations they make from their observing AI performance. This leads people to react very strongly to mistakes on easy problems or to misunderstandings that are completely humanly unreasonable. While these reactions make sense for humans who see failing an easy problem as a good sign of poor ability, they make less sense for AI, which can have a very high success rate yet sometimes fail at easy tasks.
Raux’s research also explores the broader implications of AI adoption in the economy, highlighting the importance of realistic expectations about AI’s capabilities. He argues for a more lucid understanding of AI, urging policymakers and industry leaders to better educate the public on the distinctions between human and machine intelligence to optimize AI’s integration into society—and rethink how AI systems are evaluated before they are deployed.
By challenging the assumptions underlying AI-human interaction, Raux’s work encourages a re-evaluation of how AI technologies are assessed and implemented. He hopes his research will guide more informed decisions about AI and promote a balanced approach to harnessing its potential, paving the way for economic and social advancements.
The text above was generated by GPT 4.0 and engineered and edited by Paul Massari.