Princeton professors’ new book ‘AI Snake Oil’ offers smart perspective on navigating AI

The rapid rise of artificial intelligence has sparked intense debate between those heralding its revolutionary potential and others dismissing it as overhyped, prompting two Princeton scholars to offer a more nuanced perspective in their new book “AI Snake Oil.”

The core argument: Princeton computer science professor Arvind Narayanan and Ph.D. candidate Sayash Kapoor aim to help people distinguish between AI’s genuine capabilities and empty promises.

• Their book “AI Snake Oil” focuses on providing foundational knowledge to separate legitimate AI advances from misleading claims
• The authors argue that while some AI applications show remarkable progress, many marketed AI products make unfounded or impossible claims
• A key example includes AI hiring tools that claim to predict job performance based on brief video interviews, despite lacking scientific evidence

Critical distinctions: The authors emphasize that treating all AI technologies as similar is as problematic as failing to distinguish between bikes, cars, and spaceships.

• Generative AI (like ChatGPT) and predictive AI (used for credit scoring) are fundamentally different technologies requiring separate evaluation
• While generative AI has shown impressive year-over-year improvements, predictive AI still relies largely on decades-old tools
• The authors warn that conflating different AI technologies leads to public confusion and misguided concerns

Real-world impacts: The most consequential AI applications affecting people’s lives today are predictive AI systems making high-stakes decisions.

• These algorithms determine crucial outcomes like bail amounts, hospital stay durations, and hiring decisions
• Insurance coverage may be denied based on AI predictions about recovery time
• The widespread use of similar AI vendors means individuals could face repeated rejections across multiple job applications

Industry influence concerns: The concentration of AI development among major tech companies raises significant issues about the future direction of the technology.

• Large language models can only be developed by the biggest tech labs, giving companies like OpenAI, Google, and Facebook outsized influence
• These same companies are among the biggest spenders on political lobbying
• The centralization of AI development allows industry giants to drive the technical agenda

Positive outlook: Despite concerns, the authors see promising developments in AI’s future.

• They anticipate beneficial AI tools will become seamlessly integrated into knowledge workers’ daily workflows
• Previously “intractable” problems like spellcheck have become routine automated features
• Self-driving technology could potentially help reduce the approximately 1 million annual global auto-related fatalities

Looking ahead: The evolution of AI technology suggests a pattern where controversial cutting-edge applications eventually become normalized, useful tools – much like spellcheck and autocomplete today – while the most concerning current applications may be replaced by more reliable alternatives.