Unmasking the Hype: A review of A. I. Snake Oil by Arvind Narayanan and Sayash Kapoor

Sayash Kapoor @Arvind Narayanan

Artificial intelligence (AI) has captured the imagination of policymakers, technologists, and the public in recent years. One might recall Warren Buffet's famous investment maxim: "Be fearful when others are greedy and greedy when others are fearful." Just as Buffet's insight resonated during the 2008 financial crisis, AI snake oil book calls for a similar cautious perspective in today's AI landscape.

?The Government of India is investing significantly in AI, focusing on the agriculture, education, and healthcare sectors. In October 2024, the government approved ?990 crore to establish three AI Centres of Excellence dedicated to these key areas. These centres aim to drive interdisciplinary research and develop scalable AI solutions to address sector-specific challenges. The National Strategy for Artificial Intelligence, formulated by NITI Aayog, also emphasises leveraging AI for social and inclusive growth, focusing on healthcare, agriculture, and education.

However, amidst this wave of euphoria, the book A. I. Snake Oil by Arvind Narayanan and Sayash Kapoor is a timely and necessary critique that demystifies the common misconceptions surrounding AI.

As aptly explained in the book, AI is not a singular entity but a broad umbrella term, much like the word "vehicle." Just as a vehicle can refer to a bicycle, helicopter, or train, AI encompasses many technologies, including deep learning, machine learning, generative AI, retrieval-augmented generation (RAG) etc., These technologies have distinct purposes and mechanisms, which the authors emphasise to help readers grasp that not all AI is created equal or serves the same function. This analogy demystifies the misconception of AI as a monolithic force, underscoring the need to understand its various branches before drawing conclusions about its potential and limitations.

The book goes further by categorising AI into practical, useful applications versus hyped, overestimated versions. It makes a compelling case for distinguishing between generative AI, which focuses on creating content, and predictive AI, which aims to forecast future events. This differentiation is pivotal as it highlights AI's varied purposes and boundaries. The authors illustrate that while generative AI may excel in tasks like text or image creation, predictive AI faces significant challenges due to the inherent uncertainty in forecasting future events. This nuanced classification helps readers appreciate the spectrum of AI applications and recognise where genuine technological progress ends and the hype begins.

The authors highlight the unchecked enthusiasm for AI, which they argue is perpetuated by various stakeholders—from researchers and tech companies to the media, all profiting from the allure of AI's perceived potential.

One of the book's strengths lies in its exploration of prediction, a core function of AI. It discusses why forecasting the future is fraught with challenges and lays out the limitations of prediction itself—an area not commonly addressed in mainstream conversations. This perspective is crucial, especially given that the journey toward developing generative AI has taken decades; it took nearly 50 to 60 years for computers to reach a point where they could identify and describe objects through cameras, as highlighted in the book. Not understanding the limitations of predictions in AI can potentially be career-breaking, especially for the marginalised and vulnerable groups, and can result in erroneous decisions in life-and-death situations.

The book also narrates the existential threats posed by AI, backed by references to numerous researchers who argue that AI alone cannot solve humanity's problems. The discussion on how AI-powered algorithms have contributed to polarising society rather than effectively moderating content is particularly compelling. The authors argue that social media's failure as a content moderator is compelling.

The concluding chapters of A. I. Snake Oil resonate as an appeal to policymakers and institutions, cautioning against viewing AI as a panacea for systemic issues. The authors emphasise that AI should not be seen as a shortcut to reform or replace human-led solutions, especially in developing nations like India. The authors caution against assuming AI can leapfrog broken systems without foundational reform.

This book is informative and a crucial guide for those wishing to harvest AI's fruits with an informed and balanced perspective. Written in a style that is lucid and accessible, A. I. Snake Oil demystifies AI in a way that is relatable to any reader capable of following English-language newspapers. It is a must-read for policymakers, educators, and the public who wish to see beyond the fa?ade and understand the realities masked by the AI hype. What sets this book apart is its accessibility and clear articulation, making it an invaluable read for anyone vested in understanding AI's realistic capabilities and limitations.

AI Snake Oil is a timely and critical examination of the exaggerated narratives surrounding artificial intelligence, much like Professor Raghuram Rajan's depiction of the 2008 economic crisis in his book Fault Lines. Just as Rajan's insights during the 2005 conference paper served as an early caution against unchecked financial practices, AI Snake Oil seeks to unmask the current euphoria enveloping AI technologies. The authors lay out a compelling case for differentiating between genuinely useful AI applications and overhyped ones, clarifying why generative, predictive, and anomaly-detecting AI each have distinct roles and limitations. ?This perspective is invaluable for countries like India. AI is often seen as a catch-all solution for systemic challenges, emphasising the importance of foundational reforms over naive technological optimism.