Affiliate disclosure: This post contains affiliate links. If you click through and make a purchase, I may earn a small commission at no additional cost to you.

In a world increasingly shaped by artificial intelligence, it’s easy to fall into the trap of seeing AI as an omnipotent solution for every societal challenge. From predictive policing to automated hiring, AI seems to promise a smarter, faster, more efficient future. But what if that promise is often overstated or even fundamentally misunderstood?

In AI Snake Oil: What Artificial Intelligence Can Do, What It Can’t, and How to Tell the Difference, Arvind Narayanan and co-authors issue a necessary and timely corrective to the blind optimism surrounding AI technologies. Their central argument? Much of what we call “AI” today is not only oversold but actively misrepresented, especially in high-stakes domains.

Drawing the Line Between Hype and Reality

Narayanan divides AI systems into three categories:

1. Genuine AI: Systems that demonstrably work and produce useful results (e.g., image recognition, speech-to-text, machine translation).
2. Dubious AI: Systems that work sometimes but are unreliable, unexplainable, or easily gamed (e.g., facial recognition for emotion detection).
3. Snake Oil AI: Systems that pretend to be intelligent but are built on flawed premises or deliver pseudoscientific results (e.g., AI-based criminal risk assessments).

This tripartite framework is a powerful tool for evaluating any AI product or policy. It urges decision-makers to ask: Is this tool empirically validated? Does it work reliably outside a lab? Is it transparent and explainable?

The Original Snake Oil: A Cautionary Tale

The term “snake oil” itself originates from 19th-century America, where traveling salesmen pitched miracle elixirs said to cure everything from arthritis to baldness. One of the most infamous examples is Clark Stanley, the “Rattlesnake King,” who peddled his Snake Oil Liniment at the 1893 World’s Columbian Exposition. His product was eventually exposed in 1917 by the U.S. government, which found it contained no snake oil at all—only mineral oil, beef fat, red pepper, and turpentine. Stanley’s fraudulent claims were emblematic of an era when pseudoscience thrived on public ignorance and unregulated markets.

This historical context parallels today’s AI landscape in striking ways. Just as Stanley’s potion was sold with scientific-sounding claims and dramatic testimonials, so too are many modern AI tools marketed with exaggerated promises and opaque methodologies. The lesson: skepticism and scientific rigor are timeless shields against being misled.

Other examples of historical “snake oil” include phrenology—a 19th-century pseudoscience that claimed the shape of your skull could determine your intelligence and personality. Though it was widely accepted in its time, it was later debunked and ridiculed. Yet it eerily mirrors some of today’s AI applications, like emotion detection through facial analysis or cognitive profiling through digital behavior tracking. Both depend on the idea that highly subjective human traits can be quantified into data-driven predictions. Both also carry the risk of reinforcing harmful stereotypes.

Reframing AI’s Role in Society

The book is especially critical of the use of AI in social and legal decision-making. For example, algorithmic tools used in hiring often reinforce historical biases under the guise of objectivity. Similarly, predictive policing can embed racial profiling into seemingly neutral systems. In each case, the authors argue, AI isn’t merely failing to solve a problem—it’s actively exacerbating it.

This critique connects closely with a recent blog post from Pursuit of Thought, “Are We Automating Inequality? Lessons from the Algorithms Shaping Our Lives”. That post similarly challenges the unchecked adoption of algorithmic decision-making and underscores the importance of critical, ethical engagement with AI.

What You Can Do

Narayanan encourages a more informed, skeptical public. The path forward, he argues, involves:

• Demanding transparency in how AI models are built and used
• Supporting regulation that protects users from opaque or manipulative systems
• Prioritizing human-centered design in technology development

These ideas echo themes explored in other Pursuit of Thought articles, such as “What If We Designed Technology Like We Design Communities?”, which advocates for designing tools that are transparent, accountable, and human-first.

Final Thought

The core message of AI Snake Oil is not anti-technology. Rather, it’s a call to resist techno-solutionism and refocus AI on domains where it truly adds value. As Narayanan writes, “The problem is not that AI doesn’t work. The problem is that we often apply it where it shouldn’t be used.”

Just as 19th-century consumers eventually demanded real medicine over quack cures, 21st-century citizens must demand real intelligence over AI illusions. Before deploying the next algorithmic tool, we must all ask: Is it miracle, or mirage?

---

Further Reading on Pursuit of Thought:

• “The Four Battlegrounds of AI: Strategy, Data, Talent, and Trust”
• “Co-Intelligence and the Human-Machine Partnership”

Affiliate link reminder: You can purchase AI Snake Oil via this affiliate link to support the authors and this blog.