Abstract
As a scholar and teacher of media technologies, I found this book on AI grifts very informative, especially in how it provides basic knowledge about AI products for non-experts. AI Snake Oil—a critical assessment of AI applications from two seasoned computer scientists at Princeton University—is incisive, supplying copious evidence from real life (and making it a good source of case studies for those teaching AI-related courses). In an era where the promise of AI captivates the world, few works are as dedicated to uncovering its deficiencies as this book. While not a traditional academic work, this book does a deep dive into the development of AI technology by synthesizing and analyzing secondary resources. Categorizing AI applications into three types—predictive AI, generative AI, and content moderation AI—the authors endeavor to explain differences in performance across these categories. Most chapters, except for the introductory and concluding sections, are organized thematically around one of the three AI applications, drawing upon a rich array of cases to anchor its chapter focus.
The book’s primary theoretical insight is its systematic mapping of the AI technology landscape. While acknowledging the complexities of deploying AI for content creation and content moderation, the book critiques the techno-optimism surrounding AI-driven predictive analytics and autonomous decision-making. It reviews the origins of AI hype—a complex phenomenon fueled by the amalgamation of different technologies and the hyperbolic portrayal of their capabilities, and attributes this frenzy to biased news reports, academic work, and PR statements. The authors point to cases like the news title “Epic’s Faulkner Has High Hopes for Forthcoming Cosmos Technology,” and demonstrate how news coverage often relies on corporate claims without further verifying the technology. Media narratives also frequently rely on grand metaphors that mispresent AI’s actual capabilities, a trend exemplified by a New York Times piece referring to Google’s voice assistant as the “gods of artificial intelligence.”
The authors assess three ways journalism contributes to the “AI illusion.” First, they say, news reports distort the predictive accuracy of AI applications and products; journalists elevate the predictive efficacy while remaining oblivious to the ways AI companies could massage performance metrics in their favor. The second critique posits that the press has been intellectually co-opted by corporate interests, especially when reporters regurgitate PR fluff without independent verification. A third line of critique centers on the lack of rigor when journalists integrate research findings into their reporting. In the coverage of AI scholarly work, media outlets tend to include accuracy numbers while failing to articulate the underlying methodologies used to compute them. However, explicit disclosure of methodological constraints is imperative, as performance evaluations are highly contingent upon the measurements used, leaving them vulnerable to researchers’ selective reporting inflating outcomes.
Academic research is also complicit in AI hype, representing another significant locus of AI-centric optimism. The authors’ argument—that academic contributions overstate and spread the power of AI—is provocative. Leaning on a comprehensive body of literature, this book suggests that the accuracy of predictive AI is intrinsically constrained. Scaling up data volume cannot mitigate the risks associated with algorithmic prediction. On the one hand, algorithm training for AI applications relies on sample data rather than population data, rendering them highly susceptible to representational bias. On the other hand, predictive AI is built on existing evidence. The forecasts are drawn from historical aggregates, so they fail to account for unpredictable systemic disruptions, known as exogenous shocks. Compounding the problem is the reality that “AI research relies on corporate funding” (p. 236). This financial dependency tethers research agendas to commercial viability, thereby marginalizing critical efforts to interrogate industry-perpetuated AI hyperbole.
I have seen a plethora of communication research following this flawed heuristic in AI-focused research—spotlighting the superiority of AI without acknowledging the structural limitations. Studies generalize results from finite samples, overlooking both the context-dependent nature of AI agents and the fundamental instability of performance metrics used to evaluate them. Papers make assertions like “AI outperforming humans” in their titles, further boosting AI hype. I appreciate the authors’ advocacy for a paradigm shift away from the superficial celebration of computational efficacy to address the methodological opacity that currently obscures how these computational techniques arrive at their determinations.
As with any scholarly endeavor, this book is not without limitations. While its critique of AI hype is largely compelling, the arguments lose cogency when addressing concerns at the intersection of communication and journalism. The authors failed to elucidate the pivotal role of investigative journalism in demystifying the prevailing AI narrative. Indeed, news outlets frequently publish content from PR statements created by AI stakeholders, but this trend does not negate the fundamental importance of journalism’s gatekeeping function. Though the authors did not explicitly recognize journalism’s role in debunking AI hype, their inclusion of investigative news reports—such as those involving Retorio and Social Sentinel—implicitly underscores the media’s power to expose the misuse of AI in recruitment and student surveillance. The authors even admitted that “journalists questioned city officials about the efficacy of the predictive tool” (p. 50). With only a brief section titled “news media misleads the public,” the book is conspicuously silent on the public advocacy role journalism plays. An overview of the AI landscape shouldn’t just call out journalism’s flaws while turning a blind eye to its advantages. In fact, journalists are increasingly aware of AI-related risks. From Northeastern University’s academic curriculum to the Pulitzer Center and Poynter Institute’s professional workshops, the journalism community is laying the groundwork for AI literacy. The authors mentioned a Pulitzer Center initiative that funded a group of reporters to work on “in-depth AI accountability stories” (p. 26).
This book underscores the imperative of AI literacy for navigating the modern algorithmic landscape. Looking ahead, three specific areas in the intersection of AI and communication deserve closer scrutiny. First, we need literacy frameworks to equip journalists and strategic communication professionals with knowledge about the inherent risks of AI applications. Second, we should study the regulatory frameworks governing the production and deployment of AI within the media ecosystem. Finally, we should shift the attention from an AI application perspective to understand why specific AI architectures outperform or underperform their human counterparts in complex communication tasks. More crucially, computational communication studies need to demystify AI tools and spell out exactly where and how bias can creep in—whether during the design phase, the data analysis, or the final estimation. By explicating these pathways, scholars can cultivate a rigorous, ethically grounded future for journalism and communication research in the age of AI.
