AI Snake Oil audiobook cover - What Artificial Intelligence Can Do, What It Can’t, and How to Tell the Difference

AI Snake Oil

What Artificial Intelligence Can Do, What It Can’t, and How to Tell the Difference

Arvind Narayanan, Sayash Kapoor

★ 3.4 / 5(36 ratings)
Start ListeningDownload App
Categories:
Technology & the FutureEducationScience

If You're Curious About These Questions...

You should listen to this audiobook

Listen to AI Snake Oil — Free Audiobook

Loading player...

Key Takeaways from AI Snake Oil

Learning Tools

Reinforce what you learned from AI Snake Oil

Mind Map

AI Snake Oil
Core Premise+
Generative AI+
Predictive AI+
Content Moderation AI+
The Path Forward+

Quiz — Test Your Understanding

Question 1 of 7
According to the text, why are chatbots prone to producing plausible but inaccurate statements?
  • A. They rely on outdated search engine algorithms to retrieve factual data.
  • B. They generate text by predicting word sequences rather than truly understanding context.
  • C. They are intentionally programmed to prioritize creative storytelling over factual accuracy.
  • D. They lack access to sufficient amounts of training data from reliable sources.
Question 2 of 7
What hidden labor issue is associated with the development of generative AI models?
  • A. A severe shortage of highly paid software engineers in North America and Europe.
  • B. The reliance on unpaid internships at major tech companies to write foundational code.
  • C. The outsourcing of labor-intensive data labeling to countries with low wages and high workloads.
  • D. The mandatory replacement of human artists with AI systems in corporate environments.
Question 3 of 7
What is identified as a major limitation of predictive AI when it comes to decision-making?
  • A. It requires too much real-time testing, making it too slow for modern business needs.
  • B. It focuses solely on past data and fails to account for how its own choices change the circumstances it predicts.
  • C. It is entirely random in its outputs, making it impossible to rely on for historical analysis.
  • D. It cannot process numerical data efficiently, limiting its use in fields like finance and medicine.
Question 4 of 7
How do companies often react when fully automated predictive AI systems make poor decisions or fail?
  • A. They immediately recall the AI system and financially compensate the affected users.
  • B. They blame the underlying algorithms and sue the third-party developers.
  • C. They shirk responsibility by claiming that human oversight should have been in place.
  • D. They double down on the AI's decision, arguing that the machine is mathematically infallible.
Question 5 of 7
What is 'collateral censorship' in the context of content moderation AI?
  • A. The intentional silencing of political opponents by social media executives.
  • B. The process where platforms over-moderate and remove more content than necessary to protect themselves from legal liability.
  • C. The accidental deletion of user accounts due to a glitch in the fingerprint matching algorithm.
  • D. The practice of relying entirely on human moderators to review flagged content in culturally sensitive regions.
Question 6 of 7
What approach does the book recommend for regulating artificial intelligence?
  • A. Discarding all current laws to create an entirely new, AI-specific rulebook from scratch.
  • B. Relying exclusively on self-regulation by major AI companies to prevent stifling innovation.
  • C. Using existing regulatory frameworks as a solid foundation while strengthening agencies to prevent regulatory capture.
  • D. Banning the use of AI in high-stakes areas like criminal justice and healthcare completely.
Question 7 of 7
To address the economic and employment impacts of AI automation, what policy does the text suggest might help incentivize retaining a human workforce?
  • A. Implementing a universal basic income funded by social media advertising.
  • B. Placing a strict cap on the number of AI models a company can deploy.
  • C. Implementing a 'robot tax' on companies that benefit from automation.
  • D. Mandating that all AI-generated content be legally classified as public domain.

AI Snake Oil — Full Chapter Overview

AI Snake Oil Summary & Overview

AI Snake Oil (2024) explores the myths and misconceptions surrounding artificial intelligence, particularly focusing on where AI fails to deliver on its promises. It critically examines the limitations of current technologies such as generative AI, predictive AI, and content moderation AI, emphasizing the need for a grounded understanding of what AI can and cannot achieve. 

Who Should Listen to AI Snake Oil?

  • Anyone keen to educate themselves about AI 
  • Policymakers and industry regulators 
  • Business and community leaders exploring AI adoption

About the Author: Arvind Narayanan, Sayash Kapoor

Arvind Narayanan is a computer science professor at Princeton University and director of Princeton’s Center for Information Technology Policy. Narayanan’s work explores the ethical and societal implications of artificial intelligence and digital technologies. Together with Sayash Kapoor, Narayanan was named one of TIME’s 100 most influential people in AI. 

Sayash Kapoor is a Ph.D. candidate in computer science at Princeton University. Kapoor previously worked in AI for Columbia University, École Polytechnique Fédérale de Lausanne (EPFL), and Facebook. 

🎧
Listen in the AppOffline playback & background play
Get App