Better Offline: CZM Rewind - The Academics That Think ChatGPT Is BS

Release Date: August 13, 2025

Introduction

In this rerun episode of Better Offline, host Ed Zitron engages in a thought-provoking discussion with three esteemed academics—Michael Townsend Hicks, James Humphries, and Joe Slater—from the University of Glasgow. They delve into their critically acclaimed paper titled "ChatGPT is Bullshit," exploring the limitations and implications of ChatGPT within the broader context of artificial intelligence and society.

1. Defining Bullshit: Philosophical Foundations

The conversation begins with the academics defining "bullshit" in the philosophical sense, drawing inspiration from the late philosopher Harry Frankfurt.

• Joe Slater [03:40]: "Bullshitting occurs when you speak without caring about the truth of what you say. It's not about lying; it's about not caring whether what you're saying is true."

The distinction between soft and hard bullshit is introduced to categorize different levels of disingenuous communication.

• Joe Slater [04:20]: "Soft bullshit doesn't necessarily involve malicious intent, whereas hard bullshit involves a deliberate attempt to deceive."

2. ChatGPT as a Bullshitter: Analyzing Intentionality

The discussion transitions to ChatGPT's functionality, questioning whether the model can be considered a "hard bullshitter" given its lack of consciousness and intention.

• James Humphries [06:07]: "Minimally, ChatGPT is a soft bullshitter. It doesn't have intentions because it's not sapient, but it does generate content without concern for truth."

3. The Mechanics of ChatGPT: Representation and Perception

The academics scrutinize how ChatGPT generates responses, emphasizing that it lacks true understanding or representation of the world.

• Ryan Reynolds [08:34]: "ChatGPT doesn't track or perceive the world. It simply predicts the next word based on statistical patterns, which is fundamentally different from human cognition."

• James Humphries [15:37]: "ChatGPT doesn't engage in planning or tracking external data. Its responses are purely statistical, devoid of any real-world awareness."

4. Limitations in Language Understanding and Reasoning

Highlighting specific linguistic challenges, the speakers illustrate ChatGPT's shortcomings in handling nuanced language tasks.

• Ed Zitron [25:26]: "When prompted to insert expletives correctly, ChatGPT often fails, producing grammatically correct but contextually inappropriate sentences."

• James Humphries [19:40]: "Even with grammatically correct sentences like 'The drink that I beer,' ChatGPT struggles to understand meaninglessness compared to humans."

5. Educational Implications: The Rise of AI-Generated Essays

A significant portion of the conversation addresses the impact of ChatGPT on academic integrity and student learning.

• James Humphries [44:34]: "Students are increasingly using ChatGPT to write essays, leading to a surge in mediocre work. Detecting AI-generated content remains a challenge without concrete evidence."

• Ryan Reynolds [48:23]: "In higher-level classes, ChatGPT-authored papers become blatantly poor, lacking coherent arguments and becoming repetitive, making them easy to spot."

6. Ethical and Social Concerns: Phishing and Manipulation

Beyond academia, the discussion touches on broader societal risks posed by AI technologies like ChatGPT.

• Ryan Reynolds [60:30]: "Phishing attacks have surged by 1,265% since ChatGPT's launch, as malicious actors leverage AI to craft convincing deceptive emails."

• James Humphries [62:46]: "Biases in AI algorithms persist, reminiscent of past issues with facial recognition systems, underscoring the need for careful oversight."

7. Rethinking Intelligence and Consciousness in AI

The panel debates the philosophical underpinnings of intelligence, consciousness, and how they relate to AI models.

• Ryan Reynolds [13:16]: "ChatGPT isn't designed to represent the world accurately. It lacks internal states connected to external reality, unlike human cognition."

• James Humphries [36:52]: "A model specialized in one task, like passing the Turing Test, doesn't equate to genuine intelligence or consciousness."

8. Moving Forward: Strategies and Solutions

Concluding the discussion, the academics contemplate potential strategies to mitigate the adverse effects of AI in education and society.

• Ryan Reynolds [52:59]: "Implementing assignments that focus on argumentation rather than summarization can help students develop critical skills that ChatGPT cannot replicate."

• James Humphries [57:11]: "Universities need to bolster support systems, like writing centers, to ensure students develop essential skills without overreliance on AI tools."

Conclusion

This episode of Better Offline offers a critical examination of ChatGPT, challenging the notion of AI as an intelligent or conscious entity. Through philosophical discourse and practical insights, the panel highlights the ethical, educational, and societal implications of deploying such technologies without a thorough understanding of their limitations. The conversation serves as a compelling reminder of the need for deliberate and informed approaches to integrating AI into various facets of human endeavor.

Notable Quotes:

• Joe Slater [03:40]: "Bullshitting occurs when you speak without caring about the truth of what you say."

• James Humphries [06:07]: "Minimally, ChatGPT is a soft bullshitter."

• Ryan Reynolds [08:34]: "ChatGPT doesn't track or perceive the world."

• Ryan Reynolds [60:30]: "Phishing attacks have surged by 1,265% since ChatGPT's launch."

Note: Timestamps are based on the provided transcript and reflect the approximate position of each quoted statement within the episode.