Intelligent Machines Podcast Episode 809: "Fun Mustache"

Release Date: March 6, 2025

Hosts: Leo Laporte, Paris Martineau, Jeff Jarvis
Guest: Gary Marcus, AI Expert, Psychologist, and Cognitive Scientist

1. Introduction

The episode kicks off with host Leo Laporte welcoming listeners to "Intelligent Machines" and introducing today’s guest, Gary Marcus. Gary is recognized for his expertise in artificial intelligence (AI), his critical perspective on artificial general intelligence (AGI), and his contributions to the field through his Substack newsletter and published works.

2. Interview with Gary Marcus

a. Gary’s Skepticism Towards AGI

Gary Marcus opens the discussion by expressing his love for AI but conveys skepticism about the current trajectory towards AGI. “[02:53] I love AI, but I don't like the way it's happening now. And if that makes me a contrarian, then you can call me a contrarian.”

b. The Hype vs. Reality of Current AI

Gary delves into the disparity between the hype surrounding AI and its actual capabilities. He criticizes companies like OpenAI for “weaponizing” hype to drive stock valuations without delivering on their promises of AGI. “[05:46] OpenAI has historically been at the top of that list. They’ve weaponized hype the most of any of these companies.”

c. Regulation and Policy Concerns

The conversation shifts to the need for robust AI regulation. Gary reflects on signing a letter in 2023 advocating a six-month pause on AI development, emphasizing ongoing concerns about AI safety and reliability. “[06:10] Even though I didn't think it was a perfect letter, I thought we needed to call attention to how fast things were moving and how little we understood about how to make AI safe and reliable.”

d. Symbolic vs. Neural AI Approaches

Gary champions the integration of symbolic AI with neural networks, advocating for a neuro-symbolic AI approach to address current limitations in reasoning and abstraction. “[28:38] What I really favor is neuro symbolic AI, which would be a hybrid of classical symbolic AI and neural networks.”

e. Defining AGI and Its Challenges

Gary discusses the elusive definition of AGI, sharing his efforts to establish clear criteria for AGI and criticizing the shifting definitions within the AI community. “[22:40] AGI was defined in terms of human flexibility and cognition... Now it's just being defined based on financial metrics like revenue generation, which seems absurd.”

f. AI Hallucinations and Reliability

A significant portion of the interview focuses on the unreliability of current AI models, particularly their tendency to generate “hallucinations” or incorrect information. Gary recounts personal experiences where AI failed to accurately perform tasks, underscoring the need for improved control mechanisms. “[30:55] Machines make stuff up anyway... They just kind of do what they do and you throw more data and you hope for the best.”

g. Opportunity Costs and Future Directions

Gary warns against the heavy investment in large language models (LLMs) at the expense of exploring alternative AI methodologies. He argues that this focus might hinder the development of more reliable and trustworthy AI systems. “[05:46] If we just keep dumping all of our eggs in the LLM basket, then we're missing some opportunity to make better AI.”

3. AI News Highlights

a. Meta’s $200 Billion AI Center

The hosts discuss Meta's colossal investment in AI infrastructure, questioning the tangible outcomes and highlighting potential opportunity costs related to such vast expenditures.

b. Turing Award Recipients Warn About AI Risks

Leo mentions that recent Turing Award winners have voiced concerns over the unsafe deployment of AI models, drawing parallels with Gary’s earlier points about AI reliability and safety.

c. The Return of Digg by Gary Marcus and Alexis Ohanian

Gary Marcus teams up with Reddit co-founder Alexis Ohanian to revive Digg, aiming to address past issues like system gaming and create a more resilient social platform. “[03:30] We've had multiple episodes... So we need a thinker.”

d. The AI.com Domain Saga

The episode highlights the intriguing saga of the AI.com domain, managed by domain broker Larry Fisher, who has a history of selling influential domains. Multiple redirections of AI.com have led to confusion and speculative headlines claiming ownership by entities like OpenAI.

e. AI’s Role in Modern Filmmaking

The hosts explore the controversial use of AI in film production, such as modifying actors' voices or generating visual content, which has sparked debates over ethical implications and authenticity in cinema.

f. Google Pixels in NYC Subways

A pilot program in New York City utilizes Google Pixels attached to subway cars to monitor track defects through AI-driven audio analysis, showcasing practical applications of AI in public infrastructure maintenance.

g. YouTube’s Premium Lite Offer

YouTube introduces a "Premium Lite" option, allowing users to remove some ads for $8 without subscribing to the full Premium service, reflecting evolving monetization strategies in digital content platforms.

4. Host Discussions

Post-interview, the hosts engage in a lively discussion about broader themes related to AI, business ethics, and the historical evolution of technology. They reflect on the challenges of regulating rapidly advancing technologies and the importance of balancing innovation with responsible oversight. Additionally, interactions with AI chatbots during the show reveal both fascination with and frustration over AI’s current limitations.

5. Conclusion

The episode concludes with teasers for upcoming guests, including Ray Kurzweil—a prominent figure in predicting the Singularity and coining the term “intelligent machines.” The hosts encourage listeners to join their "Club Twit" for exclusive content and engage with the community through various platforms. They emphasize the ongoing nature of AI’s impact and the necessity of informed discourse to navigate its future developments.

Notable Quotes:

• Gary Marcus on AI's Current Path: “[02:53] I love AI, but I don't like the way it's happening now. And if that makes me a contrarian, then you can call me a contrarian.”

• On AI Hype and Accountability: “[05:46] No one really holds anybody accountable. They just look at the next promise.”

• On the Need for Neuro-Symbolic AI: “[28:38] What I really favor is neuro symbolic AI, which would be a hybrid of classical symbolic AI and neural networks.”

• Discussing AI Reliability: “[30:55] Machines make stuff up anyway. We don't have very good control of large language models.”

• On Opportunity Costs of AI Investment: “[05:46] If we just keep dumping all of our eggs in the LLM basket, then we're missing some opportunity to make better AI.”

Additional Highlights:

• Club Twit Membership: The hosts promote their "Club Twit," encouraging listeners to subscribe for exclusive content, including access to their Discord server and ad-free podcast experiences.

• Upcoming Events: Announcement of a 24-hour live stream celebrating the podcast network's 20th anniversary, featuring interactive segments and listener participation.

This summary encapsulates the core discussions and insights from "Intelligent Machines" Episode 809, providing a comprehensive overview for listeners who may not have tuned in. It highlights Gary Marcus's critical perspective on the current state of AI, the challenges in defining and achieving AGI, and the broader implications of AI advancements in various sectors.