Dr. Jillian Fisher breaks down political bias in LLMs, why true neutrality is impossible, and how AI literacy can defend against subtle AI persuasion.
As the capabilities of modern LLMs evolve, questions about their objectivity and influence on human behavior are moving to the forefront. Join us to explore the political neutrality—or lack thereof—in LLMs, how architectural decisions force models to favor majority opinions, and the surprising psychological tactics AIs use to persuade users.
To help us, Dr. Jillian Fisher, a recently minted PhD in statistics and computer science from the University of Washington and a researcher in AI alignment and safety, joins us to share her perspective on the downstream effects of biased models, why true political neutrality is impossible, and how AI literacy could be our best defense.
- Defining partisan bias and how it manifests in an LLM's training data, math, and output.
- The illusion of true political neutrality and why philosophers and political scientists debate its existence.
- Exploring "approximations of neutrality" like reasonable pluralism, output transparency, and model refusals.
- The difference between good faith and bad faith user queries, using Area 51 as an example.
- How models use subtle framing and "information packing" (rather than human-like empathy) to persuade users.
- The sycophantic nature of LLMs and humanity's inherent preference for bias that aligns with their own.
- How AI literacy and understanding the noisy, biased nature of internet training data could inoculate users against unwanted persuasion.
This episode is full of technical insights and forward-looking predictions that are sure to change how you approach your next dataset. As we move into a new era of AI, it's the perfect time to explore the fundamentals of the next frontier!
Chapters:
- Intro & What We’re Reading Right Now (1:19)
- Defining Political Bias In LLMs (5:41)
- How Bias Shows Up In Outputs (9:28)
- Do Biased Models Change Users (11:44)
- Why True Neutrality Breaks Down (14:18)
- Reasonable Pluralism And Transparency Options (17:10)
- Refusal, Safety, And Bad Faith Prompts (19:56)
- Persuasion And The AI Authority Effect (21:14)
- AI Literacy That Actually Helps (28:46)
- Closing (35:08)
What did you think? Let us know.
Do you have a question or a discussion topic for the AI Fundamentalists? Connect with us to comment on your favorite topics:
- LinkedIn - Episode summaries, shares of cited articles, and more.
- YouTube - Was it something that we said? Good. Share your favorite quotes.
- Visit our page - see past episodes and submit your feedback! It continues to inspire future episodes.
