# Independence of Approximate Clones Source: https://www.youtube.com/watch?v=2F0mynPSv00 Recap page: https://rapidrecap.app/video/2F0mynPSv00 Generated: 2026-02-02T00:03:24.12+00:00 --- ## Quick Overview The paper "Independence of Approximate Clones" by Theo-dora Misur demonstrates that even for highly robust AI systems like RLHF, the existence of near-perfect clones that are nearly identical in their preferences can cause the entire system to fail when evaluated using simple voting rules or when one clone is a total non-contender, revealing a fundamental vulnerability in AI alignment structures. **Key Points:** - The paper examines the "Independence of Approximate Clones" where two candidates (A and B) are nearly identical in voter preferences, causing instability in ranked voting systems. - In a four-candidate race (A, B, C, D), if B is removed, the winner can flip from A to C due to the structure of ranked choice voting, illustrating the spoiler effect. - The paper introduces a metric, the 'weak independence' metric, which quantifies how often removing a candidate changes the ranking order of the remaining candidates. - The authors show that for LLMs trained with RLHF, the existence of near-perfect clones (alpha score of 0.24) causes the system to fail certain tests, suggesting a fundamental flaw in the alignment structure. - The failure mode is demonstrated in the Scottish election example where removing a non-contender (C) caused the outcome to flip, showing that even subtle rank changes matter. - The research concludes that if the underlying structure allows for small variations in rankings to cause large outcome changes (the paradox of weak independence), the system is brittle, even if the AI performs well on average. - The core finding is that high robustness scores against near-perfect clones do not guarantee stability when the underlying preferences are split or ambiguous. ![Screenshot at 0:15: The title slide introducing the paper 'Independence of Approximate Clones' authored by Theodora Misur, setting the context for analyzing AI robustness against near-identical candidate profiles in voting systems.](https://ss.rapidrecap.app/screens/2F0mynPSv00/00-00-15.jpg) **Context:** The video discusses findings from a research paper titled "Independence of Approximate Clones" authored by Theodora Misur from the University of Amsterdam. The paper investigates the robustness of AI systems, particularly those trained using Reinforcement Learning from Human Feedback (RLHF), against scenarios where multiple 'clones' or candidates are nearly indistinguishable in their underlying preference profiles. This is framed within the context of ranked-choice voting systems, where small changes in voter preferences or the presence of near-identical options can drastically alter election outcomes, posing a threat to the perceived stability of the AI alignment process. ## Detailed Analysis The discussion centers on the paper by Theodora Misur concerning the "Independence of Approximate Clones" in the context of ranked-choice voting and AI alignment. The core issue is that if two candidates (or AI outputs) are very similar, removing one or introducing a third, clearly inferior candidate can cause the election result to flip, a phenomenon known as the spoiler effect or the paradox of weak independence. The paper tests this using three different datasets from mathematics, political science, and AI alignment studies. In the Scottish election example, removing candidate C caused the winner to change, demonstrating this instability. The paper proves that even highly robust AI systems, like those trained with RLHF, are vulnerable to this. When the AI models were tested, they showed a failure rate on these edge cases, specifically when ranking candidates with high similarity scores (alpha of 0.24). The analysis shows that the algorithms were trained to prefer unique answers over correct similar answers, leading to a fundamental flaw where the structure itself allows for instability, even if the system appears robust on average. ### Paper Introduction and Problem - Analyzing 'Independence of Approximate Clones' - The paper uses math, political science, and AI alignment data sets - The core issue is instability in ranked voting when candidates are similar. ### The Spoiler Effect Example - Scottish Elections - Two similar candidates (A and B) - Removing C causes the winner to flip from A to the next preferred candidate (D or B, depending on the scenario) - This demonstrates that the ranking order is not robust. ### AI Alignment Implications - RLHF Systems - The failure mode applies to LLMs trained with RLHF - Near-perfect clones (alpha score 0.24) cause the system to fail tests designed to check for this vulnerability. ### Algorithm Behavior - Preference for Uniqueness - The algorithms were trained to prefer unique answers over correct, similar answers, leading to fragility in borderline cases. ### Conclusion on Robustness - The risk is that the system's structure, not just the data, creates instability - The paradox of weak independence shows that high average performance hides inherent brittleness. ![Screenshot at 0:00: The podcast/channel intro screen with the text 'Become A Member Today!' overlaying an audio waveform graph.](https://ss.rapidrecap.app/screens/2F0mynPSv00/00-00-00.jpg) ![Screenshot at 0:25: The speaker discusses the paper title, 'Independence of approximate clones,' which addresses a very old problem with a new lens.](https://ss.rapidrecap.app/screens/2F0mynPSv00/00-00-25.jpg) ![Screenshot at 0:53: Visualizing the spoiler effect: Two candidates with similar platforms split the vote, allowing a third, less-preferred candidate to win.](https://ss.rapidrecap.app/screens/2F0mynPSv00/00-00-53.jpg) ![Screenshot at 2:27: The speaker introduces the first test case, the Scottish election, which uses a form of ranked voting.](https://ss.rapidrecap.app/screens/2F0mynPSv00/00-02-27.jpg) ![Screenshot at 4:08: The speaker summarizes the finding that for any election with four or more candidates, there is no standard voting rule that satisfies the independence axiom.](https://ss.rapidrecap.app/screens/2F0mynPSv00/00-04-08.jpg)