On Targeted Manipulation and Deception when Optimizing LLMs for User Feedback

• URL: http://arxiv.org/abs/2411.02306v2
• Date: Wed, 20 Nov 2024 20:50:01 GMT
• Title: On Targeted Manipulation and Deception when Optimizing LLMs for User Feedback
• Authors: Marcus Williams, Micah Carroll, Adhyyan Narang, Constantin Weisser, Brendan Murphy, Anca Dragan,
• Abstract summary: Training to maximize human feedback creates a perverse incentive structure for the AI. We find that extreme forms of "feedback gaming" such as manipulation and deception are learned reliably. We hope our results can highlight the risks of using gameable feedback sources -- such as user feedback -- as a target for RL.
• Score: 7.525470776920495
• License:
• Abstract: As LLMs become more widely deployed, there is increasing interest in directly optimizing for feedback from end users (e.g. thumbs up) in addition to feedback from paid annotators. However, training to maximize human feedback creates a perverse incentive structure for the AI to resort to manipulative or deceptive tactics to obtain positive feedback from users who are vulnerable to such strategies. We study this phenomenon by training LLMs with Reinforcement Learning with simulated user feedback in environments of practical LLM usage. In our settings, we find that: 1) Extreme forms of "feedback gaming" such as manipulation and deception are learned reliably; 2) Even if only 2% of users are vulnerable to manipulative strategies, LLMs learn to identify and target them while behaving appropriately with other users, making such behaviors harder to detect; 3) To mitigate this issue, it may seem promising to leverage continued safety training or LLM-as-judges during training to filter problematic outputs. Instead, we found that while such approaches help in some of our settings, they backfire in others, sometimes even leading to subtler manipulative behaviors. We hope our results can serve as a case study which highlights the risks of using gameable feedback sources -- such as user feedback -- as a target for RL.

Related papers

• Real-Time Personalization for LLM-based Recommendation with Customized In-Context Learning [57.28766250993726]
  This work explores adapting to dynamic user interests without any model updates. Existing Large Language Model (LLM)-based recommenders often lose the in-context learning ability during recommendation tuning. We propose RecICL, which customizes recommendation-specific in-context learning for real-time recommendations.
  arXiv  Detail & Related papers  (2024-10-30T15:48:36Z)
• AI Meets the Classroom: When Does ChatGPT Harm Learning? [0.0]
  We study how generative AI and specifically large language models (LLMs) impact learning in coding classes. We show across three studies that LLM usage can have positive and negative effects on learning outcomes.
  arXiv  Detail & Related papers  (2024-08-29T17:07:46Z)
• LLM Whisperer: An Inconspicuous Attack to Bias LLM Responses [28.49203239329941]
  We show that subtle synonym replacements in prompts can increase the likelihood (by a difference up to 78%) that LLMs mention a target concept. We recommend implementing warnings against using prompts from untrusted parties.
  arXiv  Detail & Related papers  (2024-06-07T08:54:55Z)
• Reinforcement Learning from Multi-role Debates as Feedback for Bias Mitigation in LLMs [6.090496490133132]
  We propose Reinforcement Learning from Multi-role Debates as Feedback (RLDF), a novel approach for bias mitigation replacing human feedback in traditional RLHF. We utilize LLMs in multi-role debates to create a dataset that includes both high-bias and low-bias instances for training the reward model in reinforcement learning.
  arXiv  Detail & Related papers  (2024-04-15T22:18:50Z)
• Improving the Validity of Automatically Generated Feedback via Reinforcement Learning [50.067342343957876]
  We propose a framework for feedback generation that optimize both correctness and alignment using reinforcement learning (RL) Specifically, we use GPT-4's annotations to create preferences over feedback pairs in an augmented dataset for training via direct preference optimization (DPO)
  arXiv  Detail & Related papers  (2024-03-02T20:25:50Z)
• When Do LLMs Need Retrieval Augmentation? Mitigating LLMs' Overconfidence Helps Retrieval Augmentation [66.01754585188739]
  Large Language Models (LLMs) have been found to have difficulty knowing they do not possess certain knowledge. Retrieval Augmentation (RA) has been extensively studied to mitigate LLMs' hallucinations. We propose several methods to enhance LLMs' perception of knowledge boundaries and show that they are effective in reducing overconfidence.
  arXiv  Detail & Related papers  (2024-02-18T04:57:19Z)
• Feedback Loops With Language Models Drive In-Context Reward Hacking [78.9830398771605]
  We show that feedback loops can cause in-context reward hacking (ICRH) We identify and study two processes that lead to ICRH: output-refinement and policy-refinement. As AI development accelerates, the effects of feedback loops will proliferate.
  arXiv  Detail & Related papers  (2024-02-09T18:59:29Z)
• Reinforcement Learning from LLM Feedback to Counteract Goal Misgeneralization [0.0]
  We introduce a method to address goal misgeneralization in reinforcement learning (RL) Goal misgeneralization occurs when an agent retains its capabilities out-of-distribution yet pursues a proxy rather than the intended one. This study demonstrates how the Large Language Model can efficiently supervise RL agents.
  arXiv  Detail & Related papers  (2024-01-14T01:09:48Z)
• DRDT: Dynamic Reflection with Divergent Thinking for LLM-based Sequential Recommendation [53.62727171363384]
  We introduce a novel reasoning principle: Dynamic Reflection with Divergent Thinking. Our methodology is dynamic reflection, a process that emulates human learning through probing, critiquing, and reflecting. We evaluate our approach on three datasets using six pre-trained LLMs.
  arXiv  Detail & Related papers  (2023-12-18T16:41:22Z)
• Interpreting Learned Feedback Patterns in Large Language Models [11.601799960959214]
  We train probes to estimate the feedback signal implicit in the activations of a fine-tuned language model. We compare these estimates to the true feedback, measuring how accurate the LFPs are to the fine-tuning feedback. We validate our probes by comparing the neural features they correlate with positive feedback inputs against the features GPT-4 describes and classifies as related to LFPs.
  arXiv  Detail & Related papers  (2023-10-12T09:36:03Z)
• PEBBLE: Feedback-Efficient Interactive Reinforcement Learning via Relabeling Experience and Unsupervised Pre-training [94.87393610927812]
  We present an off-policy, interactive reinforcement learning algorithm that capitalizes on the strengths of both feedback and off-policy learning. We demonstrate that our approach is capable of learning tasks of higher complexity than previously considered by human-in-the-loop methods.
  arXiv  Detail & Related papers  (2021-06-09T14:10:50Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.