Multi-turn Reinforcement Learning from Preference Human Feedback
This work addresses the problem of aligning LLMs with human preferences in multi-turn interactions for applications like education, though it is incremental as it builds on RLHF with a focus on multi-turn settings.
The paper tackles the limitation of existing RLHF methods that only emulate preferences at single-turn decisions, which restricts their use in multi-turn planning scenarios, by developing novel RL methods from preference feedback between full multi-turn conversations and demonstrates that a deep RL variant outperforms RLHF baselines in a new Education Dialogue environment.
Reinforcement Learning from Human Feedback (RLHF) has become the standard approach for aligning Large Language Models (LLMs) with human preferences, allowing LLMs to demonstrate remarkable abilities in various tasks. Existing methods work by emulating the preferences at the single decision (turn) level, limiting their capabilities in settings that require planning or multi-turn interactions to achieve a long-term goal. In this paper, we address this issue by developing novel methods for Reinforcement Learning (RL) from preference feedback between two full multi-turn conversations. In the tabular setting, we present a novel mirror-descent-based policy optimization algorithm for the general multi-turn preference-based RL problem, and prove its convergence to Nash equilibrium. To evaluate performance, we create a new environment, Education Dialogue, where a teacher agent guides a student in learning a random topic, and show that a deep RL variant of our algorithm outperforms RLHF baselines. Finally, we show that in an environment with explicit rewards, our algorithm recovers the same performance as a reward-based RL baseline, despite relying solely on a weaker preference signal.