Procedural Adherence and Interpretability Through Neuro-Symbolic Generative Agents
This addresses the problem of managing and interpreting long-term agent behavior for interactive applications like choose-your-own-adventure games, representing a novel method rather than an incremental improvement.
The paper tackles the challenge of ensuring procedural adherence and interpretability in generative agents by combining formal logic-based program synthesis with LLM content generation, achieving at least 96% adherence to temporal constraints compared to 14.67% for a purely LLM-based agent.
The surge in popularity of large language models (LLMs) has opened doors for new approaches to the creation of interactive agents. However, managing and interpreting the temporal behavior of such agents over the course of a potentially infinite interaction remain challenging. The stateful, long-term horizon reasoning required for coherent agent behavior does not fit well into the LLM paradigm. We propose a combination of formal logic-based program synthesis and LLM content generation to bring guarantees of procedural adherence and interpretability to generative agent behavior. To illustrate the benefit of procedural adherence and interpretability, we use Temporal Stream Logic (TSL) to generate an automaton that enforces an interpretable, high-level temporal structure on an agent. With the automaton tracking the context of the interaction and making decisions to guide the conversation accordingly, we can drive content generation in a way that allows the LLM to focus on a shorter context window. We evaluated our approach on different tasks involved in creating an interactive agent specialized for generating choose-your-own-adventure games. We found that over all of the tasks, an automaton-enhanced agent with procedural guarantees achieves at least 96% adherence to its temporal constraints, whereas a purely LLM-based agent demonstrates as low as 14.67% adherence.