Cognitive Architectures for Language Agents

“We introduce such a framework, drawing parallels with two ideas from the history of computing and artificial intelligence (AI): production systems and cognitive architectures. Production systems generate a set of outcomes by iteratively applying rules (Newell and Simon, 1972). They originated as string manipulation systems – an analog of the problem that LLMs solve – and were subsequently adopted by the AI community to define systems capable of complex, hierarchically structured behaviors (Newell et al., 1989). To do so, they were incorporated into cognitive architectures that specified control flow for selecting, applying, and even generating new productions (Laird et al., 1987; Laird, 2022; Kotseruba and Tsotsos, 2020).

We suggest a meaningful analogy between production systems and LLMs: just as productions indicate possible ways to modify strings, LLMs define a distribution over changes or additions to text. This suggests that the kinds of controls used with production systems might be equally applicable to LLMs, addressing aspects including memory, grounding, learning, and decision making, among others.”