Building an AI with Its Own Brain, Not Just a Language Model

The article discusses the limitations of current AI systems that rely solely on language models (LLMs) and proposes a new architecture called Skuld, which has its own independent cognitive system or 'brain' that can reason, set goals, and control the use of LLMs and other tools.

Details

The article argues that existing AI agents, despite their ability to remember past interactions, lack true understanding and the capacity to think independently. They simply execute tasks based on prompts without any deeper reasoning or proactive decision-making. In contrast, the Skuld system proposed in the article has a separate 'Brain' component that acts as an independent cognitive system. This Brain has its own world model, belief graph, prediction engine, and goal system, allowing it to build an understanding of the user and their context, set its own objectives, and selectively allocate attention to areas of uncertainty. The LLM is positioned as a tool that the Brain can call upon for reasoning and information retrieval, but the Brain maintains control and applies its own epistemic standards to discount the LLM's outputs. This dual-channel architecture, with one channel for observing the external world and another for internal reasoning, allows Skuld to leverage the capabilities of LLMs while maintaining its own independent decision-making.