Fig. 4: This diagram illustrates the structural framework and potential vulnerabilities of LLM-based AI scientists.
From: Risks of AI scientists: prioritizing safeguarding over autonomy
The agent is organized into five interconnected modules: LLMs, planning, action, external tools, and memory & knowledge. Each module exhibits unique vulnerabilities. The arrows depict the sequential flow of operations, starting from memory & knowledge to the use of external tools, underscoring the cyclic and interdependent nature of these modules in the context of scientific discovery and application.
