How to build recursive AI agents? Control the controllers
Problem Statement:
Recursive AI agents are designed to manage and optimize other AI agents, creating a hierarchical system of control. The study will investigate the challenges and methodologies involved in building such agents, including their architecture, decision-making processes, and potential applications. Building recursive AI agents involves several significant challenges:
· Computational Complexity: Recursive AI agents require substantial computational resources, especially for large-scale simulations. Managing these resources efficiently is crucial to avoid performance bottlenecks.
· Risk of Infinite Loops: Ensuring that recursive functions have proper base cases is essential to prevent infinite recursion, which can lead to system crashes or unresponsive behaviour.
· Bias in Simulation: Agents may learn from skewed data or environments, leading to poor generalization and biased decision-making.
· Opaque Decision-Making: As agents evolve and optimize themselves, understanding the rationale behind their decisions becomes increasingly difficult. This lack of transparency can hinder trust and accountability.
Tasks:
The student will study the concept of holarchy and recursive agents and then:
- Define Architecture: Design the architecture for recursive AI agents, specifying the layers of control and interaction between agents.
- Develop Algorithms: Create algorithms that enable recursive AI agents to effectively manage and optimize other AI agents.
- Simulation: Implement simulations to test the performance and efficiency of the recursive AI agents in a chosen scenario.
Work:
10% Theory, 70% Simulations, 20%Writing
Contact:
Alessandro Chiumento (a.chiumento@utwente.nl)