Paper proposes a cognitive framework for autonomous AI systems and highlights trust gaps

A new arXiv preprint argues that autonomous systems represent the ultimate stage of AI development and outlines a comprehensive framework for their design and evaluation. The paper, titled “Bringing AI to Autonomous Systems — From Cognition to Collective Intelligence,” is authored by Joseph Sifakis and was submitted on September 11, 2026.

The proposed framework centers on a generic agent architecture in which cognitive functions are organized around a long-term memory that stores the agent’s evolving knowledge. The paper emphasizes the need to combine connectionist AI and symbolic AI while integrating AI with systems engineering to address the technical challenges of autonomous systems.

The author highlights three core implementation challenges: linking sensory data to structured data stored in memory, decision-making and planning to achieve the agent’s goals, and coordinating agents to combine individual and collective intelligence.

The paper introduces a broader conception of agent trustworthiness that includes cognitive validity — how the agent uses its knowledge in decision-making — in addition to behavioral properties typical of traditional systems. It also proposes directions for evaluating such trustworthiness.

The work concludes with a critical assessment of the substantial gap between the aspirational vision of autonomous multi-agent systems and the current state of the art, underscoring the distance between today’s capabilities and the long-term goals of autonomous AI systems.