The concept of the ‘agent’ has been a cornerstone of AI research, influencing both theoretical frameworks and practical applications. However, this paradigm is fraught with ambiguities and an anthropocentric bias that may hinder the advancement of AI. The authors differentiate between agentic, agential, and non-agentic systems, arguing that the prevalent focus on agentic AI could obscure the underlying computational processes that govern these technologies.
A critical examination of the agent paradigm reveals challenges in defining key concepts such as autonomy and goal-directedness in AI systems. The authors propose that a more fruitful approach involves exploring frameworks grounded in complex systems and biological inspirations. This shift could facilitate the development of AI architectures that are not merely extensions of human intelligence, but rather new, robust forms of intelligence independent of anthropomorphic characteristics.
👉 Pročitaj original: arXiv AI Papers
