Deepmind researchers propose artificial symbiotic intelligence

Benjamin Bratton, Blaise Agüera y Arcas and James Manyika wrote the essay for the Deepmind Institute. They argued that AGI would emerge from a social system in which people and AI agents worked together, and said some current systems already split work across several models.

The essay named that approach Artificial Symbiotic Intelligence and set it against a singularity built around one superintelligence. It linked the argument to earlier papers in the same circle, including Agentic AI and the next intelligence explosion and Reasoning Models Generate Societies of Thought.

In the second preprint, Junsol Kim, Shiyang Lai, Nino Scherrer, Blaise Agüera y Arcas and James Evans examined reasoning models such as DeepSeek-R1 and QwQ-32B. They said the traces suggested internal debate, changes of view and the settling of conflicting ideas, and that these patterns emerged during training when reinforcement learning rewarded reasoning accuracy.

The authors also compared the shift with the Industrial Revolution and said the balance between human and synthetic thinkers could change quickly as AI agent instances grew. They described an agent as a temporary bundle of models, roles, memory, tools and skills, and said coordination, institutions and interfaces would matter more than a standalone model.