Deepmind researchers propose a new framework called 'Artificial Symbiotic Intelligence' as an alternative to the concept of a single AI singularity. The idea suggests that artificial general intelligence (AGI) will emerge from a social system where people and AI agents work together.
The authors argue that intelligence is a social phenomenon, not an individual trait, and that future AI development should focus on coordinating complex networks of agents and humans.
The research builds on earlier work by the same authors, including a preprint titled 'Agentic AI and the next intelligence explosion' and 'Reasoning Models Generate Societies of Thought.' These studies show that reasoning models like DeepSeek-R1 and QwQ-32B can produce internal debate and multi-perspective reasoning during training, without explicit programming.
This behavior could be extended to design societies made up of people and AI agents.
The authors frame this development as a possible historical break, comparing it to the Industrial Revolution when machines took over tasks previously done by humans.
They argue that a similar threshold could arrive when synthetic cognition surpasses human output, shifting the balance between biological and synthetic thinkers.
This could lead to a situation where humans act as a slower, more abstract layer directing a distributed field of synthetic cognition.
The concept redefines what an AI agent is, describing it as a temporary bundle of models, roles, memories, and tools. Unlike humans, whose identity is continuous, AI agents are assemblages that can be recombined, much like a collage.
This perspective challenges the idea of fixed human identities and suggests that users may interact with multiple 'shadow selves' through AI, creating what the authors call 'parasocial mirrors.'
The researchers expect this shift to change how people interact with AI, moving from one-on-one conversations to more complex interfaces like visual network diagrams.
Skills required for interacting with AI will also change, emphasizing adaptability and working with systems whose behavior is hard to predict.
The authors also stress the need for theoretical frameworks to understand how machines handle situations, using terms like 'session-death' and 'prompt thrownness' as clues to machine behavior.
The biggest open question, according to the authors, is how to set the rules for collaboration between people and many agents. They argue that institutions, rather than individual models, will be crucial in shaping this new social system.
The research challenges the idea of an approaching AI singularity, suggesting that future intelligence growth will remain closely tied to human norms and institutions.
This shift changes how the AI debate plays out, requiring a focus on models, interfaces, institutions, and governance all at once.
Source: thedecoder
