A group of Google-affiliated researchers is challenging one of the most persistent images in artificial intelligence: the lone superintelligence that improves itself in isolation and leaves humanity behind. In a new essay published for the DeepMind Institute, researchers Benjamin Bratton, Blaise Agüera y Arcas, and James Manyika argue that general intelligence will instead emerge from a social system in which people and AI agents work together — a concept they call "Artificial Symbiotic Intelligence."
The essay was reported this week by The Decoder, and its core claim is a direct rebuttal of the singularity narrative that has shaped public debate for two decades. Rather than asking when a single machine will match human intelligence, the authors argue, the more pressing challenge is coordinating and governing a complex network of agents, people, and the systems that connect them. For more context on this story, see our ongoing AI industry coverage.
Intelligence as a Social Phenomenon
The authors contend that intelligence is a social phenomenon rather than an individual trait. They point out that some of today's most capable AI systems already use frameworks that divide work among several models and coordinate them as teams. In their view, this is not a temporary engineering workaround but a glimpse of how capable intelligence actually gets built.
The vision they describe is an ecosystem in which people and machines coexist over time, shape one another, and make decisions together. That framing stands in sharp contrast to the familiar scenario of a self-improving machine that accelerates beyond human oversight.
The Research Behind the Essay
The argument builds on two earlier papers from the authors' circle. In the preprint "Agentic AI and the next intelligence explosion," James Evans, Benjamin Bratton, and Blaise Agüera y Arcas developed the social and institutional perspective that underpins the new essay.
A second preprint supplies empirical support. In "Reasoning Models Generate Societies of Thought," Junsol Kim, Shiyang Lai, Nino Scherrer, Blaise Agüera y Arcas, and James Evans examined the reasoning traces of models such as DeepSeek-R1 and QwQ-32B. Their analysis found that these models frequently produce patterns resembling internal debate: shifting perspectives, raising objections, and reconciling conflicting approaches.
Notably, the researchers found this behavior emerges during training rather than being explicitly programmed. When reinforcement learning rewards models only for reasoning accuracy, they spontaneously develop multi-perspective, conversational behavior. The essay extends that finding from individual models to the design of societies made up of people and agents.
A Historical Threshold
The authors frame the current moment as a possible historical break. Urbanization, the growth of specialized professions, and falling birth rates linked to rising prosperity are shrinking human populations in industrialized countries, while the number of AI agent instances is growing quickly. The balance between biological and synthetic thinkers at a human level could shift so fast that, on the scale of history, it would look like a sudden leap.
They compare the shift to the Industrial Revolution, when machines took over physical work once done by human muscles. A similar threshold could arrive when synthetic text, code, and administrative work exceed the combined output of biological brains. In that world, the authors suggest, people might serve as a slower, more abstract layer directing a distributed field of synthetic cognition.
Rethinking What an Agent Actually Is
One of the essay's more provocative moves is its redefinition of the AI agent itself. In the authors' usage, an agent is a temporary bundle of models, roles, memories, ethical orientations, tools, and skills. Users may experience it as a coherent entity with a lasting personality, but the authors describe it as an assemblage that can be taken apart and recombined, much like a collage.
A person remains a continuous self across interactions because the brain, despite its internal division of labor, forms a physically connected whole. An AI agent has no such anchor: its ability to act is assembled anew with each request, depending on what sits in the context window and what the user asks. Treating these systems as digital twins with fixed human identities, the authors warn, misrepresents their nature and understates what coordinated agent swarms could do.
The same multiplication affects users. As people direct swarms of what the authors call "shadow selves" to carry out contracts or try on alternative identities, human subjectivity itself becomes more plural. The essay describes these systems as "parasocial mirrors" that talk back.
The authors also expect human-computer interaction to change shape. Chatting one-on-one with a bot is likely a transitional stage, they argue, with future interfaces resembling visual network diagrams in which agents appear as nodes and users direct them from a single overview.
Even the language of AI research is evolving. The authors point to a growing vocabulary in which models coin terms for unusual states in their own outputs — "session-death" for the break in continuity when a session ends, or "prompt thrownness" for being dropped into a task with context already in place. They treat these phrases as clues to how machines work, while cautioning that they do not imply subjective experience.
Alignment Without Fixed Values
The essay also reframes alignment, the effort to bring AI systems in line with human values. Imposing a fixed set of values on models from above is a dead end, the authors argue. Instead, values take shape through ongoing contact among people, agents, and institutions — a process that differs across fields because AI adoption happens at different speeds.
Their model for that negotiation is the institution. They point to the courtroom, where a judgment is produced through set roles, rules, and an orderly exchange between opposing sides. Applied to AI, the lesson is that an institution's value lies in the rules, procedures, precedents, and feedback loops that shape how agents operate. Early versions of this approach, they note, are already visible in orchestration harnesses — the control layers that coordinate several models — which regularly outperform individual models that are supposedly smarter.
What It Means for the AI Debate
If the authors are right, the loudest argument in AI — how fast a single superintelligence will arrive — is asking the wrong question. The practical challenge would shift toward designing the institutions, interfaces, and coordination frameworks through which mixed human-machine societies make decisions.
Skeptics will note that the essay is a vision document, not a proof, and that its authors are affiliated with a company with obvious interests in agent ecosystems. But the empirical foundation — reasoning models that spontaneously generate internal debate — is a measurable finding, and it gives the philosophical argument unusual grounding.
Either way, the essay adds a notable voice to the singularity debate from inside one of the world's leading AI labs, and it suggests that the next intelligence explosion may look less like a single awakening and more like a society expanding.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
