Omega Future Institute
In September 2026, a research team at the University of Chinese Academy of Sciences published a paper.
It introduced the first Periodic Table of Agent Capabilities.
Published in Springer’s Annals of Data Science, the paper lists 243 capability configurations, from zero to infinite intelligence.
For the first time, this table brings physical, artificial intelligence and biological systems, together with philosophical entities, into one classification of agent capabilities.
It also proposes that differences between classical mechanics, relativity and quantum mechanics may arise from the different capabilities assigned to their observers.
What does that mean?
Classical mechanics. Relativity. Quantum mechanics.
These three major theories of physics describe strikingly different pictures of the world.
Classical mechanics: absolute space and time, definite trajectories, and, in principle, precise predictions.
Relativity: light sets a speed limit, and space and time depend on the observer.
Quantum mechanics: outcomes are probabilistic, and measurement can change the system.
Why the differences? The usual answer is that nature behaves differently at different scales.
But perhaps another factor is who is observing.
Generalized Agent Theory treats an observer as an open information-processing system.
Its five capabilities are input, output, memory, generation and control. Together, they form a capability configuration, kappa.
In this framework, the ideal classical observer has instantaneous, unlimited access to information.
It observes without intervening. Its output is zero: an all-knowing spectator.
The relativistic observer has information access constrained by the speed of light.
Inside a closed elevator, local measurements alone cannot distinguish uniform acceleration from an equivalent gravitational field.
The quantum observer is modeled as finite in every capability.
Observation involves interaction, so output is no longer zero.
The move from classical to quantum is not simply a loss of capability. It also introduces interaction.
Our simulation places a Schrödinger’s cat experiment inside an equivalence-principle elevator.
Observers with different capabilities face the same scene, yet obtain different effective pictures.
The three theories may therefore describe the same reality,
while different observer capabilities yield three different effective pictures.
This is a possibility proposed within Generalized Agent Theory.
It still needs further, stronger evidence.
The capability table and the study of observers in physics are part of Generalized Agent Theory, developed by Chinese scientist Feng Liu and his team.
It is an exploratory foundational framework for artificial intelligence, centered on the concept of the agent.
Its starting axiom is that an agent is an open information-processing system. From this follow four core components.
First, the Minimal Complete Architecture: the standard agent model.
Second, the Periodic Table of Agent Capabilities: a classification of agents.
Third, the evolutionary dynamics of agents.
Fourth, the multi-agent relationship spectrum.
Source
