|A screenshot of the DARPA-funded 2009 physical intelligence project, the premise that a thermodynamics-based type of artificial intelligence can be built, that evolves. |
“The physical intelligence program will address its overall objective through a coordinated effort in three complementary domains: theory, implementation, and analysis. The objective of the theory domain is to develop and validate a physical formalism that unifies and expands ideas from diverse domains such as evolution, thermodynamics, information, and computation. The objective of the implementation domain is to demonstrate the first human-engineered open thermodynamic systems that spontaneously evolve nontrivial "intelligent" behavior under thermodynamic pressure from their environment. The objective of the analysis domain is to develop analytical tools to support the development of human-engineered physically intelligent systems and to understand physical intelligence in the natural world.”
|A flowchart diagram of the physical intelligence model. |
Synopsis: Our current understanding of the natural world lacks an effective description of the evolution of complexity evident in all living systems. The Physical Intelligence (PI) program hypothesizes that a unifying description called physical intelligence can be developed from currently fragmented disciplinary domains. The vision of the Physical Intelligence program is to develop a physically-grounded understanding of intelligence that applies to engineered systems and scales to high levels of organization. If achieved, this understanding might be applied to problems as diverse as the complexity of living systems, the nature of human intelligence, and the engineering of systems that spontaneously evolve intelligent behavior. This capability would find broad utility in the many defense systems requiring improved autonomy in unmanned systems and improved situational awareness and decision support in manned systems.
Background: For the past 50 years, the dominant paradigm for intelligence supposes that the brain is the seat of intelligence and is functionally equivalent to a computer capable of executing any algorithm. Hence, the current conceptual foundations of intelligence are rooted in the abstractions of logic. Although many valuable and practical applications have emerged from this approach, the goal of true machine intelligence remains distant and the validity of the conceptual foundation is unclear. In contrast, our understanding of the evolution of life is rooted primarily in observations of the natural world, resulting in a multitude of descriptive, domain-specific models of evolutionary processes. A particular challenge is the specification of the mechanisms of selection, as they are typically attributed to a complex environment within which a model system is situated. With some exceptions, current approaches to understanding intelligence and evolution are disconnected and often lack grounding in fundamental physical principles.
|A physical intelligence project overview schematic. |
The Physical Intelligence program supposes that the phenomena associated with intelligence and evolution can be understood as natural consequences of complex, open thermodynamic systems. In particular, evolutionary variation may result from thermodynamic fluctuations, and evolutionary selection may result from entropy production. Further, the hierarchy of organization observed in the natural world may result from the evolutionary/thermodynamic growth of “networks” of energy transduction mechanisms. These networks include neural systems, for example, but also many other complex natural systems. Although the idea that life is “a struggle for entropy” (Boltzmann) has been supposed for more than a century, acknowledged formalisms, validation in the natural world, and applications to engineered systems are scarce. The Physical Intelligence program aspires to change this situation.
Description: The outline of the program is hinged on three independent but connected points: (a) Creating a theory (a mathematical formalism) and validating it in natural and engineered systems; (b) Building the first human-engineered systems that display physical intelligence in the form of abiotic, self-organizing electronic and chemical systems; (c) Developing analytical tools to support the design and understanding of physically intelligent systems.
See main: Neumann automaton theoryThe theory of electrical-chemical intelligent systems, to note, was first proposed by American chemical engineer John Neumann in 1948, who envisaged a robot or automaton, made of wires, electrical motors, batteries, etc., constructed in such a way that when floating on a lake stoked with component parts, it will reproduce itself (self-replicate), albeit only if a source of free energy is available.