Skip to content
Create an account for full access.

The third generation of the systems approach: accounting for the evolution of time.

Vanchurin, Wolf, Koonin, Katsnelson have noticed that continuing to apply analogical reasoning in classical thermodynamics and information theory to quantities in thermodynamics, machine learning, and evolutionary biology[1], will lead to an unexpected common ontology (framework) for all of them. In other words, evolution could be described as learning, and learning has a thermodynamic nature, making it a purely physical phenomenon. Based on these observations, they formulated the theory of evolution as multilevel learning[2] (deep --- represents the multilevel nature of representations[3]). In this theory, the driving force of evolution is conflicts between objects of different system levels, leading to frustrations[4]. This concept of frustration was introduced into the system language based on the results of studies on spin glasses as an example of the behavior of nonergodic (with memory) systems, and it meant geometric frustrations (the inability to maintain stable spin geometries in glasses). Evolution thus represents a process of learning, which comes down to solving an optimization problem to find the minimum of free energy by changing the structure of multiple system levels. During this process, evolution finds quasi-minima, but not the absolute minimum. Periodically, there is a complexity growth spurt (a major evolutionary transition, the appearance of another system level, another type of whole system), resulting in a sudden minimization of the free energy of the evolving system, yet this is just another quasi-minimum, not an absolute minimum. This series of studies[5] demonstrated the physical nature of system complexity growth (using biological systems as an example, but the reasoning is scale-free, meaning the biological nature of systems or the presence of consciousness do not affect the work's conclusions). Thus, conflicts between system levels, the emergence of frustrations due to these conflicts, and the inevitability of system complexity growth by increasing the number of system levels are introduced to the system's ontology.

All these researchers are in contact with each other (for example, there were discussions between Vanchurin and Friston) because most of this work is based on thermodynamics and the understanding that all systems adhere to the principle of minimizing free energy during their existence, creation, and evolution. Translating the traditional formulas expressing this ontology (this ontology is now expressed in a familiar form to physicists, traditional for thermodynamic calculations) into a constructive form of category theory is a separate topic, but such a problem statement for mathematicians is more or less familiar, and the research program on this topic can be discussed. The same can be said about reformulating the differential form into a quantum-like one[6], which could increase the accuracy of physical modeling of biological evolution and techno-evolution (memetic evolution) due to the quantization/digital nature of most described phenomena.

Thus, the modern/SoTA (third-generation systemic approach) system ontology:

  • Provides object types for multilevel focus to ensure the evolution of target systems (continuous everything) by system-creators.
  • Considers at least three system existence times: operations, construction/evolving of phenome, evolution/development of genome/memome --- based on physics, mathematics, and computer science.
  • Views systems as stable entities within a minimal physicalism (including systems that are active towards themselves and their environment, seeking to minimize free energy by conducting active/embodied inference, regardless of their level of "reasonableness").
  • Provides scale-free descriptions of systems (thus accounting for quantum physics phenomena), explains the emergence of system levels (complexity growth) due to multilevel optimization to achieve the minimum of free energy.
  • Merely expresses mereology not through "eternal classes" and relations between them, but through morphisms and operations reflecting operations with physical systems during their interactions, as well as operations with abstract objects carried out by system-creators with a (universal in terms of Turing machine equivalence) computer within them.

  1. Vitaly Vanchurin, Yuri I. Wolf, Eugene V. Koonin, Mikhail I. Katsnelson, Thermodynamics of evolution and the origin of life, 2022, https://www.pnas.org/doi/full/10.1073/pnas.2120042119 ↩︎

  2. Vitaly Vanchurin, Yuri I. Wolf, Mikhail I. Katsnelson and Eugene V. Koonin, Toward a theory of evolution as multilevel learning, 2022, https://www.pnas.org/doi/10.1073/pnas.2120037119 ↩︎

  3. https://ailev.livejournal.com/1045081.html ↩︎

  4. Geometrical frustration, https://en.wikipedia.org/wiki/Geometrical_frustration ↩︎

  5. Yuri I. Wolf, Mikhail I. Katsnelson, and Eugene V. Koonin, Physical foundations of biological complexity, 2018, https://www.pnas.org/doi/10.1073/pnas.1807890115 ↩︎

  6. Irina Basieva, Andrei Khrennikov, Masanao Ozawabc, Quantum-like modeling in biology with open quantum systems and instruments, https://www.sciencedirect.com/science/article/pii/S0303264720301994 ↩︎