CC BY
0
UDC 004.42
ARTIFICIAL CHEMISTRY MODELS: ARTIFICIAL THOUGHTS FOR ARTIFICIAL MIND
E. A. Kol'chugina
Penza State University, Penza, Russia [email protected]
Abstract. We propose a new concept of artificial intelligence based on principles of artificial chemistry, where short programs representing artificial molecules can spontaneously react with each other forming new programs or complex computing structures. These structures representing source and results of a special kind of intellectual activity are seen as artificial thoughts. The structure of "artificial mind" following this concept and methods of interactions between programs are also proposed.
Keywords: spontaneous emergence of programs, artificial chemistry models, artificial mind, artificial brain, artificial thought
For citation: Kol'chugina E. A. Artificial chemistry Models: Artificial thoughts for Artificial Mind // Vestnik of Penza State University. 2024; 4: 75-78. (In Russ.).
Introduction and background
Artificial chemistry [1] is one of the eldest types of computer models that appeared at the dawn of computing technology: J. von Neumann's self-replicating cellular automata [2] are claimed to be the first example of models of this type. Currently, artificial chemistry models are known as a sub-branch of software models of artificial life [3].
The set of artificial chemistry models is not uniform; these models use different mathematical apparatus. A wide variety of such models are functionally oriented. Their base is 2-calculus and its analogues.
The first model of this kind, AlChemy [4], is one of the first models of artificial life that studied the formation of finite sets of functions with a steady composition capable of self-maintenance and self-reproduction.
In this artificial chemistry, functions play a dual role of both reagents (molecules) and rules for conducting reactions. During the simulation, the molecules collide. The collision can be elastic, without consequences, and reactive, in which new molecules are formed. The easiest and most obvious way to form a molecule is to create a composition of reagents.
This approach is popular in artificial chemistry, and it has been further developed in some models, for example, such as Gamma [5] and CHAM [6], where analogous of A-calculus has been proposed. In comparison to AlChemy, these newer models are focused on practical use, namely the automation of programming and the creation of new programming paradigms.
These models generously apply the metaphor of chemistry, including notions of "ions", "heating", "cooling" and "freezing" [6]. The resulting programming concepts are allowing creation of programs with dynamically changing structure and composition, in which functions and their compositions are
© Kol'chugina E. A., 2024
considered as a disposable material. The expenditure of such material means the end of the program execution.
Programming concepts based on this approach have many advantages, such as implicit parallelism, compact size of the resulting programs, high expressiveness of the programming style, and new capabilities to achieve automatic program development. But the list of disadvantages is not empty either. And the first disadvantage that should be mentioned is the function-oriented programming style used by these concepts. Function-oriented programming concepts are less popular among IT professionals due to an unusual style of thinking that is more difficult for a human to assimilate.
Another disadvantage is the non-obvious application area of such programming concepts. Currently, only a few applications are known in which dynamically changing programs are crucial.
Collectively, these disadvantages discourage a wide range of IT professionals from using such concepts. But it is obvious that the future belongs to methods based on evolution and self-organization: these methods can eliminate the influence of the human factor and make the final products more flexible and high-quality. In general, the function-oriented approach in artificial chemistry models is flourishing, evidence of which can be found in [7, 8].
The author has spent almost twenty years trying to bridge the gap between imperative programming and the paradigm of artificial chemistry and bring a decentralized and spontaneous character to the models of dynamically changing programs. The results in [9, 10] represent artificial chemistry models there functions are accomplished as active entities (possibly, program agents). In [9], the functions are understood as reactions for which initial substances are needed. Actually, reactions are looking for a substrate in the model space. In [10], the functions correspond to artificial analogues of atoms capable of spontaneous formation of molecules. Some kinds of atoms are stationary, and some kinds of atoms are mobile (volatile) and are also looking for other atoms to react with.
The results of the experiments described in [9, 10] led to the emergence of dynamic computing structures in the form of programs with an implicit [9] or explicit construction [10]. These computing structures are aggregates of logical functions of different complexity, and thus they represent some logical decisions or problem solutions. These structures are also can be seen as material inscriptions of spontaneously arising ideas or thoughts of artificial mind, emerging from the interactions between logical elements.
This led to the idea of a new form of artificial intelligence based on the emergence of computing (logical) structures corresponding to principles similar to the function-oriented models of artificial chemistry. The structures themselves are artificial molecules, and the basic functional elements are artificial atoms.
The aim of this study is to propose and consider such an idea, as well to propose also a new form of artificial intelligence that can be used independently or in ensemble with other concepts, such as neural nets, agent-oriented models, decision trees, etc.
The method
Figure 1 represents the new concept of artificial intelligence named "artificial mind", to be distinct from traditional concepts of artificial intelligence.
The main component of the diagram in Figure 1 is an artificial brain: a vessel in which spontaneous reactions occur and new molecules arise. The space inside the vessel can be cellular or have a different topology, Euclidean or non-Euclidean. The reactions proceed according to a set of reaction rules, which is also part of the artificial brain. In general, the idea of artificial brain is close to "magical stirring mechanism" from [6].
But it is necessary to mention that there is no central processing mechanism in this model: all atoms and molecules retain their own source of internal energy, hence reactions are equiprobable and spontaneous.
Traditionally, an internal energy means a processor time. This analogy is common in artificial chemistry models [1].
The crucial question is the construction or choice of the means to achieve selective asymmetric interactions between atoms and molecules.
Fig. 1. The structure of artificial mind based on artificial chemistry approach
The analysis of various models of artificial chemistry [1-10] makes it possible to identify various kinds of such means. We state that the most powerful and perspective kinds are:
- connection by program labels [1];
- connection by common shared data structures [9];
- connection by specially designed tools imitating charged particles [10].
The latter two kinds of means provide decentralized spontaneous interactions between artificial atoms and molecules and the simulation of self-organization processes.
The initial and resulting molecules of the artificial intelligence model, or artificial thoughts, form an analog of the "solution" from [6]. The composition and structure of these molecules can be written in a graphical bracket notation from [10], which represents the forces of attraction and repulsion between artificial atoms and their strength.
These molecules are sources and results of activity of artificial mind itself or other kinds of artificial intelligence. For example, the atoms can represent primitive computing entities such as neurons in neural nets, transputers, or agents in agent-oriented models. Then the molecules correspond to neural nets with different topologies, transputer networks or self-assembling robots consisting of independent agents, software or hardware.
Results and discussion
We proposed a new concept of artificial intelligence and introduced the definitions of artificial mind, artificial thought and artificial brain.
Artificial thought is the material embodiment of a dynamic logical or computing structure consisting of dynamically interacting functions, just as a molecule consists of interacting atoms.
An artificial brain is a vessel in which artificial thoughts interact in a given model space in accordance with a set of reaction rules.
Artificial mind is a general notion for a new concept of artificial intelligence based on the principles of artificial chemistry and consisting of an artificial brain and artificial thoughts. Literally, in this concept, artificial thoughts interact with each other to form new artificial thoughts.
The advantages of the proposed concept are its decentralized nature and focus on the processes of emergence and self-organization. It also makes possible to store emerging dynamic structures as artificial thoughts using a form of graphic notation [10].
The disadvantages are related to the general problems of artificial chemistry models, for example, the limited capabilities of tools for organizing interactions between artificial atoms and molecules.
Conclusion and further studies
The proposed new concept of artificial intelligence can be used independently or together with other concepts and models, enabling the results of modeling or decision-making, such as emerging dynamic computing structures, to be stored in a tangible form. This can be used in the development of neural nets topologies, agent-based modeling, design of decision trees, etc. Future studies should be devoted to both improving the concept itself and expanding the possible scope of application.
References
1. Banzhaf W., Yamamoto L. Artificial Chemistries. Cambridge, Massachusetts; London, England: The MIT Press, 2015. 576 p.
2. Von Neumann J. Theory of self-replication automata / ed. by A. W. Burks. Urbana, London: University of Illinois Press, 1966. 416 p.
3. Artificial Life Models in Software / ed. by M. Komosinski, A. Adamatzky. London: Springer, 2009.
462 p.
4. Fontana W. Algorithmic Chemistry, Artificial Life II, SFI Studies in the Sciences of Complexity. Vol. X / ed. by C. G. Langton, C. Taylor, J. D. Farmer, S. Rasmussen. Redwood City, CA: Addison-Wesley, 1991. P. 159-209.
5. Banâtre J. P., Le Métayer D. The Gamma model and its discipline of programming // Science of Computer Programming. 1990. Vol. 15, № 1. P. 55-77.
6. Berry G., Boudol G. The Chemical Abstract Machine // Theoretical Computer Science. 1992. Vol. 96, № 1. P. 217-248.
7. Kruszewski G., Mikolov T. Emergence of self-reproducing metabolisms as recursive algorithms in an artificial chemistry// Artificial Life. 2021. Vol. 27, № 3-4. P. 277-299.
8. Cudennec L., Goubier T. A short overview of executing r chemical reactions over the £C and tC dataflow programming models // Procedia Computer Science. 2015. Vol. 51, Iss. C. P. 1413-1422.
9. Kol'chugina E. A. Spontaneous Emergence of Programs from «Primordial Soup» of Functions in Distributed Computer Systems // Automatic Control and Computer Sciences. 2018. Vol. 52, № 1. P. 40-48.
10. Kol'chugina E. A. Self-Synthesis of Programs Based on Artificial Chemistry Model // Programmnaya Ingeneria. 2022. Vol. 13, № 10. P. 440-448.
Information about the authors
Kol'chugina Elena Anatolyevna, Doctor of Technical Sciences, associate Professor, Professor of the department of mathematical support and application of electronic computing machines, Penza State University.
The author declare no conflicts of interests.
