Research on creativity modeling and machine implementation of artificial intelligence (hereinafter referred to as AI) has suddenly become a hot topic. Various creative software emerges in an endless stream. Its outstanding achievements include software that can write academic papers and beat the world Go champion. Multi-level artificial systems for creating works such as novels and paintings, etc. Basic theoretical research has also made outstanding achievements, and a new thing that can fill the gap in AI has been born - "computational creativity". It has two references. One refers to the creativity or AI creativity realized by artificial systems; the other refers to the AI branch field that specializes in how to make artificial systems express creativity and integrates theoretical discussion and engineering practice.
To create an AI branch of computational creativity with Chinese style, in addition to comprehensively and in-depth research on the successful experience of foreign computational creativity construction and completing the "make-up task", we should also overtake others in a corner, directly enter its frontier, and focus on studying the prerequisites, Forward-looking basic theoretical issues, focusing on the mind-cognitive philosophical issues, such as the prototype instance problem of creativity modeling, the possibility of computers realizing creativity, the "grounding" and "lack of authenticity problems" of software engineering, etc.
Disenchantment and Computationalization of Creativity: Model Considerations of Computational Creativity
AI allows computers to achieve creativity by modeling human or non-human creativity as models or "prototype instances". To do this, of course, we must first answer the prerequisite spiritual-cognitive philosophical questions, such as what exactly is creativity itself? Is there an independent creativity that is different from cognitive abilities such as thinking? Can its own structure, essence and secrets be solved? Open to human cognition? The trouble is that, if not all, the traditional view of innovation is at least partly an obstacle to research on computational creativity. For example, according to the traditional romantic and mystical view of innovation, creativity itself is a kind of mystery or mystery, or even a paradox. Even when creativity occurs in the world, it is a quality unique to the Muses.
To remove the above-mentioned barriers to computational modeling of creativity, we must undoubtedly try to bring creativity into the center of basic scientific research on AI, disenchant it, eliminate its mystery, and pull creativity down from the altar of unattainable cognition. , returning it to an objective process or force in nature. The so-called mysteries and mysteries in the world are relative to people's existing cognition. Things that were not clearly understood in the past were mysteries and were full of mystery. For example, thunder in the sky was considered a mysterious force in an era when science was underdeveloped. With the development of physics, it was disenchanted and returned to its original nature. status of natural phenomena. The same goes for creativity.
The reason why creativity is not mysterious and why it can be modeled by machines is intrinsically because it relies on our ordinary cognitive abilities, such as thinking, imagination, association, analogy, etc. When they are brought together in a certain way, innovation will emerge. In scientific language, the performance of creativity is determined by distributed cortical networks, and its reality does not depend on a single brain region; the neural basis of creativity changes with changes in task requirements and forms; most innovative tasks Completion is all about the dynamic coupling of the default mode network and the execution control network. Activation of the default mode network reflects the spontaneous generation of ideas or information from long-term memory, while activation of the executive control network reflects the process of constraining thinking to complete specific target tasks. Creativity is therefore a natural phenomenon that can be computationally modeled and machine implemented. In the case of divergent thinking, on which creativity is most dependent, they consist of the activation of nodes. If these nodes are connected very firmly, they behave like regular psychological phenomena. People have divergent thinking, but there are weak and indirect connections, which then awaken the system to hit the cerebral cortex with non-specific activation. As a result, common phenomena such as flashes of inspiration appear in people's spiritual lives.
As far as inspiration, epiphany, etc., which are the most cognitively closed and regarded as mental miracles, as long as human cognition has a way to enter, its mystery will slowly dissipate. Its so-called mystery is just that the way to solve problems is different from the standard way of analyzing problems. The characteristic of the latter type of approach is that it recognizes that the problem to be solved is easy to express in words and that such problems can be solved in a direct and logical way. Experiments have shown that when subjects are faced with a problem that can be solved logically, they use language to report the step-by-step steps they took to solve the problem. Problems that are suitable to be solved by methods such as inspiration are different. Subjects are unlikely to use logical arguments to solve problems. In this case, the problem-solving process will show indescribable characteristics. Despite this, such a process of solving problems with the help of inspiration is also completed by the biological brain. There will be no supernatural power in it. It is still a natural process that can be explained by science, such as the process in which relevant information is encoded and processed in a specific way. .
For machines to realize creativity, it is also necessary to calculate creativity. The so-called computationalization is to provide operational definitions for concepts, or to translate creativity into attributes that can be realized on artificial systems, to re-express creativity and its components in computational terms, and to reveal its essential characteristics of formal or symbolic transformation. Computationalization can also be understood as the formalization of relevant concepts in computational terms. There are various ways of formalization, such as algebraic formalization, logical formalization, etc. To meet the requirements of formalization, research like algebraic semiotics has been born. It attempts to logically formalize the structure of symbols, symbol systems, and their mappings. With the deepening of the discussion on computational creativity, many computational schemes for creativity with theoretical basis and practical value have been born, such as cognitive schemes, procedural schemes, situationalism schemes, computationalism schemes, etc. According to the computationalism plan, computing creativity means using computational terms such as conceptual space, heuristics, and search to explain creativity and reconstruct the concept of creativity, such as formalizing the components and mechanisms of innovative capabilities in machines. among. Only by doing this kind of work can machines realize these formalizations or parts of them, and then complete innovative tasks.
How is creativity AI modeling possible: Exploring the mechanism of computational creativity
Another project in the construction of the basic theory of computational creativity is to resolve the skepticism that computers and creativity are unrelated, because creativity is the miracle of the human mind and the thing that best embodies the essential characteristics of human beings. The computer runs according to the program, and everything it does is arranged by the programmer. Its characteristic is that it is programmed. Being programmed is the opposite of autonomy, which is an inevitable feature of creativity. Even if a computer can express so-called creativity, it should only be attributed to the programmer. The instructions and rules in the program determine all possible performances of the computer, and these cannot be surpassed.
But as long as you do research that keeps pace with the times, you will find that the above cognition is based on a narrow and outdated understanding of the program. According to new research on programs, the problem with the above view is that it does not see that the program contains changes in the rules itself, that is, the program contains rules that stipulate how to change, and can be embedded in "living algorithms" or even creative algorithms that change as the situation changes. algorithm. Furthermore, programs are embedded with algorithms that can learn and respond to unexpected inputs from the environment. Importantly, it also includes genetic algorithms, which make random changes to a program's task-oriented rules. These changes are similar to the point mutations and crossovers that drive biological evolution. Many evolutionary programs also include a fitness function that selects the best members from each new generation of task program members as the "parents" for the next round of random rule-based changes. When there is no fitness function, such a choice is made by humans, but with such a function, the machine can make it "by itself". This means that machines have a specific sense of autonomy and creativity due to changes in the concept of programming, and can also generate output that meets the two criteria of human creativity (i.e. novelty and usefulness). Taking evolutionary programming as an example, it can lead to preliminary transformational artificial intelligence, that is, to allow machines to have transformational creativity. For example, some images generated by a program are completely different from the original images, which are new and useful images. This is so because genetic algorithms allow not only point mutations within a single programmed instruction, such as changing a number, but also the continuous and hierarchical nesting of entire image-generating programs.
Since how AI modeling and realization of creativity is possible is both a theoretical issue and a practical issue, and the latter aspect is more fundamental and critical, we can take a two-pronged approach and explore the expression of creativity in artificial systems from both theory and practice. How to solve the problem and focus on solving key problems in engineering practice. In fact, AI adopts a strategy of discussion and practice, and focuses on how to design artificial systems with greater innovative capabilities. It has achieved a large number of world-renowned results, such as the aforementioned ability to write innovative papers and creative works. Software from literary and artistic works, such as AlphaFold, which can make the most accurate predictions of the most difficult-to-predict protein structures far exceeding those of human scientists. That being the case, there now seems no need to waste precious human energy on the question of whether creativity is possible. In fact, computational creativity research already has this way of progress, which is to put aside grand theoretical questions such as "whether it is possible", and on the basis of dissecting specific forms of creativity, do some specific and small things to allow artificial systems to achieve creativity. work.
Software Engineering Innovation: The Technical Key to Computational Creativity
Software engineering is the process of applying systematic, strictly constrained, and quantifiable methods to the research and practice of engineering technologies such as software development, operation, and maintenance. In the study of computational creativity, software engineering is both its main driving force and its main work, such as researching, designing, and writing creative software in the application fields (painting, games, scientific discoveries, etc.) in which it is engaged. In philosophical terms, it is a veritable "big nose" because no matter how much disenchantment, computation, and model building work is done for creativity, it must ultimately be implemented and realized through software.
It should be admitted that at the beginning, because people held this understanding of the nature and role of software, that is, designing software is nothing more than writing codes and algorithms, so most of the software that appeared in computational creativity became a means to achieve an end. If awareness and practice remain at this level, the ideal of computational creativity and creative creativity cannot become a reality. Based on careful reflection and research on software engineering from the perspective of innovative software, people have made such adjustments to the goal, that is, making the codes and algorithms generated by the software an innovative achievement at the same time, and making the software a creative software generator. One of its functions is to pose questions to the world, not just solve them. To do this, the methodology must be changed. Based on this understanding, the code in computational creativity programming is not just a tool like elsewhere, but can be like the results or processes in science or art, that is, such code also has its own life and can Be studied, modified, applied in unforeseen fields, can be admired by culture, etc. To design and develop software according to this concept is not only to engage in engineering and technical work, but also to engage in philosophical discussions on creativity. The performance is that philosophical issues such as the nature of creativity will definitely be rethought here. According to new research, the role of creativity is not just to solve problems, but more importantly to ask questions about the world, or problematize the world. By problematization I mean that the generated code exposes opportunities that either help to better understand the world through problem solving, such as exposing an unexpected anomaly or hypothesis about a data set, or that applies the code to changes to change the world.
The key work of software engineering is programming, because computers express creativity through programs. Specifically, in order for machines to express creativity, in addition to studying creativity and computing it so that it can be implemented in programs, we must also explore the affordability of programs, such as what is the relationship between programs and creativity, and whether creativity can be realized. Strength, to what extent it can be achieved, etc. Seeing this, many computational creativity research experts practice this engineering approach, that is, first study the affordability and nature of the program, and then use this to calculate the creativity to solve specific engineering implementation problems.
The “loss of authenticity conundrum” and its resolution: “grounded” modeling of computational creativity
Calmly reflecting on the existing computational creativity software and programming research work, philosophers and some AI experts who focus on basic theoretical work have admitted that existing software that simulates various intelligent phenomena has a "lack of authenticity problem." . This problem is actually a manifestation of what Searle et al. calls the “lack of intentionality problem” in modeling computational creativity. As long as you examine it, you will find that there are two situations when it comes to creativity displayed by humans: First, the system truly realizes creativity, such as the system is either a true collaborator in innovation or an autonomous subject that can independently innovate; second, the system is used to realize creativity. Evaluated and interpreted as creative, that is, having a superficial sense of creativity. According to the internalist point of view on creativity, existing creative software is only evaluated as creative, but not really creative. This is the problem of lack of real creativity.
In order for computing systems to be truly creative, we must first clarify the reality of human creativity. For example, what does it mean to say that they are truly creative? What are the standards and expressions of authenticity of creativity? The reason why human creativity is real is that, in addition to being really determined by human purpose, motivation, and power and can be adjusted at any time, it is also embedded in, penetrated into, and integrated with human culture. The constraints and influences of culture also serve culture. The scope of "authenticity" can also be expanded to many aspects of life, such as whether the description of one's own experiences and experiences is true. Furthermore, in order to make artificial innovation systems have real creativity, one of the ways out is undoubtedly to study the roots and conditions of the authenticity of human creativity. As long as we analyze it, we can find that the reason why people and their creativity are real is because people and their creativity have the characteristic of "grounding", that is, living and being embodied in their world, as Heidegger said , people exist in the world. Looking back at the computing system, the reason why its creativity is unreal is because it is not grounded and has no basis in life. Therefore, to solve the problem of authenticity, the key is to solve the "grounding" problem of computational creativity software, that is, when designing truly creative software, let it have its own life world, let it be embedded and embodied in its in the world. In fact, computational creativity research experts are already working on solving this problem, and the concept of "situationalist computational creativity" is its positive result. Of course, to solve the problem of lack of authenticity, specific engineering and technical research is also a necessary condition. The key here is to solve the problem of how to automatically generate code. To realize this vision, two principles must be adhered to: first, the world should be problematized; second, creating programs should be seen as work in its own right, not just a means to an end. In this way, automated code generation provides a suitable testing ground for cutting-edge computational creativity techniques. Here, the role of conversational generation technology is also very important, because only through it can users believe that the generated code products are useful, and then solve relevant philosophical issues, such as how computing systems can have autonomy, intentionality, etc. .