Modeling creativity in artificial intelligence: possibilities and limits

Full Text

The key issue of the philosophy of artificial intelligence is the designation of the boundary of the modeling of natural intelligence in artificial systems (Hawk, 2021). In connection with constant discoveries in the field of artificial intelligence in recent decades, the ideas about this border have changed significantly. The appearance of generative neural networks in the XXI century has opened up new opportunities for modeling human intelligence. The structure of natural intelligence includes both functions that are successfully modeling, for example, images recognition, and those that are not yet implemented in artificial intelligence systems, primarily creative and creative activity. This problem was set by Ada Lavelis, who argued that the main criterion for the presence of intelligence is creativity (Stein, 1985). The creation of neural networks that generate images, music and texts makes us look at the problem of creative activity in a new way, and the possibility of its transfer to the car. The ratio of the concepts of "creativity" and "creativity" in the context of artificial intelligence The category of "creativity" is introduced by Aristotle, who defines him as the ability of a person to find new solutions and the creation of new things (Aristotle, 1927). For him, the ability to work was an integral part of human nature, and he believed that it could manifest itself in various fields of life, including art, science, philosophy and technology. This gave rise to a wide interpretation of the concept of creativity, within the framework of which creative activity is defined “as the processes of generation of the new” (Nesterov, Demin 2020, p. 84). Within the framework of this approach, the meaning of creative activity is seen in the “increase in the diversity of the human world in the process of cultural migration” (Encyclopedia of the epistemology and philosophy of science 2009, p. 954). Creative activity is not limited to the actions of the subject and applies to society and culture as a whole. The setting of the task of modeling creativity requires its concretization. I.A. Beskova and I.T. Kasavin associate creativity with the level and quality of the cognitive activity of the individual and define it as “hierarchical structured unity of abilities, which determines the level and quality of mental processes aimed at adapting to changing and unknown conditions in sensorimotor, visual, operational-activity and logical and theoretical forms "(Ibid.). The study and modeling of the creative activity of the subject in the cognitive sciences required a differentiation of understanding of creativity as a cultural phenomenon and the type of cognitive activity. Creativity as a scientific term is introduced by the psychologist D. Simpson, who defines it as “a person’s ability to abandon stereotypical and historically established ways of thinking” (Simpson 1948, p. 172). J. Gilford in the work “Nature of human intelligence” developing the concept of creativity, offers a similar approach and describes it as “the ability to generate new ideas that differ from traditional thinking schemes, as well as skills that contribute to the solution of problem situations” (Guilford 1967, p. 197) . In this work, we will rely on the definition that M. Boden gave the defined creativity as “the ability to invent ideas and artifacts that are considered new, unexpected and valuable” (Boden 2004, p. 16). Summarizing these definitions, it can be noted that creativity is considered as building new opportunities for the subject, and creativity as the creation of new opportunities for culture. The key problem in determining creativity is the understanding of what is considered new and what new object should appear so that we consider it the result of creative activity. To solve this problem, three levels of novelty can be distinguished: ontological, subjective and semantic. The ontological level of novelty can be attributed to everything that was not before. With this approach, any object that has not previously been in space and time will be considered new regardless of the degree of novelty of its characteristics. In this sense, any image generated by a neural network can be considered new. Having studied for some set of examples, the neural network actually continues the sequence available in the study sample. In this case, we are not talking about creating new meanings or qualitative changes in something, but there is a continuation in new objects of what already exists. The second, subjective, level of novelty lies in the fact that something new and unexpected for the subject is created, but at the same time as a result of a more thorough analysis, the subject can describe the algorithm for obtaining such a result. Knowing this algorithm, another subject can come to the same or similar result, regardless of its complexity. At the highest, semantic, level, there are processes of creating objects that are currently not amenable to description and algorithmization. In this case, we do not understand how they are obtained and cannot “decompose these processes into atoms”. G. Haken expressed a similar thought, who defined creativity as “the birth of ideas that were never born before and, moreover, the birth of which is highly unlikely” (Haken 2001, p. 313-314). This level is characterized by finding non -standard methods of solving problems. The complexity of modeling such activities is associated with its uniqueness and non -consumption. Creative activity of this level cannot be implemented in algorithmic, programmable models. Creativity Modeling: Problem As has already been mentioned in the introduction, the problem of creating machines capable of creative activity was set by Ada Lavelis. In modern studies of artificial intelligence, a whole area has been formed, specializing in the development of creative models and systems. Creative artificial intelligence in this work means a direction in artificial intelligence, in which machine learning methods are used to generate new objects based on patterns found in available data. The main methods and approaches to the creation of creative systems of artificial intelligence have been developed a long time ago, but only in recent years the developers received on the one hand, the opportunity to use quite large computing power, and on the other hand, the rapidly developing sphere of creative industries appeared, to solve the problems of which these models necessary. To assess the creativity of artificial intelligence, the modified Turing test is widely used (Turing 1960). Currently, there are many of its options in particular: musical, humorous, creative Turing test. To pass the test, the machines set the task of creating a work that will be indistinguishable from the results of a person’s creative activity. Further, the expert community is proposed to determine which of the works are created by a person and which machine (Winters, Nys, De Schreye 2018). An example is a survey conducted by us, during which the respondents were offered two jokes, one of which was written by a person, and the second GPT-4 chat. As a result, 35% of respondents expressed the opinion that the second joke was written by man. There are more successful examples of passing creative tests of turing. Since cars often win such tests, it must be recognized that either we must agree that artificial intelligence has become creative, or revise the turing test itself. An important issue of developing creative artificial intelligence is whether artificial intelligence itself can determine the level of creativity of the objects he produced. The above definition formulated three signs of creative ideas or artifacts, namely novelty, surprise and value. The first parameter of novelty can be appreciated by the machine through correlation with existing analogues. An example is the competition generative network “Can: Creative Adversarial Networks” (https://arxiv.org/abs/1706.07068), designed to create new artistic styles and evaluate their novelty. It consists of two neural networks. The task of the first is to create a new style of fine art on the basis of existing examples and add it to the available database. The second part of the system analyzes the data and tries to identify which of the styles were offered by the machine. This model is simultaneously able to generate new objects and evaluate their novelty. The parameter of surprise is subjective and is determined by the emotional reaction of a person’s surprise. Value is a parameter that is introduced by the subject. Theoretically, the machine can evaluate functionality, but the endowment of an object by meaning is carried out only by a person. The scope of creative artificial intelligence includes not only the traditional types of creative activity, such as writing music, poems, creating images, but also the creation of forecasts for the development of diseases, the design of prostheses in medicine, management of monetary assets on stock exchanges. But all these systems are not capable of goal -setting and cannot initiate creative activities. Therefore, often such systems are used as creative assistants. Setting a goal, objective and evaluation of results is carried out by a person, and the use of artificial intelligence allows you to expand the possibilities of the subject. 

The possibilities and boundaries of humor modeling in creative artificial intelligence One of the areas of development of creative artificial intelligence was the modeling of humor. The ability to joke and understand humor has always been considered exclusively a human line. The problem of modeling humor helps to more accurately understand the border between what we can teach the machine and the fact that remains an exceptional feature of human intelligence. Humorous activity is difficult to simulate, as it is comprehensive. The generation of humor requires a high level of ownership of the language, flexibility and mobility of thinking, creative abilities, the ability to identify contradictions, knowledge of the laws of logical thinking and the ability to violate them, the ability to respond to relevant events and understand the social context as a whole. Despite the difficulty of modeling humor, modern neural networks successfully cope with a number of tasks in this area. So, cars generate jokes, stand-up committees create monologues and perform with them, generative systems, such as GPT, write a humorous performance script, come up with jokes and puns (https://openai.com/gpt-4). The study (Nijholt 2018) compared the perception of jokes written by volunteer people and the GAG artificial intelligence system. As a result of the experiment, it turned out that the jokes that were written by the machine were twice as often as ridiculous than those that were written by people. The key advantage of such systems is the ability to learn. Consequently, a neural network is like a person to learn from his mistakes and improve his capabilities of humor generation. Examples of generative systems show that neural networks are able to correctly reproduce the structure of jokes, identify contradictions and paradoxes, use a tool such as language games. At the same time, neural networks have a number of significant restrictions. A key problem is the lack of a sense of context, a sense of space and time. In addition, the most important condition for successful writing jokes is the ability to laugh at them. People consciously choose the theme and format of humor, and cars do not have their own goal -setting. As a conclusion, we give an answer that GPT-4 gave us the question of what the capabilities and boundaries of the Machine Generation of humor: “As for my capabilities, I can offer jokes and jokes on various topics, but I have no sense of humor, therefore I can’t laugh or evaluate jokes. ” Conclusion In Russian -speaking literature, it is customary to share the concepts of “creativity” as an activity aimed at the development of society and culture as a whole, and “creativity” as a construction of new opportunities for the subject. In relation to artificial intelligence, the use of the term “creativity” is correct. By creative artificial intelligence is understood to mean a direction in artificial intelligence, in which machine learning methods are used to generate new objects based on the laws found in the available data. In modern creative artificial intelligence, two main tasks are distinguished. The pragmatic task is to expand the creative abilities of a person by creating tools. The research task is to model human creativity. The technology developed to date can be summarized by the term “weak creative artificial intelligence”, as they can create new objects, but are not capable of goal -setting and reflection. An example is humor generation systems that can create jokes, but do not have a sense of humor. Despite this, the possibilities of artificial intelligence are constantly expanding, which changes our idea of the boundaries of the modeling of natural intelligence.

About the authors

Nikita A. Krylov

Vologda State University

Author for correspondence.
Email: krylovna@vogu35.ru
ORCID iD: 0009-0009-1378-0548
SPIN-code: 9300-0207

PhD student, a lecturer at the Department of Philosophy

Russian Federation, 15, Lenina str, Vologda, 160035, Russian Federation

References

Supplementary files