Comparison of Cognitive Differences of Artworks between Artist and Artistic Style Transfer

Abstract

This study explores how audiences responded to perceiving and distinguishing the paintings created by AI or human artists. The stimuli were six paintings which were completed by AI and human artists. A total of 750 subjects participated to identify which ones were completed by human artists or by AI. Results revealed that most participants could correctly distinguish between paintings made by AI or human artists and that accuracy was higher for those who used “intuition” as the criterion for judgment. The participants preferred the paintings created by human artists. Furthermore, there were big differences in the perception of the denotation and connotation of paintings between audiences of different backgrounds. The reasons for this will be analyzed in subsequent research.

1. Introduction

Artistic creation can be regarded as one of the ways that humans express their cognition of themselves and the interweaving of various things in their lives and experiences. Along with literature, dance, drama, and music, art is a medium of self-expression. The intervention of artificial intelligence (AI) in artistic creation has changed the mode of creation and brought unlimited possibilities. It is believed that human decisions are the foundation of science and technology. Therefore, before we talk about whether the work completed by AI is art, we should make it clear that, when we put together the paintings completed by AI and human artists, we asked the audience to correctly find out which one was completed by a human artist.

The art of painting has long been regarded as an imitation of nature and the re-production of things in the real world. The pace of technological progress is accelerating, and the application of AI in the fields of art and design has also become more mature [1,2,3]. When technology evolves from being an auxiliary tool of artistic creation to its medium, it has a great impact on our understanding of the nature of artistic creation and how to appreciate and interpret artworks.

Relatively speaking, existing research, which focuses on the development and application of algorithms, and audience acceptance and viewing experience of artworks completed through AI, still has a lot of space to be further explored and discussed. Most people have a stereotype of aesthetics or design, thinking that it is the profession of an artist or designer. Sense of beauty is innate but because we have specialized it and imbued is as being a gift, we think that this talent belongs to a specific few artists or designers. We must not give up our intuitive feelings for beauty. The best proof is that more and more amateur artists are constantly changing their modes of artistic creation and appreciation through their efforts.

In the past, experiments or research have allowed computers to imitate human behavior or work. Our research team did the opposite, inviting the amateur artist to simulate and synthesize the paintings by AI, and then create their own works. From this point of view, one may ask: What are the main differences between the paintings created by AI and by human beings? Specifically, for this article, we must first clarify whether the audience can correctly distinguish the paintings completed by AI or human beings. What is the basis that influences their judgment? Among different ethnic groups, is there a consensus on cognition of the paintings completed by AI or human beings? Or are they completely different?

On 26 April 2019, BBC Science and Technology reporter Eleanor Lawrie pointed out: “Can a computer, devoid of human emotion, ever be truly creative? Is this portrait really art? Does any of that matter if people are prepared to pay for it? And as artificial intelligence evolves and eventually perhaps reaches or surpasses human level intelligence, what will this mean for human artists and the creative industries in general? [4]” This article is trying to answer the question raised by Lawrie.

1.1. From Trying to Popularizing: The Implementation of AI in Painting and Its Value

In 2005, futurist Ray Kurzweil predicted the coming of the era of AI in his book ‘The Singularity is Near’ [5]. As AI technologies, such as AlphaGo, are widely used in many fields [6], research and applications on AI have begun to emerge. In recent years, the development of computer vision technology has provided opportunities to analyze paintings through high-level calculation methods [7,8]. These calculation methods can increase the knowledge and abilities of artists, scholars, and curators [9]. Machine Learning (ML) is a technology that realizes AI and Deep Learning (DL), such as Artificial Neural Networks (ANNs), which are currently one of the most widely used types of machine learning. Deep learning methods include many hidden units of deep layers [10], which allow the learning of complex relationships in data [11].

The first application of neural networks to the field of artistic generation is ‘A Neural Algorithm of Artistic Style’, based on a neural network proposed by Gatys et al. [12], which can realize the transfer of any style. Since then, there has been endless research on styles, conversion efficiency, application fields, and related technologies. According to training and feature extraction methods, this mode can be divided into “Paired” and “Unpaired”. The so-called “Paired” method is a pre-trained model. The method involves using a style map and a content map to complete artistic style conversion after the computer’s deep learning, such as the method of Gatys et al., AdaIN, WCT, etc. The “Unpaired” method involves extracting the common features of multiple images from a data set, and then using an algorithm to complete the purpose of style transfer, without any paired training stimuli, such as Pix2Pix, CycleGAN, etc.

In short, the learning and imitation effects of algorithms are the key to perceiving the value of art in current research, without considering the feelings and cognition of the general audience. Therefore, this research can be valuable as a bridge to span the gap between technology and human nature. Based on this, the head of the research team has also carried out a series of researches in recent years. For example, Lin, Chen and Lin [13] discussed how to use desktop virtual reality (Desktop VR), or head-mounted display virtual reality (HMD VR), to enhance the audience’s experience when viewing artworks. In addition, Lyu, Lin, Lin and Lin [14] proposed a framework for evaluating artistic style conversion from a cognitive perspective, based on the artist’s coding attributes (Color, Stroke, Texture) and the audience’s decoding awareness level (Technique, Semantics, Effect). It can be used to explore the audience’s perception of the transfer of artistic styles.

Actually, the AI algorithms are quite varied. Selecting and implementing a detailed AI algorithm to imitate artwork will have a significant impact on outcomes. There can be large differences in terms of quality between “the paintings created by AI”, due to the different AI algorithms used. It should be noted that the focus of this research is whether the audience can distinguish paintings completed by AI or human beings, and what criteria their distinctions are based on. Therefore, this article does not involve a comparison between “paintings generated by different AI algorithms”. However, this will be carried out in a follow-up study.

1.2. Amateur Art and Artist

Amateur art is usually defined as visual art that is created by a person who lacks the formal education and training that a professional artist undergoes [15,16,17,18]. When all kinds of creative or painting materials and tools have become more and more abundant, anyone who wants to create art can put artistic tools to creative work anytime, anywhere. Since amateur artists have not received the basic training of art school, they can express their ideas more freely and casually, without worrying about restrictions, such as what to draw first, how to compose the picture, how to shape it, etc. Many studies have conducted in-depth research on the creative mode, communication mechanism, display method of amateur artists and the audience’s cognition of the interpretation of art works [19,20,21,22], which provided much inspiration for this article.

Whether it is the history of art or design, in a sense it is a history of “imitation” and “innovation” [23]. Imitation is different from “copying” in general understanding. So, it can be said that: “The creative process should be based on close imitation of the masterpieces of past creators.” The inspiration and enlightenment obtained from imitating others’ works are “decoded” by the creator, and then “coded” according to his or her ideas and creative goals, and then transformed into new work [24,25].

The expression of creativity is not limited to the form of expression but pays more attention to the ingenuity and connotation behind the creativity. If imitation is the study and application of the external form (denotation) of painting creation, then creativity is more dependent on the inner meaning of the work and the creator’s thoughts (connotation). The gradual process of “copying-sketching/creating”, for creators, is their way into the art world, from seeing and understanding to being moved by the work. This recognition is also reflected in the process of others looking at their work.

What’s more, the use and actual effects of different tools and media cannot be effectively grasped if you have not operated and experienced them, but just watched the operations of others.

1.3. From Creation to Appreciation: Artistic Creation and Cognition

Artistic creation is an expression of the artist’s pursuit of beauty. It has the following two inter-influencing characteristics: (1) Experience the “connotation” with “form”; (2) Enrich the “form” with “connotation”. “Painting” has a “form” context that trans-forms abstract “connotation” into concrete “form”. Form (style) and connotation (idea) are different in expressing artistic creation [21]. How does the relationship between “intention” and “form” become the concept of creative thinking in painting? There seems to be a certain degree of correspondence between “form” and “intention”, that being: to find clues to “connotation” in “intention”. From the perspective of the symbolic communication model, “artistic creation” occurs in the process of “coding” by the artist and “decoding” by the audience [24,25,26]. Therefore, from the perspective of “decoding” of readers, exploring the cognition of “artistic creation” is helpful to understand the artistic process of creation [21]. Such principles and models also apply to our viewing and evaluating of works created by AI.

1.4. Purpose

To a certain extent, the machine can also be regarded as a work of art; in fact, the machine has always been involved in artistic creation [27,28]. The machine is operated by people or controlled by programs designed by people. Therefore, it can be said that the machine itself does not create art, but is executed by the “people” behind the machine. So, in terms of the relationship between AI and artistic creation, in addition to scientists and engineers constantly innovating on the technical level, the intervention of artists is equally important and may even produce unexpected results. General experience tells us that artworks are created by so-called artists (professionals). However, the existence of a group of amateur artists first broke the notion that art can only be created by professionals. AI, to some extent, is not an artist, but more like a generalist scientist (closely related, of course, to the big data behind it and the setting of the algorithm). Therefore, the audience can effectively recognize the transfer of artistic style through the computer, which will have a great impact on the mode of AI intervention in artistic creation.

In short, while AI provides various possibilities for artistic creation, we may need to think about how to evaluate artistic creation under the intervention of AI from the perspective of users. So, this research is an attempt to respond and explore the question “ Can machines create art?” Can work created by AI be considered art? If the answer above is ‘Yes’, does that mean we can treat everything as art? What kinds of things cannot be counted as art? The audience’s ability to distinguish paintings originating from different creative modes, and the factors that influence their judgment, will be investigated.

2. Materials and Methods

2.1. Stimuli

In fact, there is not just one AI algorithm. The selection and detailed implementation of an AI algorithm to imitate artwork will greatly impact the results. There can be large differences in terms of quality between “the paintings created by AI”, due to the different AI algorithms used. It should be noted that the focus of this research is whether the audience can distinguish paintings completed through different creative modes and what criteria their distinctions are based on. Therefore, this article does not involve a comparison between “paintings generated by different AI algorithms”. However, this will be carried out in a follow-up study.

As part of a series of studies, this study argues that home is unique to each individual’s mind. Therefore, we have made ‘Home: Sweet Home’ a central theme in the creation of paintings for our study. Although this study does not address the audience’s attitude towards the subject of the painting, it is a prerequisite for the continuation of this series of studies. Therefore, we will also follow this principle in the process of selecting paintings.

Figure 1 shows how these stimuli are produced. The details are as follows:

• The author invited experts in the field of art and aesthetics to select six world-famous paintings (Row 1 in Figure 1) with the theme of “Home: Sweet Home”.
• Using AI technology, the computer simulated the six original paintings and obtained six new paintings, as shown in the second row in Figure 1.
• We invited amateur artist A to imitate the six paintings after the computer simulation and obtained the new paintings (Row 3 in Figure 1). Then, we asked artist A to create six new paintings (Row 4 in Figure 1), based on his or her views on the paintings created by AI.
• At the same time, we invited another amateur artist, B, to create directly from the painting completed by AI, and obtained six paintings (Row 5 in Figure 1).

2.2. Research Design and the Content of the Questionnaire

This study hypothesized that people would use “intuition” to make judgments when confronted with paintings where it could not be determined whether the creator was human or AI, and that they would also rely on the basic elements of the painting. Such a judgment process may be short-lived, but they apply to the models constructed by the relevant theories mentioned in this section (see Figure 2), and these six attributes further constitute the evaluation criteria for this study; which form the basis for examination of how the audience distinguishes between different creators.

After the content of the questionnaire was drafted, the authors invited more than ten teachers and students from the fields of art and design to fill out the questionnaire and then asked them to give their opinions on the framework of the questionnaire, the setting of questions, and the attributes chosen for evaluation. The authors made three major revisions based on opinions given (and asked them to review each revision). In addition to personal information, the main part of the questionnaire consisted of two parts (see

Table A1. Example of the stimulus assessment section of the questionnaire.

Part I Take P01 as Example
	Q01. What do you think is the rigor of the composition of this painting? Low ← [ ] 1 [ ] 2 [ ] 3 [ ] 4 [ ] 5 → High Q02. How do you like the color matching of this painting? Low ← [ ] 1 [ ] 2 [ ] 3 [ ] 4 [ ] 5 → High Q03. How do you think the details of this painting are portrayed? Low ← [ ] 1 [ ] 2 [ ] 3 [ ] 4 [ ] 5 → High Q04. How do you feel about the feeling about “Home” conveyed by this painting? Low ← [ ] 1 [ ] 2 [ ] 3 [ ] 4 [ ] 5 → High Q05. What do you think of the creative mode of this painting? [ ] Computer Simulation [ ] Amateur Artist Imitation [ ] Amateur Artist Creation
Part II Take G01 as an Example
	Q01. Would you please, according to your instinct, judge which one is imitated by an amateur artist? [ ] A [ ] B Q02. What is your confidence index of the previous question?Low ← [ ] 1 [ ] 2 [ ] 3 [ ] 4 [ ] 5 → High Q03. Which of the following attributes did you use to distinguish paintings imitated by amateur artists? (Select One or More Answer Choices) [ ] Line [ ] Color [ ] Light and Shade [ ] Stroke [ ] Detail [ ] Only “intuition” Q04. Which one do you like best? [ ] A [ ] B

).

We considered that the teachers and students had filled out the questionnaire many times, and knew the creative modes and code names of different paintings. Therefore, the questionnaire they filled out would not be included in the formal statistical analysis. All participants in this study saw the relevant paintings and questions for the first time, which helped us better understand their attitudes.

2.3. Participants

The questionnaire was officially launched on 12 April, 2021, and we waited 19 days in the hope that more people would participate. In the end, a total of 750 valid questionnaires were received (see

Table 1. Profiles of the participants.

Category		n (%)
1. Gender	Male	308 (41.07%)
	Female	442 (58.93%)
2. Age	Under 20	87 (11.6%)
	21–40	312 (41.6%)
	41–65	300 (40.0%)
	Over 65	51 (6.8%)
3. Educational Attainment	High School	48 (6.4%)
	College	61 (8.13%)
	University	327 (43.6%)
	Graduate	314 (41.87%)
4. Background	Pedagogic	69 (9.2%)
	Humanities	31 (4.13%)
	Fine Arts	77 (10.27%)
	Design	323 (43.07%)
	Sociology	38 (5.07%)
	Science	19 (2.53%)
	Engineering	68 (9.07%)
	Other	125 (16.67%)

n = 750.

).

It is important to note that the participants were treated as a whole and divided into three groups, based solely on how they identified the creators of the works (see Table 4 for details). Cognitive differences between different participants (e.g., audiences with different genders) will be explored in subsequent studies and will not be covered in this article.

2.4. Procedures

This research is divided into parts (see Figure 3). First, the literature was reviewed to clarify and grasp the general situation and latest achievements of AI in painting creation. At the same time, a summary was made of the characteristics of amateur artists and related cognitive models in artistic creation. Second, based on previous research and the opinions of experts with artistic and aesthetic backgrounds, stimuli were screened, questionnaires were drawn up, and content validity analysis of the results was carried out. Finally, analysis and discussion were conducted based on the answers of all participants, and the research conclusions and follow-up research ideas summarized.

Participants were asked to carefully read the introduction and related instructions of the questionnaire and answer the questions in order. In the first part, they evaluated the six paintings, including the mode of composition, the color matching, the depiction of details and the connotation of the paintings. At the same time, they judged the creative mode of the stimuli based on subjective intuition. The second part focused on identifying the creative mode of the stimulus and its confidence index, and the criterion for making judgments. Finally, they chose their favorite painting.

2.5. Statistical Analysis

After a preliminary inspection, all questionnaires met the requirements of a valid questionnaire and could be used for further statistics and analysis. Data analysis and interpretation were carried out in the following order: First, descriptive statistical analysis was used to determine the accuracy of the audience in distinguishing the creation mode of the stimuli; Second, One-way ANOVA was used to analyze whether the audience had cognitive differences in related evaluation criteria and for stimuli that reached the 0.05 significance level. The specific differences were found through post-hoc comparison. A discussion of the stimuli with large cognitive gaps will be undertaken.

3. Results and Discussion

This research assumed that the audience could distinguish the stimuli in different modes, but the accuracy would fluctuate to a certain extent when the participants were faced with different paintings. Furthermore, it was assumed that participants would be more willing to use “intuition” to distinguish the stimuli in different modes. Therefore, this section will analyze and discuss the data from the following three perspectives: First of all, the audience’s ability to distinguish the completion of the stimuli in the different modes. The analysis included accuracy and confidence indices and whether there were cognitive differences between the groups. Next, the criteria used by the audience to judge different creative modes and the one having the greatest effect were analyzed. We used “intuition” as another criterion to make judgments for group differences. We then explored the accuracy of the groups and whether there were cognitive differences between them. Finally, we analyzed any differences in the audience’s perception of the stimuli from the same painting completed by different modes.

3.1. Identified the Creative Mode and Preference

Question 5 in the first part of the questionnaire, and question 1 in the second part, mainly reflected whether the viewer could correctly distinguish paintings completed in different modes. The accuracy for these two questions was 28–67.2%, and 65.3–76.4% respectively.

In the first part, the stimuli were divided into three categories according to the creative mode, and the accuracy was then summed and averaged. The participants who could correctly distinguish “computer simulation”, “amateur artist imitation” and “amateur artist creation” were 52.8%, 28%, and 46.7%, respectively. This showed that when there is no reference, the audience’s accuracy for distinguishing different creative modes is not ideal.

In part 2, over 65% of the audience could make a correct judgment on the four stimuli (G01, G02, G03, and G06). The number of people who correctly judged the creative modes of G04 and G05 were 76.4% and 74.5%, respectively. In addition, those participants who made the correct judgment had an average confidence between 3.56 and 3.68. The audience’s favorite paintings were consistent with the paintings they judged to be imitated by amateurs, and there was a certain gap between the voting rate and the accuracy (see

Table 2. The accuracy, confidence index and the preference for G01–G06.

Stimuli	Imitated by Amateur Artist	Accuracy	Confidence Index	Preference
G01	B	497 (66.3%)	3.62 (0.971)	454 (60.5%)
G02	A	511 (68.1%)	3.63 (0.963)	417 (55.6%)
G03	B	499 (66.5%)	3.64 (0.979)	455 (60.7%)
G04	B	573 (76.4%)	3.68 (0.946)	449 (59.9%)
G01	B	497 (66.3%)	3.62 (0.971)	454 (60.5%)

n = 750.

).

The accuracy of the stimulus in the second part was much higher than that in the first part. Perhaps the second part of the stimuli was related to the reference object, or it may have been because the audience had gradually become familiar with the stimulus after the first part. Although in the first part the accuracy of the four stimuli did not reach 50% and the accuracy of P05 was less than 30%, the answer rate for the remaining stimuli were relatively in line with expectations. Therefore, we can further explore the factors affecting the audience’s accuracy. In addition, according to the answer in Q3 of the second part of the questionnaire, the audience could also be divided into three groups, as shown in

Table 3. Division of the audience into three groups, based on different criteria.

Groups	n (%)
Group I: Just rely on “intuition” as a criterion.	34 (4.5%)
Group II: Rely on “intuition” and “other 5 attributes” as the criterion.	155 (20.7%)
Group III: Only rely on the “other 5 attributes” as the criterion.	561 (74.8%)

n = 750.

.

Table 4. The accuracy of the stimuli (G01–G06) was judged by the three groups based on “intuition”.

Stimuli	Group I (n = 34)	Group II (n = 155)	Group III (n = 561)
G01/B	22 (64.7%)	93 (60.0%)	382 (68.1%)
	Group III > Group I > Group II
G02/A	25 (73.5%)	99 (63.9%)	387 (69.0%)
	Group I > Group III > Group II
G03/B	24 (70.6%)	93 (60.0%)	382 (68.1%)
	Group I > Group III > Group II
G04/B	25 (73.5%)	114 (73.5%)	434 (77.4%)
	Group III > Group I = Group II
G05/A	28 (82.4%)	117 (75.5%)	414 (73.8%)
	Group I > Group II > Group III
G06/A	23 (67.6%)	98 (63.2%)	369 (65.8%)
	Group I > Group III > Group II

n = 750; Groups that rely solely on “intuition” as the criterion for judgment is marked in red.

and

Table 5. Cognitive differences between different groups.

Stimuli	Question (Core: 1–5)	Mean (SD)			F	Post Hoc Comparison
		Group I (n = 34)	Group II (n = 155)	Group III (n = 561)
G01	Part II-Q02. What is your confidence index of the previous question?	3.32 (1.34)	3.24 (1.10)	3.67 (0.88)	13.533 ***	Group III > Group I; Group III > Group II
G02		3.35 (1.25)	3.32 (1.03)	3.70 (0.87)	11.661 ***
G03		3.32 (1.32)	3.34 (1.03)	3.64 (0.90)	6.887 **	Group III > Group II
G04		3.21 (1.34)	3.41 (0.96)	3.75 (0.91)	11.842 ***	Group III > Group I; Group III > Group II
G05		3.29 (1.31)	3.32 (1.00)	3.65 (0.90)	9.016 ***
G06		3.32 (1.22)	3.37 (1.03)	3.70 (0.88)	9.456 ***

n = 750; ** p < 0.01, *** p < 0.001.

show the accuracy and confidence of these three groups in distinguishing paintings imitated by amateur artists. Specifically, from the perspective of the accuracy of distinguishing paintings, the accuracy of the three groups were all above 60%. The accuracy of the group based solely on subjective intuition (Group I) fell between 64.7% and 82.4% (see Table 4). Except for G01 and G04, the accuracy of Group I was higher than Groups II and III; while for G01 and G04, the accuracy of Group I was lower than Group III, but slightly higher, or equal to, Group II (see Table 5). The confidence index of the audience in Group I, who judged by “intuition”, was relatively lower than that of the other two groups (see Table 5).

3.2. The Cognitive Differences of the Audience

In the first part, among audiences of different genders, ages, backgrounds, and education, the perception of the six stimuli on the four evaluation criteria was more complicated.

Since the audience did not know the true creative mode of each stimulus, we believe that they used subjective intuition to evaluate them. In order to avoid the confusion of explanation caused by analyzing the stimuli one by one, the six stimuli were divided into three categories, according to the creative mode and the average value taken to analyze the variance. Corresponding analysis and discussion were carried out.

In addition, due to limited space, the large amount of content that could be analyzed for different audiences could not be presented. According to the purpose of the research, the authors believe that the differences between gender and background had the greatest impact on the cognitive experience of the audience and therefore the study’s focus was on these two demographic variables.

The cognitive differences of the stimuli through three modes between participants of different genders are shown in

Table 6. The cognitive differences between audiences of different genders.

Mode	Question (Core: 1–5)	Mean (SD)		F	Post Hoc Comparison
		Male (n = 308)	Female (n = 442)
Computer Simulation	Part I-Q03. How do you think the details of this painting are portrayed?	3.26 (0.80)	3.15 (0.73)	2.017 *	Male > Female
Amateur Artist Creation	Part I-Q01. What do you think is the rigor of the composition of this painting?	3.33 (0.81)	3.21 (0.79)	2.042 *
	Part I-Q02. How do you like the color matching of this painting?	3.21 (0.83)	3.03 (0.83)	2.914 **

n = 750; * p < 0.05, ** p < 0.01.

. It shows that there is a cognitive difference in question 3 (t = 2.017, p < 0.05) for the computer-simulated stimulus, and there is a cognitive difference between question 1 (t = 2.042, p < 0.05) and question 2 (t = 2.914, p < 0.01) for the stimulus created by amateurs. Among the above questions, the evaluations given by male audiences were significantly higher than those given by females.

The cognitive differences between the three creative mode stimuli among audiences of different backgrounds are shown in

Table 7. The cognitive differences between audiences of different backgrounds.

Mode	Question (Core: 1–5)	Mean (SD)		F	Post Hoc Comparison
		Art and Design (n = 400)	Non-Art and Design (n = 350)
Computer Simulation	Part I-Q03. How do you think the details of this painting are portrayed?	3.10 (0.76)	3.30 (0.76)	−3.595 ***	Art and Design < Non-Art and Design
Amateur Artist Creation	Part I-Q01. What do you think is the rigor of the composition of this painting?	3.20 (0.82)	3.33 (0.77)	−2.218 *
	Part I-Q03. How do you think the details of this painting are portrayed?	2.99 (0.89)	3.12 (0.82)	−2.145 *

n = 750; * p < 0.05, *** p < 0.001.

. For better statistical analysis and interpretation, we simplified the original eight different types of backgrounds into two categories: “art and design” and “non-art and design”. This study holds that for the “computer simulation” stimuli, there is a cognitive difference in question 3 (t = −3.595, p < 0.001); for the “amateur artist creation” stimuli, there is a cognitive difference between question 1 (t = −2.218, p < 0.05) and question 3 (t = −2.145, p < 0.05). Among the above questions, the evaluations given by audiences with non-art and design backgrounds were significantly higher than those given by audiences with art and design backgrounds.

The six stimuli were then divided into three categories as independent variables according to different creative modes, and the four evaluation criteria were analyzed as dependent variables by One-way ANOVA. The results show that among stimuli of different modes, none of the four questions had significantly different responses, which means that there was no perceived difference between paintings of different creative modes. We can speculate that when the participants were evaluating and identifying, they may not have cared what kind of creative mode the work used, but rather cared about the feelings the work brought to them.

Combining the results of the above analysis and comparing the accuracy of the two partial stimuli showed that viewing artwork is a more subjective experiential process and that everyone will have different opinions. This may be exactly what works of art are. The charm of art is that it allows us to get very diverse feelings in the process of viewing artwork. In addition, the authors found that audiences prefer, or agree with, paintings completed by humans and that paintings created by amateur artists were definitely recognizable.

3.3. Key Factors for Distinguishing the Stimuli

The researchers hypothesized that the audience was more inclined to use personal subjective intuition as the main criterion for distinguishing paintings that were imitated by amateur artists. Therefore, in each group of stimuli in the second part of the questionnaire, a question (Q03) was designed to understand the audience’s reason for distinguishing the paintings imitated by amateur artists. The results are shown in

Table 8. Audiences used only “intuition” as the criterion for judgment.

Stimuli	Only by “Intuition”	Stimuli Imitated by Amateur Artist	Accuracy
G01	87 (11.60%)	B	49 (56.23%)
G02	85 (11.33%)	A	56 (65.88%)
G03	82 (10.93%)	B	52 (63.41%)
G04	83 (11.07%)	B	59 (71.08%)
G05	88 (11.73%)	A	66 (75.00%)
G06	80 (10.67%)	A	51 (63.75%)

n = 750.

.

Although only about 10% of all participants relied solely on “intuition” as the criterion for judging the creative mode, among all the participants who made correct judgments, the proportion of those who made a correct judgment only by “intuition” increased between 56.32–75%. This shows that the hypothesis of this article can be established. If we exclude the option “intuition only” and sort the votes of the remaining five attributes from highest to lowest (see

Table 9. The number of votes for the other five attributes.

Stimuli	Attributes
	Line	Color	Light and Shade	Stroke	Detail
G01	47 (6.3%)	44 (5.9%)	15 (2.0%)	82 (10.9%)	19 (2.5%)
	Stroke > Line > Color > Detail > Light and Shade
G02	55 (7.3%)	25 (3.3%)	36 (4.8%)	50 (6.7%)	39 (5.2%)
	Line > Stroke > Detail > Light and Shade > Color
G03	28 (3.7%)	26 (3.5%)	31 (4.1%)	80 (10.7%)	28 (3.7%)
	Stroke > Light and Shade > Detail > Line > Color
G04	25 (3.3%)	30 (4.0%)	46 (6.1%)	48 (6.4%)	23 (3.1%)
	Stroke > Light and Shade > Color > Line > Detail
G05	39 (5.2%)	41 (5.5%)	25 (3.3%)	54 (7.2%)	21 (2.8%)
	Stroke > Color > Line > Light and Shade > Detail
G06	38 (5.1%)	25 (3.3%)	32 (4.3%)	44 (5.9%)	48 (6.4%)
	Detail > Stroke > Line > Light and Shade > Color

n = 750.

), excepting G02 and G06, the largest number of people used “stroke” as the criterion for discrimination. In G02 and G06, the number of people who chose “stroke” also ranked second.

It can be inferred that among the elements of painting, audiences are more willing to use the difference or effect of “stroke” to distinguish paintings created by amateur artists (regardless of “imitation” or “creation”). Therefore, we can make the following inference: whether it is imitation or creation, various painting tools (mainly brushes) are one of the important means for us to transform thoughts into paintings with the necessary depiction of lines, colors, light, and shade, and other such details. The possible reason for the above findings is that, excluding “intuition”, the audience is more willing to use “stroke” as the main criterion for distinguishing creative modes.

The number of votes regarding which criterion audiences use to distinguish paintings imitated by amateur artists is shown in

Table 10. The number of votes for the attributes selected by the audience.

Stimuli	Accuracy	Attributes
		Only “Intuition”	Line	Color	Light and Shade	Stroke	Detail
G01/B	497 (66.3%)	49 (9.86%)	25 (5.03%)	25 (5.03%)	6 (1.21%)	62 (12.47%)	12 (2.41%)
		Only by “intuition” > Stroke > Line = Color > Detail > Light and Shade
G02/A	511 (68.1%)	56 (10.96%)	36 (7.05%)	19 (3.72%)	19 (3.72%)	33 (6.46%)	25 (4.89%)
		Only by “intuition” > Line > Stroke > Detail > Color = Light and Shade
G03/B	499 (66.5%)	52 (10.42%)	14 (5.38%)	9 (3.59%)	16 (3.21%)	24 (9.56%)	11 (4.38%)
		Only by “intuition” > Stroke > Light and Shade > Line > Detail > Color
G04/B	499 (66.5%)	59 (10.30%)	17 (2.97%)	20 (3.49%)	31 (5.41%)	47 (8.20%)	18 (3.14%)
		Only by “intuition” > Stroke > Light and Shade > Color > Detail > Line
G05/A	559 (74.5%)	66 (11.81%)	30 (5.37%)	31 (5.55%)	17 (3.04%)	41 (7.33%)	15 (2.68%)
		Only by “intuition” > Stroke > Color > Line > Light and Shade > Detail
G06/A	559 (74.5%)	51 (10.41%)	27 (5.51%)	16 (3.27%)	15 (3.06%)	24 (4.90%)	36 (7.35%)
		Only by “intuition” > Detail > Line > Stroke > Color > Light and Shade

n = 750; The attribute with the highest number of votes is marked in red.

. Further analysis of different audiences when judging whether the painting is imitated by an amateur artist and whether there is a cognitive difference in confidence shows the following: First, among the six groups of stimuli, audiences from different genders, backgrounds and education had cognitive differences in their confidence in judging whether the work was imitated by an amateur artist, specifically, for each individual stimulus. In G01 and G03, audiences of different ages had cognitive differences in their confidence in judging whether stimuli were imitated by amateur artists. In addition to G02 and G05, the other four groups of stimuli had cognitive differences in the participants’ confidence in making their own judgments.

P05 and G06 refer to the processing of the same original work in different modes. P05 is imitated by an amateur artist and is code-named A in G06. According to the results, we found that when there was only P05, 28% of the audience correctly judged that it was imitated by an amateur artist. When the reference object (painting B in G06) appeared, the audience who made the correct judgment rose to 65.3%. This shows that whether there is a reference object will affect the accuracy of the audience in distinguishing the paintings completed by different modes. At the same time, 61.2% of the audience liked A (P05).

Why the difference between the two results is so large remains to be further explored. Figure 4 may help us to further discover the reasons.

It can be seen from Figure 3 that when P05 appeared alone, there was no other reference, which may have caused the audience to be unable to determine the mode used to complete it. From the specific analysis of the results of the audience’s answers to the four questions (Q1–Q4 in the first part) on P05, it can be found that:

• From the audience’s perception of P05 on the four evaluation attributes, the cognition among people of different ages, backgrounds and education was significantly different. This shows that the audience’s cognition of the denotation and connotation of P05 was very different.
• Interestingly, regarding the three groups classified according to the criterion in Table 4, their perceptions of almost all issues did not show significant differences.

It can be said that P05 has more characteristics of paintings completed by human artists. In addition, the audience preferred the paintings created by amateur artists. Therefore, we can speculate that when P05 appeared alone, the accuracy of the audience in distinguishing its creative mode was low compared with other stimuli, which may have just been an accidental event.

4. Conclusions and Suggestions

Authors should discuss the results and how they can be interpreted from the perspective of previous studies and the working hypotheses. The findings and their implications should be discussed in the broadest context possible. Future research directions may also be highlighted.

4.1. Summary

The intervention of AI in artistic creation not only brings more possibilities, but also subverts the mode of artistic creation and changes our thinking about viewing and evaluating artworks. The results of the research show that a certain percentage of the audience can correctly distinguish the paintings completed by different creative modes, and would like to choose “intuition” as the main criterion. These audiences who use “intuition” as the criterion for judgment have higher accuracy than other groups. This shows that intuition is obviously also an ability, not an excuse for random guessing. Those groups who use “intuition” as the criterion for judgment actually had their own opinions. They only choose “intuition” because they are relatively modest. Or it can be said that it would leave some leeway for themselves. There are still obvious differences in the perception of the relevant evaluation criteria for different audiences. This also further validates the subjectivity, complexity, diversity, and interest of the artworks themselves.

The process of AI performing DL is an alternative “imitation”. It is just that the “imitation” performed by AI can include almost all the details, which is closer to “copying”, and then AI uses different algorithms to start creation. For AI and amateur artists, imitation and innovation remain main ways to learn painting. AI has a massive database and is constantly innovating new algorithms. In theory, it can create countless kinds of works; but the learning process of AI is still be led by humans, at least at this stage. Therefore, artists and scientists can cooperate with each other and learn from each other. The authors believe that with the help of technology, the boundaries and possibilities of art can be expanded.

4.2. Follow-Up Research

The stimuli selected in this study did not include the original paintings, nor were all the processed paintings in Figure 1 used as stimuli. For the audience, this caused their understanding of the paintings to be relatively abstract. Therefore, in our follow-up research, we will put the original and processed paintings together to get a more comprehensive and in-depth understanding of the audience’s feelings. Secondly, this study found that there are big differences in the perception of many stimuli and corresponding evaluation criteria among different audience members. Due to space limitations and the focus of this article, this study did not examine or report this in depth, except from the two aspects of “gender” and “background”. This will also be a subject that the follow-up research will focus on; that is, what cognitive differences exist in the appreciation of artworks between audiences of different demographics.

In addition, intuition plays a very important role in daily life, and most people often use “intuition” as the initial standard to judge various things, including artworks. For example, when we visit a museum or art gallery which has a variety of different artworks. When someone asks whether you like an artwork and how you make a judgment, we believe that, in most cases, answers are more subjective and often rely on “intuition”. Just as English playwright William Shakespeare (1564–1616) said: “There are a thousand Hamlets in a thousand people’s eyes.” Everyone’s preference for artwork is difficult to measure by a certain established standard. For any artwork, some people like and dislike it. However, intuition may only be used for initial judgments, and other factors must be relied upon to arrive at relatively objective conclusions. Therefore, the underlying reasons behind this and the role of “intuition” in art appreciation will also become one of the focuses of subsequent research.