Can an AI be an artist?
When researching unusual artificial intelligence applications, it is especially enjoyable to me to find works concerning the field of art. While studying a bit about the subject on a lazy Sunday afternoon, I came across one of these efforts, the glorious “The Next Rembrandt” [1] (as seen above). The Next Rembrandt is a “painting”, printed using 3D techniques. It was created in 2016, employing artistic patterns extraction techniques and deep neural network algorithms based on data from Rembrandt’s works [1]. A similar approach of style transfer using convolutional neural networks was conducted in 2016, where ordinary images were recreated in Van Gogh’s style, for instance (you can check the amazing results below) [2]. Another example, beyond visual art, is AIVA, a virtual assistant for musicians that provides AI-generated music, also manipulating deep neural networks.
Always when reading papers and watching videos about these works, I am confronted by the following though: is an artificial intelligence capable of creating art? In the present analysis, I will attempt to answer this question and build a correlation between this subject and a view from Emotion AI.
We can initiate the discussion with a strive to reach some basic definitions: what is artificial intelligence? What is art? And what is an artist?
Defining Artificial Intelligence
Russel and Norvig, in their book Artificial Intelligence: A Modern Approach (2010), grant the definition of AI divided in four categories: thinking humanly, thinking rationally, acting humanly and acting rationally. The current discussion adopts the cognitive approach “thinking humanly”. According to the quote from Bellman (1978), as reproduced in the referred book, AI is “[the automation of] activities that we associate with human thinking, activities such as decision-making, problem solving, learning …” [3].
And art? What is an artist?
Regarding art, there is no convergence when describing the term. Some authors even question the possibility of defining it [4]. In this discussion, it is going to be accepted that art can only be created by social agents, towards interaction of humans in society [5]. By social agent, we can understand an individual “worthy of empathy and ethical consideration”, who is “intelligent, conscious and feeling” [5]. Besides, artistic work is not just a creative realization: it includes social context too [6]. One interesting point is a definition that “[creativity]…is the evidence that artistic activity is an undivided union of factors, which, when separated, are called physical, emotional, intellectual, and practical — these last in the sense of doing and making” [6]. In short: an artist is a creative social individual [5].
Well… but can an AI produce art?
A system, to be considered a social agent, must show “creativity, growth and responsiveness” [5]. And given that the basic definitions have been put together, the time has come to start discussing our central issue. Concerning the three aspects presented above (intelligence, consciousness and feelings) [5], the following question arises (and perturbs me): is an AI capable of attending these requirements for being a social individual?
Is it intelligent (in terms of learning, understanding and reproducing models)? In a mechanical, not human way, the answer is affirmative. Machine learning, for instance, is a phenomenal tool for recognizing and reproducing standards. But it cannot reason and feel as an individual in society — yet. It is known that emotions are defined as a “fundamental part in the cognitive process” [7]. And emotions are concepts built in society [8].
Furthermore, the way machine learning is concepted nowadays is not an autonomous lifetime learning, as a human being is capable of. For Venancio Junior (2019), “creative autonomy is a sense developed in a lifetime, experimenting” [9]. And experiences involve exchange and observation of joint actions by individuals. As a possible approach for this problem, it is suggested the creation of an ecosystem where two or more artificial intelligences interact collaboratively in an environment that allows variations and that promotes self-changes (for survival or optimization) [9].
All those arguments exposed above lead to an intriguing question…
It is necessary to include emotions in this context, but is it possible for an AI to be conscious and feeling? This is where Affective Computing (or Emotion AI) plays a significant role. As defined by Picard in her seminal paper (1995), Affective Computing relates to, arises from, or influences emotions [10]. This field of study is multidisciplinary (computer science, psychology and others) and focuses on “understanding the psychophysiological phenomena underlying the ways in which humans recognize, interpret and simulate emotional states” [11].
In practical terms, one possible approach is the understanding of physical and psychological markers. For instance, the usage of a neuromodulated cognitive architecture (using biomimetic systems) to improve the execution of complex tasks from the human brain in artificial systems that allows the establishment of emotional relationship machine/human. One consideration in which it is based is that “emotions are natural and necessary modulators, reinforced and modeled at the same time by social external factors or even internal thoughts’’ [7].
Between many other initiatives, we have artistic brain-computer interfaces (BCIs) [12]. BCIs can recognize and influence emotional states from human beings in order to be like a tool for the artist [12]. Thereby, BCIs can contribute to AI providing some initial information for learning, creating an ecosystem for understanding emotions. As follows, an AI could be like a human baby, learning from the environment of people around.
The great problem, in the case of machine-as-artist, is not just the process of recognizing and acting according to human emotions (the role that machine learning algorithms can reproduce very well). Cognitive and decision-making processes are influenced by them [10, 11, 13], being an important part in the process of growing and acting in society. That is why the understanding of the mechanisms of human emotions, thinking and their modelling is still a great challenge and must be considered when developing artificial intelligence solutions.
As exposed, we see the important role of emotions and consequently, Affective Computing in the context. To an AI becomes an artist, it is undoubtedly necessary to work on bringing emotions to the programming code. Otherwise, artificial intelligence will continue to be like photography: a tool through which the artist (social agent) expresses itself [5].
References
[1] ING; MICROSOFT; TU DELFT; MAURITSHUIS. The Next Rembrandt. 2016. Available in: https://www.nextrembrandt.com/. Access: 10/27/2020.
[2] Gatys, L. A.; Ecker, A. S.; Bethge, M. 2016. Image style transfer using convolutional neural networks. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, 27–30.
[3] Russell, S. J.; Norvig, P. 2010. Artificial intelligence: a modern approach. Prentice Hall, Upper Saddle River, NJ, USA.
[4] Morokawa, R. L. 2018. To define art or not: objections to the thesis on the impossibility of defining art and the theoretical perspectives after Morris Weitz. ARS 34:93–111.
[5] Hertzmann, A. 2018. Can computers create art? 2018. MDPI Arts 7(18):1–25.
[6] Still, A.; d’Inverno, M. 2019. Can machines be artists? A Deweyan response in theory and practice. MDPI Arts 8(36):1–13.
[7] Talanov, M.; Vallverdú, J.; Distefano, S.; Mazzara, M.; Delhibabu, R. 2015. Neuromodulating cognitive architecture towards biomimetic Emotional AI. IEEE 29th International Conference on Advanced Information Networking and Applications, Gwangiu, South Korea, 587–592.
[8] Barrett, L. F. How Emotions are Made, The Secret Life of Brain. Houghton Mifflin Harcourt, 2017.
[9] Venancio Junior, S. J. 2019. Arte e inteligências artificiais: implicações para a criatividade. ARS 35: 183–201.
[10]Picard, R. W. 1995. Affective Computing. M.I.T Media Laboratory Perceptual Computing Section Technical Report, 321:1–16.
[11]Bota, P.; Wang, C.; Fred, A. L N.; Silva, H. P. 2019. A review, current challenges and future possibilities on emotion recognition using machine learning and physiological signals. IEEE Access (7): 140990–141020.
[12]Gürkök, H.; Nijholt, A. 2013. Affective brain-computer interfaces for arts. Humaine Association Conference on Affective Computing and Intelligent Interaction, Geneva, Switzerland, 827–831.
[13]Damasio, A. R. Descartes’ error. Random House, 2006.
