Are AI more creative than us?

In 2018, Christie’s became the first auction house to sell artwork made by an Artificial Intelligence (AI). The Portrait of Edmond Belamy was sold for $432,500. This was the first time an artwork “made”, “created”, by an AI was sold (Jason Bailey, 2018). Behind this transaction, many saw the end of art made by humans, or the advent of machines as artists. Nonetheless, the authorship of the “painting” was not attributed to the AI, but to the newly named digital art collective “Obvious”, which produced the piece. Further concerns appeared later, unfolding new layers of complexity, as it was found that the three French students of Obvious used a General Adversarial Networks (GANs) built by Robbie Barrat in 2017 (the official project was named “art-DCGAN”) with data run by American painter Tom White. Should Robbie Barrat and Tome White own copyright of the Portrait of Edmond Belamy? Should the AI own credits as well? 

            Such questions are connected to moral, ethic, social and philosophical concerns: What is the place of machines in our society? How is AI practically used for design today? If Machines will be developed enough to be independent from a human touch, shall they have a new, anthropomorphic status in society? Finally, these questions bring us back to the core interrogation on how we should define creativity. As a process? An idea? An object?

What is true Creativity?

Research Professor of Cognitive Science Margaret Boden categorises creativity into three forms: combinational, exploratory, and transformational¹The Creative Mind, 2009, p. 24-25. According to her, combinational creativity generates “unfamiliar combinations of familiar ideas, and it works by making associations between ideas that were previously only indirectly linked”. In the field of art, the Surrealist collage of Maw Ernst, The Chinese Nightingale²Le Rossignol Chinois, 1920 is a good example. When people mention how computers and especially Artificial Intelligence are creative, meaning how they appear to be creative because they generate unseen, novel and original images, they refer to the combinational type. Nonetheless, the two other types are important. As Boden declares it: “Exploratory creativity rests on some culturally accepted style of thinking, or “conceptual space” (e.g. social sciences, sciences, art etc…). Transformational creativity raises (…) ideas that are not only new but fundamentally different from any that went before. As such, they are often highly counterintuitive”. For example, when Galilee discovered that the earth was not at the center of the solar system, he revolutionized how humans see their places in the universe. 

 […]

How can AI be creative? 

What is an AI?

From automatic chatbots, to Siri, to DeepLearning models, AI has been used from the 1950’s to accomplish tasks usually done by humans. How is an AI working? To answer this question, two aspects must be considered: first, AI have different models of learning and training, and second, an AI is more or less efficient. The more intellectually independent, creative, the AI is, the stronger it is³Fefegha, 2020. For example, automatic chatbots do analyse, categorise and “answer” customer’s questions, but are incapable of designing a specific speech depending on the person’s emotions (stress, panic, impatience…). 

            The main inspiration for computer scientists to create AI as we call them today, was the human brain. AI are based on the creation and concept of Neural Networks (NN). NN read inputs (images, sounds, texts…), process it and generate output (drawing, music, poems…). Depending on the AI’s NN, the machine will be able to classify (Machine Learning) or produce data (Artificial Intelligence).

            There are three types of Machine Learning: supervised (the data in which you feed the machine is labeled), unsupervised (the machine try itself to learn from unlabelled data – this is what happens when Amazon and Netflix try to “guess” your preferred choice) and reinforcement learning (the machine learn through trial and error and punishment and reward). Depending on the task, different models of NN will be chosen. For example, Recursive Neural Network (RNN) will be more adapted to generate poetry because it analyses and generates sequences, while Generative Adversarial Networks (GANs) is a popular model of AI for visual artists because this model produces more “realistic” images than the other NN models.  

            As a consequence, AI cannot be considered as creative yet. Even if existing models come up with unprecedented combinations (e.g. DeepLearning psychedelic images) and design solutions (e.g. Developing intelligent Geographical Information Systems (GIS) and Computer Aided Design (CAD) systems for Architecture and Urban Planning), existing models are not complex or trained enough to be considered as efficient as humans. Moreover, Neural Networks on which Machine Learning and Artificial Intelligence are constructed have millions of “neurons” while the human brain, which inspired the system, has trillions of them. Even if AI related research advances fast, and even if this field of research holds many possibilities, AI is neither intelligent nor intuitive. But then, why is the rise of AI perceived as a threat to creativity? 

AI: Creative threat or new medium?

In his article ‘Can computers create art?’ Hertzmann explains how technology has, over time, been perceived as a threat by artists before becoming a medium. For instance, when painter Paul Delaroche saw a photography for the first time in 1839, he said “From today, painting is dead!”⁴Hertzmann, p. 3, 2018. Painting was not dead, but the exclusivity of painting as a representative tool was indeed over. 

            In reality, the question of technology in art, and eventually the question of AI in art, is linked to developments that precede them. As regards AI, well before it was accessible enough for artists and designers to use it, art history was already taking a critical turn with automation and creation. The turn started in the 1950s: “In the art world, there is a long tradition of procedural artwork. Jean Arp created artworks governed by laws of chance in the 1910s (or so he claimed), and, beginning in the 1950s, John Cage used random rules to compose music”⁵Hertzmann, 2018, p.8. A new turn in Abstract art appeared with Sol Lewitt’s Wall Drawing in the 1960s, then later in the 1970s when AARON generated paintings which were exhibited by Harold Cohen⁶Hertzmann, p. 8, 2018. The idea of the artist as the “creator”, but not “maker”, of her/his/their work was accepted. As photography and mass industry redefined the role of the artist in the creative work, so will AI, as noted by Jason Bailey in ‘Why Love Generative Art?’: “as AI technology becomes increasingly available, artistry and technical advancement will only become more important in separating the remarkable AI artists from those repurposing old tools built by other”⁷2018.

Will AI be more creative in the Future? 

The field for AI researchers is promising⁸Boden, 2009, p. 23. Does it mean that AI will ever be as creative as humans in the future? In ‘Of Human and Hoverflies’⁹The Creative Mind, 1992, p. 261 Boden states that the interrogation: “Will AI ever be creative?” is not computational but philosophical. She distinguishes four “arguments”: the “brain stuff argument” (1), the “empty-program argument” (2), the “conscious argument” (3) and the “non-human argument”¹⁰Boden, 1992, p. 271.

            The “brain argument” (1) is summarised by the idea that silicon and metal cannot beat neuroprotein: “no biochemistry, no creativity”¹¹1992, p. 271. However, there is so much to understand about how the brain works that it is still simplistic. 

Hence another approach: even if computers were able to produce complex symphonies and poems, all the symbols they would deal with would mean nothing to them. To illustrate her point, Boden mentions John Searle’s “Doodles experiment”¹²1992, p. 273. If Searle was in a closed room with doodles, an opening with doodles coming and a book in his native language telling him what to send depending on the one he received, he would send messages that make sense for the receptor but would not learn the actual meaning of what he does. In that, computers are semantically empty (2), and therefore are deprived of consciousness(3) and Human juridic status (4)¹³see p. 279-281. 

            As a conclusion, one can say indeed that AI cannot be truly creative, but also remarks how profoundly human this concern is. Instead of asking the validity of creativity in AI, should we not ask ourselves how much fantasy the idea of AI holds? 

Illustrated by Constance Leterre-Robert