NOS PROJETS
NOTRE APPROCHE
À PROPOS
CRÉER AVEC NOUS

Ai, art and what it means to be human

June 28, 2019

Abstract

Artificial intelligence is no longer reserved to scientific and mathematical applications, but is now used to create music, visual arts and even poetry. Its impact on society relating to its usage in creative endeavours will be explored. The discussion will be centered around the very definition of being human and how artistic AI can reshape it and break it down. Also, The question of human biases being reflected through AI creations will be explored to see how that technology can impact our perception of the world.

Technology

In one of her lectures, Ursula Franklin defined technology as being ”a multifaceted entity, [that] includes activities as well as part of a body of knowledge. It includes also structures as well as the art of structuring”. She also mentions that technologies are like a house, and that most of human activities happen inside that house (Franklin,1992). In that sense, technologies are processes or objects useful to humans, created by humans, that may portrait human traits without being humane.

AI

AI, or artificial intelligence, is a broad concept in computer science where a program seems to learn and reason in order to perform some task. It often seems like magic, but in reality AI is a result of clever mathematical tricks that allows a program to find patterns in data; patterns that would be very hard for humans to find. One of the most common use for intelligent systems are for classification tasks, where a model is shown millions of data points and told how to interpret them. The connections in the model (like neurons and electrical signals in our brains) get tweaked with each data sample the model sees. This phase is called training. Once trained, a model can then see new data and interpret it according to what it learned from the training data. Classically this type of model is extremely accurate for tasks like image recognition or labeling. For example, show a deep learning model 1 million different pictures of dogs and cats and make it guess, adjusting its brain based off of if his guess was right or wrong. At the beginning, the model is going to perform purely randomly, but as you show it more and more data it will find patterns and will become better and better. Once all the training data seen, show the model a picture it has never seen of a dog or a cat and it will accurately label it. These type of models can have accuracy close to 100% (97-99%).

There are a lot of AI paradigms, but another one relevant to this discussion is a neuro-evolution network. As the name suggests, this technique mimics the strategies of evolution. When building this type of model, a programmer needs to have a task (what he / she wants the AI to do) and a way to evaluate the performance at that task. Then, generations of hundreds of AI agents are generated and the most skilled of every generation are breed. Of course, the term breeding is not accurate in that context, but its idea applies. Once again, the first generations are completely rubbish at completing the task, but through luck and through enough generations, favorable properties emerge and the model becomes skilled at the task, much like evolution and natural selection created species highly adapted to their environment over the course of hundreds of thousands of years.

Artificial intelligence dates back to the 1970’s, but recent developments in machine learning coupled with an increase in computing power greatly improves its utility in multiple fields. Classical STEM (science, technology, engineering and math) fields were the first to see applications of this technology. Medical professionals can use machine learning models to analyze brain scans to detect tumors, the automotive industry relies more and more on AI for semi-autonomous driving and software companies use it heavily to analyze data. Weather it be to form a prediction on stock markets, to detect fraudulent activity in credit card data or to ban fake accounts from social media platforms, AI is used in some way, shape or form to perform a task that would otherwise require incredible amount of man-hours.

Art

Defining art can seem trivial at first, but the Definition of Art in the Stanford Encyclopedia of Psychology shows that it is subject to much debate in the academic community. Difficulties sprout from the notion that “art[s] are, by their nature, too diverse to admit of the unification that a satisfactory definition strives for, or that a definition of art, were there to be such a thing, would exert a stifling influence on artistic creativity” (Adajian, 2018). This notion will be useful in our discussion, but a palatable definition can be formed by glossing over the majority of the complexity of the question : art is ”the conscious use of skill and creative imagination especially in the production of aesthetic objects” (art, 2019). The word object will also refer to music and writing in our context, which both are non-tangible objects.

Art predates history, with the first examples of drawings found in the Blombos Cave in South Africa which would date back to 73 000 years ago (Blakemore, 2018). Undisputed artistic sites (much like prehistoric museums) like the caves of Chauvet, Lascaux, and Altamir, date back to less than 40 000 years ago (Adajian, 2018). These caves were not inhabited and are covered with drawings which portrays the capacity of early humans to depict abstract concepts, to picture themselves in situations that are not occurring at the moment and to communicate symbolically. These aspects of art are still present today, with artist aiming to convey feelings to the audience through the abstraction of physical concepts or situations.

The human definition

The human definition is a term that will be used throughout this essay. By this term, the author wants to convey to the reader the idea of what defines humanity. In other words, what makes us humans ? What differentiates humans today from our pre-human descendants ? At what point in human history did we become human, and what elements formed that boundary ? The answers to these questions being far from definitive, a solid definition will not be given, but useful pieces of that definition will be outlined. These pieces of the human definition will be useful to evaluate if artistic AI requires a redefinition of what it means to be human.

The conflict

AI can now create art indistinguishable from human-made one, what does that mean for the human definition ? Does that make us less human, or does it mean that technology is more human than we thought ? Are we not special ? If art can be recreated through mathematical models, can emotions be as well?

The realization that mathematical models can create art can be deeply unsettling, and the implications of this technology on society is more philosophical than tangible, but is very much present.

The role of art in the human definition

The first question that needs to be answered is to what extent does art define humaneness? One may argue that art does not define humanity in the least and that AI and art has absolutely no bearings on his or her life, which would put a stop to this discussion. On the other hand, art played a role in human history that we deem is not negligible and that in turns justifies this discussion.

Language and writing are two steps in human history that were turning points for our species. With language, came the ability to communicate and organize more efficiently, while writing marks the start of history, where knowledge could be passed down through generations and is deemed to be one of the greatest invention of mankind. These two technologies share similarities with art, through their abstraction and portraying of the real world, which leads one to think that art may have been an indicator that mankind had reached a maturity and a ripeness to develop those technologies. As Gibeault & Uhl said, “Both art and advanced language are based on the capacity for symbolic representation, abstraction, integration, and probably affect recognition and regulation. The capacity for symbolic representation, allowing for symbolic communication, is a unique landmark in human evolution” (Gibeault & Uhl, 1998). The capacity to symbolize, recall and anticipate shown through art may be one element that unbound early humans from their instincts, making them in turn closer to modern humans (Blum, 2011).

With the implications of art-generation in mind, we can say that art is an indicator of the boundaries of the human definition, that humans before art were similar to us, but lacked a maturity that is essential to what we are today. If art is an indicator of when humans started being humans, the creation of art through artificial intelligence must have an impact on either the human definition, or the definition of that technology.

The impact on the human definition

If we accept that the creation of art is an indicator of a border between pre-humans and humans, the fact that our technology can independently create art must be an indicator that the human species has reached another stage of development. Through the ages, the technologies we created helped with our tasks or improved our quality of living. For examples, an iron axe makes the task of cutting a tree easier than doing it with a stone, the wheel facilitates the task of transportation, like the boat and the airplane, whereas vaccines and the internet greatly increase our quality of living. That being said, art-generating AI is a technology that does not help with a task, and arguably, does not improve our quality of life. This technology is in a new category in the sense that, on its own, it makes us reflect and inspires us. What does that mean for the human definition ? We can argue that art-generating AI is an indicator that the human species has evolved into creating technologies in the hope to expand the horizons of our creativity and to possibly find new avenues of thinking. We leave behind us the era where technologies were mechanical muscles and computing units, and enter in an era where humans can create mechanical brains that help them expand the horizons of thinking and creativity.

A re-definition of technology

Technologies were said to not be humane, but rather reflected human traits without exhibiting ones of their own. For example, a hammer reflects humanness through the shape of its handle, but a hammer in itself is not humane, it is only a weirdly shaped piece of metal. With more complex technologies, like artistic AI, that opinion is falling apart since the technology in itself can portray human characteristics. For example, AICAN (artificial intelligence creative adversarial network) a creation of the American Scientist, is an AI agent that can autonomously create art. AICAN is so good, that 75% of humans cannot distinguish its work from non-AI creations and one of its pieces was sold for 16 000$ at auction (Elgammal, 2019). This not only means that a new form of technology can be humane to the point where humans don’t distinguish between its creations and man-made ones, but also that it can convey emotions and creativity to such an extent that humans buy its creations.

The fact that artistic AI goes against the definition of technology, since it is in itself humane in some ways, may require a redefinition of technology, or at least the creation of a new category. If humans are distinguished from pre-humans in part because of their capacity to create art, maybe technologies need to be differentiated from technologies that display humane traits. Much like species evolving, maybe some technologies will evolve into post-technologies where they help us on the creative and psychological level.

Concrete impacts on the artistic world

OBVIOUS, a Paris based collective created an AI painter that produced the first auctioned work of art algorithmically created. One of its creation, part of a series of portraits of a fictive family, is a “portrait in [a] gilt frame depict[ing] a portly gentleman, possibly French and — to judge by his dark frock coat and plain white collar — a man of the church. The work appears unfinished: the facial features are somewhat indistinct and there are blank areas of canvas. Oddly, the whole composition is displaced slightly to the north-west. A label on the wall states that the sitter is a man named Edmond Belamy”. This piece sold for 432 500$ at a Christie’s auction (Christie’s, 2018). There is a lot of talk about AI replacing jobs, mostly in the transport industry, but this is a real life example that even creative work can be performed by machines, further increasing the uncertainty of the job market in the future.

More generally, An artistic AI may function in a vacuum, “in an isolated creative space that lacks social context”, whereas humans create to tell stories and make sense of the world (Elgammal, 2019). Having AIs be part of the creative process could lead to a decrease in human creativeness as it can become another task in which robots outclass humans.

Human biases through AI creations

AI can be deceiving in the sense that its creations can be seen as objective when they are not. Since AI models are trained through human-sourced and processed data, patterns in this data will be reflected in the model. For example, an AI trained solely on Picasso’s paintings will most likely produce pieces in that style. By nature, data is hard to make senses of, thus biases may hide in a data set and be reflected through the AI model. If we open the door for AIs to create art, texts, music, etc. there is a risk that it amplifies bad aspects of our society like sexism, racism or any other discriminatory behaviour, without us even noticing. Indeed, depending on who trains and builds the model, and on if that person was careful or not in the data collection process, AI-generated artistic products could be following a hidden agenda and be consumed by the general public thinking it is completely subjective. That layer of abstraction that an AI creator brings can be insidious because it hides bad intent or incompetence behind a curtain of objectivity.

[1] Franklin, Ursula. (1992) The Real World of Technology. (CBC Massey lecturesseries.) Concord, ON: House of Anansi Press Limited. ISBN 0-88784-531-2

[2] Adajian,Thomas, “TheDefinitionofArt” , The Stanford Encyclopedia of Philosophy (Fall 2018 Edition), Edward N. Zalta (ed.).

[3] art. 2019. In Merriam Webster. Retrieved June8, 2019.

[4] Blakemore, Erin. (2018), “73,000-Year Old Doodle May Be World’s Oldest Drawing”, The National Geographic.

[5] Blum, H. (2011). The psychological birth of art: A psychoanalytic approach to pre-historic cave art. International Forum of Psychoanalysis, 20(4), 196–204.

[6] Gibeault, A., & Uhl, R. (1998). De l’outil a l’oeuvre d’art: l’invention de la sym-bolization. In F. Sacco and G Sauvet (Eds.), Le propre de l’homme.

[7] Elgammal, Ahmed. (2019) “AI Is Blurring the Definition of Artist”, American Scientist.

[8] Christie’s. (2018) “Is artificial intelligence set to become art’s next medium?”, Christie’s.