Art is considered a ‘creative’ field, and creativity itself is a subjective concept. Even the process of being creative is difficult to explain – some people might find sudden inspiration, while others may be naturally gifted. Moreover, how can you evaluate a creative artifact? It is against this background that the question arises of whether computers can make art.
Artificial Intelligence and Computational Creativity
Computers are usually considered scientific objects and are mostly known for their mathematical logic and precision. This intrinsically goes against the concept of being creative – principles of logic do not govern art. However, the creative landscape has changed significantly with the development of new technologies, such as artificial intelligence (AI). Computers are now playing a key role in creative activities such as writing stories, poems, jokes, making paintings, and even generating music. However, instead of just using computers to assist human beings in creative tasks, they should be seen as creative entities. This has led to the development of a new field of Artificial Intelligence known as Computational Creativity. In this article, the question of whether a computer can be fully autonomous in creating art is discussed, along with a final discussion on what the future holds for computational creativity. There are several creative systems, a few of which will be reviewed here.
Artificially intelligent painters
One of the most well-known examples of software creating art is The Painting Fool. The aim of its creator Dr. Simon Colton was to develop a program that could be considered an artist in its own right. But how does it exactly work? There are two main ways the software does this: using user-defined input like a photograph and pairing the system with emotion-detection software, creating pictures based on the subject’s mood.
Researchers from Facebook’s AI research Lab, the College of Charleston, and Rutgers University also developed a system using AI to create paintings. Creative Adversarial Networks (CANs) are used, which build on Generative Adversarial Networks (GANs), systems that have generators (create art), and discriminators which try to identify if the generated image is art or not art. The model was trained using about 81,000 paintings spanning about five centuries – from the 15th to the 20th century, and covering a diverse range of painting styles. But are these systems fully creative? How is it possible to evaluate their creativity?
The Dancing Salesman Problem created by the Painting Fool (Source: https://www.newscientist.com/gallery/painting-fool/)
Evaluating creative systems
According to Colton, for software to be considered creative, it should be skillful, appreciative, and imaginative. Just as a tripod has three legs, these are the three fundamental behaviors required in a creative system. In addition, the legs of a tripod are split into three parts. These represent three parties that could be considered to contribute creatively when a user sees a computer-generated artwork – the programmer, the computer, and the consumer. As every party can contribute skill, appreciation, and imagination to this experience, each of the three sections in the tripod’s legs can be used to symbolize the contribution of each party to their respective behaviors. Software, therefore, will not be considered creative without all three behaviours.
Another method to evaluate creativity is to ask human beings for their opinions. For the system using CANs, human participants were asked to evaluate the generated art, and it was found that participants could not distinguish between actual art and AI-generated art. However, basing the evaluation of a system on opinions would be quite subjective.
Autonomy in creative systems
Based on Colton’s Creative Tripod method, being imaginative and appreciative qualities will play a major part in determining whether a system is autonomous. However, the meaning of the term autonomous is still debatable – how can an AI system be completely independent? Therefore, the definition of autonomy should be narrowed down to better answer this question. While there are several different creative systems, two will be discussed here – the What-If machine and Dall-E.
Sometimes, a writer may face a creative block or run out of ideas. The What-If machine, a part of the WHIM project, comes up with creative ideas with the help of its complicated algorithms using information from the internet. A few years ago, it even created a script that was successfully turned into a play in London. The system uses machine learning to determine if people will like these ideas or not. Therefore, the system is also evaluating its work – which is quite important for an artist. To some extent, this makes the system imaginative since it uses feedback to improve the quality of its work. However, the computer may not fully understand the value of its work, and this is where the challenge arises.
While this may be challenging, it is worthwhile to consider if the system needs to be appreciative at this stage. The use of AI is rapidly growing, and its contribution to the development of creative systems is also growing. This contribution has been significant to human beings, particularly in the sense that it encourages co-creativity. The process by which we can enable a system to make creative artifacts itself is fascinating and will assist in the idea-generation process in the long run.
For example, Dall-E 2, an AI system that generates new images based on text, can combine concepts, styles, and attributes. While writing this article, I landed on its page and was again impressed by its abilities. In the image below, the flamingo changes form depending on the location I select in the image on the left.
The above is just one of many things Dall-E can do. It is undoubtedly skillful and imaginative, and I fully appreciate its value, although the system itself might not.
On the website, one of the objectives for Dall-E is that it will enable us to see how ‘advanced AI systems see and understand our world.’ This is quite interesting. One had always heard of being able to understand another individual’s perspective, and here the creators wish to understand how AI sees our world. If we look at the wording again, it seems that AI is almost being treated as a human being in its own right. Does this mean that Dall-E is an autonomous system? I strongly believe it is. The fact that the system is adaptive and creates images using caption-to-image generation shows how it deeply understands concepts.
Are these systems creative?
The answer to the above question is a grey area. While AI is enabling the development of creative artifacts, should we consider it a creative being, especially since human beings themselves have created it? Intelligence is artificial at the end of the day. Moreover, I would again refer to Colton’s creative tripod method. Are these creative systems skillful, appreciative, and imaginative?
They are skillful and imaginative. Examples of The Painting Fool, The What-If Machine, and Dall-E are all brilliant examples of creative systems. Some might even produce better results through machine learning techniques (this feature is present in The What-If Machine). However, are they conscious beings on their own? Machines do not have consciousness. Nevertheless, this lack of consciousness is not a reason to overlook the potential for creativity these systems offer nor the potential they provide in terms of intelligence.
What does the future hold?
The future seems quite bright for computational creativity. Computers can now assist human beings in being more creative, and while they may not be completely original, they are still acting as co-creators. It is vital to acknowledge this. A possible reason for this may be that not all human beings may be able to write exciting stories or make funny jokes. However, with the help of a creative system, this might become possible.