Human artists VS machines: How artificial intelligence is learning to be creative

It seems computers will soon be able not only to help human artists but even compete with them. Machines are already creating work for galleries, along with human guidance. Recently a team of scientists and art historians, including researchers at Rutgers University in New Jersey, USA, and Facebook’s AI lab in California, have trained an artificial intelligence to autonomously generate images that do not follow established styles. When the researchers asked humans to assess the images, in many cases people rated the AI generated art works slightly higher than those created by human artists. Will we see artificial intelligence becoming a true artist any time soon? What would the consequences be? RENDERU.COM spoke about it to Ahmed Elgammal, Professor of Computer Science at Rutgers University, and Director of the Art and Artificial Intelligence Laboratory.

1.jpegExamples of images generated by the system. Source: Ahmed Elgammal's blog

RENDERU.COM: You have trained a computer to create novel images with slight style variations from scratch. Before computers could only generate images which looked like already existing art. Does this mean that artificial intelligence has finally learnt to be creative?

Ahmed Elgammal: It has been a goal of many AI researchers since the beginning of AI - to make algorithms that are creative. And I’ve been working on that for a long time because, basically, creativity is a sign of intelligence. If you want AI to be really intelligent then you have to add creativity. Researchers have been working on creating an AI able to generate music, poetry, literature, jokes and different types of things. Art being one of these areas, but it is not easy. So I think we have reached now a position where AI is really making things that are creative. Humans wont tell whether this came from human artist or from machine.

RENDERU.COM: Do you consider this as a turning point in the history of AI?

A. E.: I think it is, yes.

RENDERU.COM: I have some doubts. Do you think the algorithm is really creating something new or it is just combining different elements of what already exists.

A. E.: Combining different things to create something new is also novel. That is how human artists do new things. Sometimes new ideas are deformations of old ones. However, looking at the arts generated by AI it is very hard to tell that this is a combination of things. Most of the things I have seen are totally novel. We have done an experiment with human subjects and, most importantly, even people with background in art and culture liked the images, they considered them novel and they found them aesthetically appealing. So I have no doubts that machines can create something novel.


Example of generated images: Top ranked images by human subjects. Source: Ahmed Elgammal's blog

RENDERU.COM: How about insights and inspiration? These are things computers can not experience.

A. E.: Yes, that is a tricky issue. It is very hard to tell what is an inspiration here other than the machine trying to solve a problem, which we framed for it. In this case the task was to look at the history of art and then try to generate something that does not fit any of the existing styles but not going too far. That is a problem that we put into the machine and the machine finds a solution for this problem. So the only inspiration here, I would say, is coming from this frame that we put. However, it is very hard to tell if any other inspiration at this point to come from semantics, or from what is going on around us, or from any other context. That is something that we definitely would be working on.

RENDERU.COM: What are the main problems researchers need to solve to create an artificial mind which is able to draw as a real artist?

A. E.: One important thing is to understand what art is about. The reason why we have succeeded is that we have been studying things like art history for a long time. We have a lab called ‘Art and AI lab’. In the lab we have been working on problems of style justification and creativity so we have the understanding of these concepts. It’s really important for people in this field to have this connection to artists, historians, psychologists to be able to develop such algorithms.

RENDERU.COM: How far into the future until this happens?

A. E.: Very soon. There are many attempts by different scientists, big and small companies and startups - all of them are working on things related to visual art, music and literature trying to push the boundaries. It is a very hot topic now - how the machines can be creative.

RENDERU.COM: Will we see significant results in 5-10 years?

A. E.: It depends on how people will receive it. In the end in our culture any products created by machines (music or painting or anything else) are made for humans. So if the machines are successful in generating things that inspire people and if people like it, that will push things forward. However, people may react to this negatively. One reason may be that it is not aesthetically appealing to them. Another reason might be that people push back against the idea of the machine being creative and don’t want to accept that. People’s reaction and acceptance is crucial in this process.

RENDERU.COM: There are fears that certain specialists may lose their jobs because of AI development. Do you think it may ever happen that machines replace the artists?

A. E.: It’s absolutely up to the artists to decide. Innovative artists would always use technology as a new tool in their toolbox and generate something on top of that. However, if artists stick to traditional techniques, there are definitely doubts on how they would survive in the art world. New technologies have always been coming to the art world and art adopted it. The most relevant example is photography. The difference is that the AI technology now is creative. It is not a passive device but part of the creative process, so there is a room for collaboration here between human artists with human creativity and smart machines to create something new. I imagine in the future artists will be able to create a framework of what they want to do and then the machine will find a solution or will come with creative renderings of these ideas. If this happens, I think, that would be really great, because that would be a nice tool to use by artists to really explore the creative space.

RENDERU.COM: How should we classify the art created by a computer? How about art contests with AI participants?

A. E.: Again, it depends on how humans will perceive it. If there is a competition and if they allow the machine to compete, definitely, that would be great. The problem is there might be a push back and restrictions: people might think that the contests are only for human artists, and this is a threat.

Also, someone can claim to be a human artist or to use AI as a tool. Then the artwork can be thought as a human art, because there is an artist behind the scene. This is already taking place. In the recent couple of years there have been artists who have been using a sort of AI to help them in generating something. They make great art and sell it. Although, the machine is really doing the background work here. The difference in our case is that the machine is totally autonomous. Would that be allowed to compete? We will see.


Ahmed Elgammal at the the exhibition, Frankfurt Book Fair

It would be interesting to see how people react to this artworks. I have seen some reactions before. We showed these things to collectors and art lovers and they liked it very much. But once they knew that these were coming from the machine, they came reluctant to accept it. So that is the test here how people would react, would we have collectors willing to buy this art, generated by the machine, without having a human behind at all? It is very tricky.

We showed some of the AI created works at the Frankfurt Book Fair, the world's largest trade fair for book. The reception from the audience was great, totally positive. The exhibition showed 12 pieces as the first materialization of the AICAN algorithm. What was interesting is that everyone has his/her own favorite from the collection, which shows that the collection has variety that suits the taste of everybody.


RENDERU.COM: Do you think people should develop special rules to regulate AI usage in art?

A. E.: In this domain, I do not think that is needed because in the end, as I said before, in cultural domain the recipients of the culture are humans anyway. There is no need for regulation because ultimately humans are the final judges. There are millions of artists in the art world and very few of them really push the envelope and make the art history. So this filter is there: the filter by humans, by collectors, by critics and by the other artists. So there is no need in regulation in this domain.

RENDERU.COM: How about other areas? Elon Musk recently tweeted “If you're not concerned about AI safety, you should be. Vastly more risk than North Korea”.

A.E.: Yes, I accept that. Other domains that are related to military application and anything related to human lives definitely need regulation.

RENDERU.COM: You mean art is safe?

A.E.: I totally agree, art is safe. In the sense that humans are the observers, humans are the recipients and humans are the ones who decide what to accept what not to accept. And again, artists are smart - they will use this new technology, and a combination of human creativity and machine creativity will lead us to the things that we have never imagined.

RENDERU.COM: I hope so.