When discussing artificial intelligence and the capabilities of different machines, a typical argument heard is that computers are not creative. This argument is often used to console those people in knowledge jobs who might fear that eventually the advancements in robotics will replace them as well. However, AI has already demonstrated artistic skills. As we are quick to appraise such artificial works of art as aesthetically pleasing, but not truly creative, we should also ponder what creativity really is, and who gets to define it.
If there was a clear consensus of what exactly creativity is, there might be no need to discuss whether artificial intelligence can or cannot be creative. John Smith from IBM Research suggests that creativity has something to do with ideas that are novel, unexpected, and useful. This is a fairly good definition, as usefulness emphasizes creativity as a means for problem-solving. If, however, we try to assess modern art based on its usefulness, we may easily end up dismissing its creative value altogether. How many times have you heard someone commenting on a piece of art with words such as “I could have done that too”?
An important part of creative endeavours is evaluating its quality. An artist probably has notebooks filled with sketches not good enough to be implemented in their final work. A musician has lots of unfinished melodies and lyrics, and unrefined demo recordings that will never be released on their albums. Sarvasv Kulpati argues that this ability to judge the quality of art is what humans possess, but AI does not.
Humans have a tendency to evaluate things from a human perspective, perhaps to such an extent that it can be called a bias. We have had trouble studying and understanding animal intelligence, for we have tried to assess their intelligence in relation to our own. When we judge a bird based on its illiteracy, we ignore its many other exceptional qualities that we humans do not have. A bit of the same attitude still prevails when evaluating artificial intelligence and its creativity.
GAN, a generative adversarial network, is a system which has already been harnessed to create art. In brief, GAN comprises two networks: one that generates content and the other that evaluates it. Essentially, a process like this which comprises a creation phase as well as a validation phase is something that artificial intelligence needs in order to better convince humans of its creative potential. GAN may not yet be able to tell novel and useful ideas from the poor ones, but how far are we from that moment?
Instead of comparing artificial and human creativity, perhaps we should embrace them both for their unique potential. As “quantity over quality” is often a good method to eventually end up with one or two excellent, creative ideas, we should remember that machines are much better than us at processing large quantities of any type of data. If artificial intelligence will be able to assess its own work in the future, it will probably be able to help humans with difficult creative decision-making. But in the end, creativity flourishes in large networks ‒ and many creative teams will probably warmly welcome their new AI teammates.