The idea of creative machines is challenging the notion of human uniqueness

Creativity is seen as an emphatically human attribute. And yet, we have already come to accept that machine intelligence outdoes that of humans in computing. When will we accept that machines might also be creative?

Computers are already able to imitate the styles of the great masters, learning the essence of their style and the ability to emulate it. Professor Hannu Toivonen and his research group in FCAI have been developing computer programs that produce, among other things, music, rap lyrics, poetry, images and audio plays. Toivonen has also developed a theory of creative artificial intelligence. Artificial intelligence does what it is programmed to do, but is it capable of creativity?

It is a widely accepted fact that machines have the upper hand in computing. Computers are effective in reviewing a range of options or large quantities of data, as well as in logical and statistical inference. However, Toivonen believes these qualities are also related to creativity.

– It’s difficult to be creative without being smart. There must be some kind of rational action underlying creativity. For example, when a person associates things in their mind, they have at their disposal the enormous computing capacity of the natural neural network, making it possible for associations to emerge from their subconscious. Computers can do the same in a more systematic manner. They can go through a huge number of possible connections, assess them and pick the most promising ones. Depending on the situation, this can appear creative if the result is unexpected and pleasing, Toivonen says.

Creative artificial intelligence, or computational creativity

The research field represented by Professor Toivonen – computational creativity – is a fairly new one under computer science. Despite its novelty, an active and international group of scholars has already emerged in the field. Research in computational creativity touches upon, among other topics, data mining, computer graphics, psychology and neuroscience.

Computational creativity, or more colloquially creative artificial intelligence, focuses on investigating how to use computers to produce and support behaviour that is considered creative in humans. Research can be targeted at the behaviour of both natural systems, such as humans, and artificial systems, such as machines.

Toivonen’s research group has counted among its interests methods of linguistic creativity in particular, including automated poetry writing, research on word associations, humour and automated text summarisation. In addition, the group has conducted research on algorithmic composition, the creation of pictorial metaphors as well as human–computer co-creation.

Human–machine collaboration

Creativity famously has a number of definitions. According to one of them, creativity is the ability to produce something new and valuable. In art, value can denote beauty, or it can be related to, for example, a message someone wishes to convey. In creative problem solving by computers, value is linked to how competent they are at solving specific problems.

– With regard to artificial intelligence creating art, it would be natural for humans and computers to collaborate, Toivonen says, citing as an example the Poetry Machine, an AI solution developed by his research group.

Underlying the Poetry Machine is a creative artificial intelligence solution that has learned from existing poetry. The Poetry Machine utilises tens of thousands of lines of Finnish poetry in service of its computational creativity.

Toivonen’s group has investigated in school classes how the Poetry Machine can help pupils in creative work. The computer program helps them get started with writing poetry, providing assistance in finding the right words and rhymes.

– Pupils wrote poems together with both the AI solution and their friends. Some children were inspired to write rap lyrics that they then performed in front of the class; others got into poetry writing. It’s interesting how the computational creativity of machines can support human creativity, says Toivonen.

The researchers aim to determine how to enhance the completion of different creative tasks by having a computer solve parts of them or help humans complete them. For instance, the Helsinki Central Library Oodi was designed with computer-assisted means. Architects bear the creative responsibility, but software assists them.

–However, the assistant role is constantly expanding. AI programs are better able to take into account both various constrictions and user preferences, producing increasingly tailored and ready-made suggestions, Toivonen says.

Artificial intelligence can be creative in itself

Creativity, however, is not only connected to art but all creative problem solving. Smart systems and robots must have the capacity to cope in unforeseen circumstances and with systems that did not even exist at the time they were designed. Here, Toivonen wishes to highlight the classical definition of creativity: the ability to produce something new and fit for purpose.

– Creativity does not mean that anything randomly created by a computer should be thought of as creative. Key to creative activity is that the system is capable of carrying out independent and reasoned experiments, assessing their results and adjusting its actions accordingly. If a solution created by a computer is not fit for purpose, it must be able to not use it. For example, systems and robots must take their surroundings and circumstances into consideration. While the actions of AI solutions may appear surprising, there usually needs to be a justification for them, at least in retrospect, Toivonen explains.

In terms of artistic activity, Toivonen posits that artificial intelligence will carry on being one of the tools of art, since in artistic tasks AI solutions lack not only the ability for overall assessment but also the innate need to create art. Commercial content production may be another matter.

– In the future, music can probably be produced according to the personal tastes of individual listeners. Even now, video games contain a great deal of different procedurally generated content, backgrounds and music, Toivonen says.

Instead of being limited to creative arts, the scope of creativity extends to the functioning of a broad range of smart systems.

– Creativity will make the smart systems of the future even smarter and more flexible if such creativity enables them to better adapt to different situations and problems, says Toivonen.


This article has previously been published on the website of FCAI - Finnish Center of Artificial Intelligence.