Artificial Intelligence and Creativity Theory
by Simon Das
An article based on an on-going collaboration between Simon Das and Matthew Kershaw, MD of Content at Iris Worldwide . Having worked together in the magazine publishing and music industries over the decades, the article came about as part of a conversation based around ‘resolving’ an AI debate around Matthew’s interest in artificial intelligence (AI) and Simon Das’s PhD research on creativity theories and media
Back in October 2018, a painting entitled ‘Portrait of Edmond Belamy’ went under the hammer at Christie’s in New York’s swanky Rockefeller Centre. It was part of a series of 10 portraits in a series titled “La Famille de Belamy”.
The artwork sold for $432,500 (around £350k), which was way in excess of its original estimate of $7,000-$10,000.
Slightly blurry, sort of looking like an Old Master and oddly positioned on the canvas, the print was signed “min G max D x [log (D(x))] + z [log(1 – D (G(z)))]” – which was a clue to its provenance.
In fact, the picture was created by an algorithm put together by Obvious, a Paris-based collective exploring the space around art and artificial intelligence.
There are numerous other examples of machines creating artistic works:
- As far back as 1958, a computer algorithm was used to compose The Illiac Suite, a string quartet in the Classical style. When played by real musicians, the work is indistinguishable from one written by human hand, and uses well-established musical compositional elements like counterpoint, rhythm and pitch.
- In 1997, JAPE, the ‘Joke Analysis and Production Engine’ was built at Edinburgh University by Kim Binstead. It was able to generate puns such as “What do you call a Martian who drinks beer? An ale-ien!” and “what’s the difference between money and a bottom? One you spare and bank, the other you bare and spank.”
Not great, perhaps, but probably up to the standard of many a Christmas cracker and when tested, many were found genuinely funny – at least they were by an audience of children.
- In 2017, AlphaGo Zero, an algorithm designed by Google’s Deep Mind division, played 4.9m games of Go against itself over 70 hours. The neural network initially knew nothing about Go beyond the rules of the ancient Chinese board game.
(Go is more complex than games like Chess, with many more potential strategies and moves.)
Not only did the AlphaGo Zero turn out to be the best Go playing algorithm ever designed, easily beating actual masters of the game, but according to its makers, David Silver and Demis Hassabis it “discovered new knowledge, developing unconventional strategies and creative new moves that echoed and surpassed the novel techniques it played in the games against [human Go Masters] Lee Sedol and Ke Jie.”
Go players are now themselves learning from, and imitating, the AI’s moves and strategies.
DEFINING CREATIVE OUTPUT
But are these examples really ‘creative’? What do we even mean by the word?
One of the earliest thinkers about machine-based creativity was Professor Margaret Bodenfrom Sussex University.
Her classic paper, ‘Creativity And Artificial Intelligence’, now over 20 years old, defined a creative idea by its properties. According to her, something was creative that was “novel, surprising, and valuable(interesting, useful, beautiful…)”.
By that definition, these outputs above are creative – they are new, surprising and for the most part they have some value in that they’re funny, clever, etc. ‘Portrait of Edmond Belamy’, in particular, also has an explicit monetary value.
Question that this definition begs though is, valuable to whom?
Rather than defining creativity by a property of the output, another thinker in this space, Teresa Amabile, a professor at Harvard Business School, says something is creative if it the right people agree it’s creative. It’s sort of like a Turing test for creativity.
“A product or response is creative to the extent that appropriate observers independently agree it is creative.”
(The Social Psychology of Creativity, 1983)
‘Appropriate beholders’being “those familiar with the domain in which the product was created or the response articulated”. In other words, other artists/creators and critics or curators.
By both these definitions, ‘Portrait of Edmond Belamy’, The Illiac Suite, JAPE’s jokes and AlphaGo Zero’s outputs are ‘creative’.
But can the computers themselves be said to be ‘creative’?
COMPUTERS AS CREATIVE
For many, the idea of a machine being ‘creative’ is just odd. Doesn’t creativity require ‘feeling’or ‘soul’? Something which computers obviously lack.
Computers have no cultural or human frame of reference to draw on, so how could they create something that resonates with humans? No lived experience.
For some it’s about explainability; a computer can’t tell you why it’s done what it’s done so how can it be creative? It sort of requires consciousness, something which computers are a long way from having (and which we’re a long way from even being able to explain).
Computers are like those elephants who’ve been taught to paint, or chimpanzees at a tea party. They’re just imitating human behaviour or doing what they’re told.
But how can something NOT creative make something that is creative? It’s a conundrum.
But there’s another way to look at this, which is that creativity isn’t the product of any individual, computer or human, but actually arises from a wider system.
This is the view of Hungarian-American psychologist Mihaly Csikszentmihalyi(surname pronounced ‘chick-sent-me-hi’)
“Creativity is as much a cultural and social as it is a psychological event… what we call creativity is not the product of single individuals, but of social systems making judgements about individual’s products”
(A Systems Perspective on Creativity, 1999)
It’s worth going a bit deeper into his model which has essentially three forces interacting.
There is the practitioner, the creative who ‘makes’ the work. Then there’s the culture that they live in. And finally, the ‘field’ – which is to say the gatekeepers who evaluate the work and decide what’s in or out (what Amabile might call “appropriate observers”).
Interestingly, the algorithm that created ‘Portrait of Edmond Belamy’ employed a form of machine learning called a ‘Generative Adversarial Network’ or GAN, which has three elements which are analogous to the elements within Csikszentmihalyi’s model:
DOMAIN: the team behind it fed the system 15,000 images, spanning 600 years of European portraiture, so the algorithm ‘knew’ what portraits look like. It understands the ‘cultural system’ it exists within.
PERSON: Then the ’generative network’part of the algorithm output candidate images, just as a practitioner does.
FIELD: And finally, a part of the GAN called a ‘discriminator’did the job of a gatekeeping – deciding which images to keep and which to rule out.
While the GAN mirrors the system model, clearly it doesn’t do the same job as it. The machine isn’t itself impacting culture.
A machine can’t in itself be part of a cultural system, and get past the relevant gatekeepers. It’s just a machine, so how can it live in the world or influence anyone?
In fact, in the case of ‘Portrait Of Edmond Belamy’ the machine needed its programmers, the Obvious Collective to represent its work, raise its profile, persuade the art world and Christies to take it seriously.
Indeed, one thing I would add to Csikszentmihalyi’s model is that what he calls ‘practitioners’ are not always solo people but can also be a team. This is especially true for certain creative endeavours such as cinema and marketing/advertising.
And I think that might be where the role of computers in the creative process becomes more understandable.
Driven by depictions of intelligent robots in popular culture, there’s a myth that AI’s should be completely autonomous self-contained ‘brains’.
In fact, the best AI-based systems allow machines to do what they’re good at and humans to do what we’re good at, combining both into a complete human-machine system.
Think of the role that the hundreds of people who work on feature films play – everyone from editors, cinematographers, actors and costume artists. The final film doesn’t belong to any one of them, but they all play a role, inputting into and impacting on the final film. It’s a creative system.
In the same way, computers really can be creative, but they do this as part of the creative team.
Sure, they can’t necessarily to do all the stuff that humans are great at, like negotiating with whoever is commissioning the work, influencing gatekeepers, or making the work culturally relevant but they can still be ‘creative’.
“It is not our goal to recreate the human mind – that’s not what we’re trying to do. What we’re more interested in are the techniques of interacting with humans that inspire creativity in humans.”
Rob High, Vice President and Chief Technology Officer for IBM Watson.
TRANSFORMATIONAL CREATIVITY – THE BIG ‘C’
Margaret Boden categorised three kinds of creativity, Combinational, Exploratory and Transformational.
Transformational creativity is the most extreme, representing big breaks from previous thinking. Creativity with a big ‘C’.
For example, Cubism’s break with representational art, the Bauhaus movement’s break with ornamentation in architecture, Bebop’s break from previous forms of jazz, or Rap’s break from sung lyrics to lyrics spoken in poetic forms.
Other people call it ‘eminent’ or ‘domain changing’ creativity.
Boden’s point was that while machines can produce all three types of creativity, the third, Transformational, was the most difficult as it also “depends largely on unarticulated values, including social considerations of various kinds. These social evaluations are often invisible to scientists. For sure, they are not represented in AI-models.”
To which I say, ‘not yet’.
Computers using machine learning, adversarial networks, ‘recurrent independent mechanisms’ and a host of other as-yet uninvented techniques, are poised to do amazing things in the creative industries. The faltering first steps of the machines that produced ‘The Illiac Suite’,‘Portrait Of Edmond Belamy’, JAPE’s jokes or AlphaGo Zero’s incredible Go moves are really just the beginning of a whole journey which will see machines becoming more and more an accepted part of the creative process.
Artificial Intelligence means different things to different people, but I will be using it in its most general sense to cover all the tools and techniques that allow computers to learn from data, spot patterns and make decisions based on them.
And others, also see Søren Klausen:
“A creative product is something that would appear creative to an appropriate audience under suitable conditions…
“Being creative is having a propensity for bringing about products that have a propensity for being recognized as creative”
(The Notion of Creativity, 2010)
Or even David Gauntlett:
“Everyday creativity refers to a process which bring together at least one active human mind, and the material or digital world, in the activity of making something which is novel in that context, and is a process which evokes a feeling of joy.“
(Making is Connecting, 2011)