Using Artificial Intelligence for Creativity
What do you get when you cross a frog with a street?
I'll come back to that later.
For at least the last 70 years, many attempts have been made to create art using computers and more recently artificial intelligence. Some of these attempts resulting in truly amazing outcomes, but many others were not so groundbreaking. Regardless of the outcome, this question comes up time and time again:
‘Can computers really be creative?’
To begin answering that question, we need to understand what creativity is. Everyone from Albert Einstein to Steve Jobs has defined creativity in their own way and the Oxford English Dictionary defines it as ‘’the use of the imagination or original ideas." This definition makes sense to me when we think about what we typically consider to be creative work. Art, music, and marketing all require the producer to use their imagination and form original ideas.
“Creativity is intelligence having fun.” - Albert Einstein
An alternative look at defining creativity is Sir Ken Robinson’s three-step relationship which he explains in his TED Talk, Do schools kill creativity? Sir Ken talks about imagination being the process of bringing to mind things that are not present in our senses, which could be things that exist or do not exist. Creativity, the process of developing original ideas, often has imagination at the source of these ideas. Finally innovation, the process of putting these new ideas into practice.
What does AI actually do?
Well, what AI does is remember. AI remembers what it has seen before in a very efficient fashion and generalizes slightly on that information. In mathematical terms, an AI model takes a past training sample and generalizes around it in small increments.
A very simplified example of this can be found from Untapt, an AI technology company that uses algorithms to match job descriptions with candidate profiles. The model (demonstrated below) uses millions of data points to read the natural language in the candidate profile and job description, to create an Artificial Neural Network (ANN) score. This score is then used to rank candidates and find jobs. The model then uses small incremental changes in order to develop and improve accuracy over time.
Amongst companies deploying and building AI tools or products, it is generally accepted that the quality of data is the key to success or failure. It doesn't matter how brilliant an algorithm is, if it is set to run on poor or incomplete data, the outcome will not be a success. This hypothesis is defensible if we accept that what AI does is remember and the data is incorrect or poor in quality then the learning will be inefficient or incorrect.
Is Human Creativity Really Creative?
If AI is just reproducing what a human has invented, according to the definition, how can that be considered creative? Isn’t a lot of human creativity just re-contextualizing content?
Shakespeare, for example, took the vast majority of his playwright from sources other than his own imagination. It is believed the source of Romeo and Juliet was Brooke’s 1562 poem, and Hamlet was inspired by ‘Gesta Danorum’ written by Saxo Grammaticus in 1200 AD. In which the main character was Amleth, an Anagram of Hamlet.
Isn’t Clueless just Pride and Prejudice set in the modern-day? and how many superhero movies are there? Yet we consider these to be creative work, we don't say that’s derivative art. These are all examples of taking a concept someone else has done and applying it to a new area, or ‘art’.
What we decide is creative or not comes down to context. How we as humans base the decision on the experiences we have. We have experienced art class in school, baking a cake and seeing paintings sell for millions of dollars. Do we, therefore, give ourselves too much credit when it comes to creativity?
Is an AI-based model using historical data, really any different?
Computers Being 'Creative'
AI being used to create new ideas isn't new. Some of the earliest reported examples are computer-based joke-telling, which dates back to the early 1990s. Luckily, since then the types of programs being built have developed.
In 2015 film and music fans at South by Southwest Festival in Texas were able to use the Chef Watson app, the app used IBM’s Watson to create unique recipes by combining ingredient data about the way we perceive food. The app mapped similar flavor profiles and replaced ingredients with ‘relevant’ alternatives, to some interesting outcomes, like the Baltic Apple Pie, which includes a layer of pork.
By using AI rather than development chefs, the tool manages to remove any human bias or preconceived ideas of what may or may not work.
Arguably this example shows that AI has taken the work of a human, taste, and re-contextualized the data. This is an important point. When building ethical AI it is important to consider the effect on humans, I believe the intention should be to augment our daily lives, not completely automate them, and what better way to demonstrate this, than with a new pizza flavor!
Chef Watson is a great example of assessing if an AI-based tool can meet Sir Ken Robinson’s three-step creativity model. Innovation; Chef Watson brought to mind combinations humans had never previously thought of. Creativity; Chef Watson developed these flour combinations based on its learnings to find commonalities, and finally Innovation the system developed recipes that could be brought to life.
Is this the last blog I'll ever need to write?
Another well-known example of AI being creative is the GPT-2 tool. ‘Talk to Transformer’ allows you to enter a word or phrase and the tool will start writing sentences that follow on. While the output doesn't always make 100% sense, it does create coherent work.
There is also a major advancement in using AI for creativity, and even more advancements have been made when it comes to picture-based creativity. Multiple examples like the GAN (Generative Adversarial Network) technique has been used to create completely original, random images of people. The best example I could find of this is the website ‘thispersondoesnotexist.com’.
Or Google’s Inceptionism, which takes huge amounts of image data to create a model for recognizing and creating images based on word input. The neural network creates lots of layers, which are gradually refined in order to display these interpretations. This example used the word 'Banana' and the picture on the right shows a pretty good outcome.
While not to everybody's taste, models like this have already successfully created ‘art’. Inducing “Edmond de Belamy, from La Famille de Belamy” the first portrait generated by an algorithm, which sold at Christie's for over $400,000.
And the punchline is...
The final example I’ll talk about is probably the most well-documented. AI-Generated Jokes. Multiple examples of this have been created over the last few years and they follow structures that humans have been using for years.
So, “What do you get when you cross a frog with a street? A main toad"
Ok maybe I did hype it up a little too much, but it's funny because the joke has context, Road rhymes with Toad, and Toad is another name for a Frog.
What's next? and can AI be creative?
Each of these examples highlights a key theme for the next few years of AI advancement, not just for creativity but for AI in general. As we learn how to live alongside and augment humans rather than automating them, there will be many more cases like creativity in which we have the opportunity to define the future of AI. Success will depend on how we learn from what is developed and how we accept the outcome may not be perfect initially, but the opportunity to better our lives is vast.
Can what a computer does be considered creative? I believe so, If we agree that not all human creativity has to be unique to be creative, the same can easily be said for computer-based creativity, which will have some fascinating real-world applications in the years to come, and maybe, even more, $400,000 art sales.