Can machines be creative?
AI is enabling a computer to act like a human and with modern advances in machine learning, computers are now able to learn from data autonomously. When will they or when did they cross the line into creativity? AI models can now generate images, sound and video, but can we really replace the need for art and creativity? Invention and innovation is so much a part of being human that you might even say that it is what makes us human.
I was recently at a talk by Demis Hassabis, founder of DeepMind, who described the journey he went on to develop AlphaGo, a model that could beat the worlds greatest Go players at their own game. The model was not trained on a database of existing moves, but instead it was given the rules of the game and at first played at random. The model played against itself and improved. That improved version then replaced the previous version. The model continued to learn against itself until it was better than a human.
It was a remarkable achievement in of itself, but more extraordinary was the ability of the model to come up with moves and strategies that had not been seen before by experts. In particularly move 37 in a game against Lee Sedol has been praised as creative and unique. Something that the machine came up with itself that it did not learn from observing human games; to the community it really did appear as if the machine had come up with something that a human had not.
DeepMind went on to refine their deep learning techniques and apply them to real world problems. In particular looking at protein folding. A computationally expensive problem that looks to discover the 3D structure of proteins. It might take a PhD student their entire degree to discover the structure of a single protein. There are a hundred thousand proteins in use in the human body and many millions across other organisms. Understanding their structure is important for drug discovery and medical research.
Last year Demis Hassabis and John Jumper at DeepMind won the Nobel prize for chemistry for their contributions to Alphafold, an AI model that can discover the 3D structure of a protein as accurately as a human researcher. In this way contributing to human knowledge and solving a problem that with conventional computational techniques and know how we were struggling to make progress. This phenomenal achievement is already changing the field of biology with two million researchers accessing the database of protein structures so far. The potential for applying similar techniques to other problems is indeed exciting, but are we reproducing human creativity or are we generating faster computational techniques?
No doubt I can ask the various available AI models to draw a picture of my village in the style of Van Gogh, but will we ever get to the point where the AI generated art no longer needs Van Gogh to have lived? Edvard Munch created several versions of his famous work The Scream in his lifetime. The painting has inspired many reproductions, including an emoji which I once saw made of post-it notes in the DeepMind offices in London. The painting has a number of interpretations, one being the horrifying screams of the lunatic asylum where Munch’s sister had been committed. To witness the pain in the subject is one thing when you first see the painting, but to think of the unending suffering of a loved one is another. The image is powerful and I have no doubt that AI will be able to generate images that invoke deep feelings, but the back story also matters.
Tracey Emin is a British artist known best for her work entitled My Bed. The unmade bed, litter and dirty sheets represented days she spent in her room overcome with depression. The work divided the critics at the time. A messy bed is nothing without her suffering, the shock from the public exhibition of something so ordinary, the repulsion at the way she had been living. When we look at the bed we don’t just see the bed, we see her pain, and we see our judgement of the way she has dealt with that pain. The work challenged how we think about women and about what it means to be a woman.
If an AI generated an image of a messy bed now we’d attribute it to an imitation of Emin’s work. If it generated something new and similarly odd we might say it was hallucinating. Surely that is not a pipe! Without the human experience, can it be art or is it just a bed? To me the human process of creating the art is very important. Every piece of art in my home has a personal connection to me that goes beyond the pure aesthetics of the work. I watched David Nash work at Kew Gardens and the death of each tree was part of the story of each sculpture. I have a book of drawings that Picasso did for and of his daughter and the love that spills from the pages of sleeping babies and colourful chickens is so joyful.
But can we develop models that understand the human connection? I don’t know Emin, Nash, Munch or Picasso. Would it matter to me if they existed or if their lives and experiences were generated as a back story to embellish and explain the art? Would it prevent me from feeling a connection to my own life? If I don’t care if artists really exist, will I one day not care if my friends exist? Certainly we have demonstrated that machine learning can create inventive processes that go beyond the reapplication of existing knowledge and that is a fantastically powerful step, but creativity and the way it makes us human is multi faceted and so I think we are long way off from being obsolete.
However, if someone told me that the banana was a joke thought up by an AI then I would totally believe them.
