Are machines capable of being creative? The concept at first appears to be an oxymoron—“mechanical” is an antonym of “creative.” Yet for centuries, people have imagined the potential of automated thought and creativity. While they made some naïve mistakes, they often had a broader perspective on how this problem fits into the larger context of philosophy and society.
The history this book presents is a miscellaneous one. It draws from the history of the arts, magic, religion, toys, games, staged entertainment, philosophy and language. All of these threads are part of the same story: the invention of machines to automate the creation of new designs or new ideas.
The Kaleidoscope Pattern
For all their diversity, there is a pattern common to many of these devices, one that will show up again and again throughout the book. The pattern can be seen most easily in the construction of a kaleidoscope. A kaleidoscope is a very simple machine consisting of three parts:
- Colored bits of glass.
- Two mirrors that impose a regularity, or formal structure.
- A means of randomizing the arrangement.
These three parts show up in virtually every attempt to make creative machines (of which the kaleidoscope is one example). They embody a theory of creativity that is centuries old: that the random rearrangement of interesting ideas or images along with an enforced logical structure is the way that our minds are able to invent new objects.
Such machines are wonderful and fascinating in their own right. One of the main purposes of this book is simply to gather many examples of these machines and exhibit them together as examples of early efforts at automating creativity. Because they fall in the cracks between art and science, many of them have been nearly forgotten by both artists and scientists, and this is a tragedy.
For all their beauty and intricacy, however, these machines are ultimately unable to deliver on their promise of true, sustained creativity. Eventually, the new images created by a kaleidoscope no longer have the power to delight and intrigue. We come to see the theme behind the variations, and each individual work no longer brings anything new to our understanding of that theme. A machine that was truly creative would be able to find ways to keep being new, and to be new in new ways.
No one has built such a machine. Despite the promise of evolutionary algorithms and machine learning techniques, every attempt so far has eventually petered out. After some initial surprises, all of these programs in the long run end up coming up only with new variations on the same themes. No programs have been run for year after year without human interference, coming up with new creations that are enthusiastically exhibited and admired.
Probably the most famous program to create visual art is Harold Cohen’s AARON. AARON’s work has been exhibited in major museums around the world, and its output has been described as creative by both artists and computer scientists. Each new painting created by AARON is original, and can be surprising even to Mr. Cohen. Yet he feels that despite AARON’s success in the art world, it is still not creative. The reason he feels this is that AARON is essentially a kaleidoscope at heart. It has a model of a human figure, something like a paper marionette. The pose of the model, the proportions of the limbs, the placement on the page, the colors of the body segments, are all chosen randomly. The random values have constraints that keep them within reasonable bounds. A similar process generates plants and backgrounds for the scenes. Once all of this has been placed, a separate routine traces the outline of these shapes in a semi-random fashion. How is it, then, that so many people take AARON to be creative? The answer is that the program is taking advantage of certain very powerful illusions.