Introducing Computational Creativity

Computers that Create Computers Computers that Create Computers © Tony Veale, 2014

Available from all good Web megastores Or see: http://RobotComix.com

Computational Creativity (CC) is the scientific study of the creative potential of computers. As such, it is several things at once:  The engineering study of how to build machines that “create” to a human standard (if not in a human fashion)  The algorithmic analysis of human creativity, using the mind-as-computer metaphor to explicate how humans “create”  The study of what it means to be “creative” in a world where humans are not the only creators

Meta-Creativity is achieved by creators who build systems to be creative on their behalf. The combination of computational system and human creator often yields creative results that could not be achieved by either on their own. Much of what passes for CC in the field of AI is really Meta-Creativity: All builders of CC systems are Meta-Creative insofar as their systems create for them. But not all meta-creative systems are true CC systems. Some meta-creative systems are merely generative tools.

Meta-Creativity: When we create machines that that create for us

Strong versus Weak Computational Creativity  Strong Computational Creativity is the study of fully autonomous machines that can truly “own” their own outputs  Weak Computational Creativity is the study of semi-autonomous software tools that exploit CC concepts to merely support and foster greater creativity in humans  The distinction is clear in in principle, but blurred in practice

So how do we distinguish between “strong” and “weak” CC systems?  Strong CC systems not only produce novel and useful outputs, but critique, rank and filter their own outputs to select only the very best.  Weak CC systems produce outputs that they themselves cannot appeciate as creative. The human user must filter and classify its outputs for it.  A strong system can be creative on its own, without a human in the loop.  A weak system can be generative on its own, but needs a human to be creative.

Who owns the actual creation? Who is the main creative agent? Can our system ever truly surprise us? Ever truly surpass us? Important Questions to Ask about any “Creative” System Gread and Gutter Bread and Butter, you moron!

Weak Meta-Creativity with Programmable Minions Large groups of uncreative minions can collectively achieve a creative outcome for a meta-creator just by following a simple algorithm. Just think of the poor unfortunates who hold up the colored cards in North Korea’s mass games. By holding the right colors in the right sequence, each person contributes a single pixel to a larger stadium-wide image that they themselves cannot see or appreciate. Instead of poor benighted humans, we can use software ants. Our ants are let loose on a digital image of our choosing, and given some simple rules of life. Each ant is programmed to look for food, which might be defined as pixels of brightness or high-intensity color. Each is programmed to leave a pheromone trail of color pixels in its wake, so that other ants can follow a successful forager to the most bounteous areas of the image. Let the ants loose, and they will collectively overpaint the underlying image in quite artistic ways.

A software swarm is a powerful tool for a human meta-creator

CC Software can be influenced by its users (and vice versa) in surprising subtle ways

Color-Mapped Depictions of Evolved Mathematical Formulae -- Penousal Machado’s NEVAR Penousal Machado’s NEVAR system uses genetic algorithms to simultaneously explore a space of complex mathematical formulae and a corresponding space of rendering functions for turning this high- dimensional formulae into colorful 2D images. NEVAR interacts with its users in interesting ways. As it explores its spaces, it presents its works in progress to be critiqued (like/dislike) . NEVAR uses this feedback to taior the fitness function of its genetic algorithms, so as to adapt to the aesthetic sensibility of its user.

Portrait of the Artist’s Owner As A Young(-ish) and Strange Man -- Simon Colton’s The Painting Fool Simon Colton’s Painting Fool system aims to produce more conventional, less mathematical forms of visual art, simulating a wide variety of media (canvas, paper, ink oil, acrylic, pencil, pastel, charcoal) and tools (pens, pencils, brushes and stroke styles). The Painting Fool typically begins with a digital image or a live video grab, and uses this pixel map as a guide to producing a corresponding painting. The Painting Fool uses a variety of user cues (such as simple affective analysis of the user’s mood) to make artistic choices and yield a non-deterministic choices.

Co-Creation may involve Subtle Interactions between the software and its human user Creativity needs Engagement: Can Machines Handle the Unexpected? Can a rule-defined computational system genuinely engage with a problem and react sensibly to the truly unexpected? If a CC system relies on rules to handle the unexpected, in what sense is the unexpected really unexpected? Rules define the known knowns and the known unknowns of a system (to quote a much-maligned military philosopher of sorts). But what about the unknown unknowns? Well, how do humans handle the unexpected? Let’s call the unknown variable “WTF!?!?!”

Many problems with technology are caused by its human operators. Consider Wegman’s bakery in New York state. This cutting-edge business allows customers to email in the images they want on their cakes, and a special printer uses food dyes to print the image onto the cake icing. But what if the image makes no sense? What if the email contains some unexpected content, like HTML markup? This happened when a customer used Microsoft Outlook to email the desired text for a cake. The Wegman employee blindly cut’n’pasted Outlook’s added HTML markup directly to the printer, resulting in the bizarre cake overleaf. Can we really blame the technology? Happy Birthday

Can a CC System do better than this human? Could it do worse? Hmmm, Delicious!

Acquiring Aesthetics: Can Machines Evolve their Own Aesthetics? I’m a crockpot chef designed by Dan Ventura that uses ML to learn to concoct my own chili dishes by analyzing scads of online recipes from FOODnetwork.com Machine Learning (ML) is a branch of Artificial Intelligence that allows computers to learn from experience, or from large amounts of past data. ML can, for example, learn the aesthetics of cake design from a large catalogue of professional cake designs, or from similar (easier to find) data such as Hallmark’s online inventory of greeting card designs. By learning a robust model, a CC system can reject goofs like Wegman’s cake.

Defining Creativity: Physics Envy? Creativity is a folk notion, not an objective mathematical concept for which we can stipulate a clear-cut definition. There is no formula for creativity, no hidden formula waiting to be be discovered by theoretical scientists. It is a social construct that we can explore empirically, with computational models.

Creative is as Creative does Rather than search for some hidden formula underpinning creativity, CC seeks to observe creativity in individuals and in groups, to understand our implicit criteria for applying the label “creative” to an artifact, idea or solution.

1. The answer has novelty and usefulness, either for an individual or for a society 2. The answer demands we reject ideas that we had previously accepted 3. The answer results from intense motivation and persistence 4. The answer comes from clarifying a problem that was originally vague Alan Newell, Cliff Shaw & Herb Simon There are no qualities that are necessarily present in all instances of creativity, nor groups of qualities that are collectively sufficient to guarantee the aptness of the label “creative”. Instead, we tell each other narratives of creativity.

We can call on different narratives of creativity to understand useful novelty in different contexts. One commonplace narrative goes as follows: a novel response to a common problem achieves a disproportionately effective outcome. For example, when controversial ex-prime minister Tony Blair released his autobiography A Journey offering a defense of his tarnished legacy in Iraq, protesters reached for the familiar response: they threw shoes at him when he spoke at book shops. Student Euan Booth launched a quieter but more subversive (and ontologically disruptive) protest: Booth asked people to move Blair’s autobiography from the Non-Fiction shelves of bookshops to the True Crime and Fiction sections instead.

“This is a peaceful and mischievous way of making your point if you feel the same way. It’s a non-violent display of anger using the materials given to me – his book and the crime section – they're both there, I just put them together.” Facebook protester Euan Booth, quoted in The Telegraph on 4th September, 2010 Booth’s solution is cheap, easy, fun and resourceful; it makes non-obvious use of familiar objects and settings; it makes a virtue of limitations; and achieves multiple outcomes at once.

Creative Psychology 101 Computers do not have to be creative in the same way as humans, or produce creative outputs that can pass for human outputs. However, the more knowledge of human psychology that a CC system can draw upon, the better positioned it is to produce outputs that humans will appreciate.

A Tale of Two Velocities The psychologist Daniel Kahneman offers a useful metaphor for the competing forces that make up human cognition. The mind is not one unitary system but two interacting, and often adversial systems: System 1 is fast, eager, always on and always ready to jump to conclusions based on scant but suggestive evidence. System 1 needs no prompting, and underpins many of our intuitions and rapid “blink” responses. Yet this quickness is also a weakness: it relies on shortcuts and is often mislead. System 2 is slower, less eager, and easily depleted. However, System 2 is capable of analyzing a situation in greater depth, of using rules and conscious reasoning to reach a conclusion. System 2 lacks System 1’s rapid responsiveness, but is more likely to be right!

So I bought a bat and a ball for $1.10 in a yard sale. A bargain! The bat cost $1 more than the ball. Hmmmm. So much did they each cost? Let’s see … $1 for the bat, and just 10¢ for the ball? No! The right answer is $1.05 for the bat. System 1 is mistakenly drawn to the $1 and 10¢ as primed by the question itself. A Simple Math Problem: System 1 vs. System 2

Creativity involves a told- you-so dialogue between System 1 and System 2.

OK, big brain, memorize these twelve words: Charter, Voyager, Analogy, Density, Cottage, Tonight, Crumpet, Trilogy, Fixture, Brigade, Cluster, Holster

Done? Now fill in the missing letters in these word grids.

makes this task harder! The memorized words are primed, and any primed elements can become fixations for System 1 regardless of whether they are relevant or not to a problem

Jokers play mind games with an audience’s System 1 processes

System 1’s unstated assumptions can push our minds into mental ruts when solving even simple puzzles. Escaping from these ruts via System 2 can yield a sense of creative satisfaction. Re-assemble these four pieces to make a letter “T” shape. Quickly Now!

The yellow dot marks an outside corner, not the internal hole that System 1 assumes needs to be filled! This simple insight is a tiny breakthrough

Even tiny deviations from System 1’s assumptions and habitual scripts can yield big differences in outcomes! Most jokes have a butt, yet the biggest butt is System 1. Jokes rely on System 1, yet expose the failings of System 1. System 1’s automatic associations hide the ambiguity of everyday life. A successful creator revives this ambiguity by subverting System 1

Minor variation on an established success story is deemed “creative” when it yields disproportionate results, like humor. But “safe” variations that just achieve incremental results produce “pastiche”. Daring to Go Beyond Pastiche and “Safe” Generation

The simplest computer script can generate a billion novel outputs. But creativity is more than mere generation: it requires the selective generation of outputs with not just novelty but demonstrable utility. “Mere Generation” is easy, but creativity involves more!

Borges’ Library of Babel To appreciate the importance of selectivity, and the potential for mere generation to overwhelm a semiotic system, consider the short story ‘The Library of Babel’ by Jorge Luis Borges. Borges imagines a vast (but finite) library containing every book of fixed upper-size and alphabet that was ever written, or that ever could be written. Want a sequel to War and Peace, or the next Harry Potter that JK Rowling may one day write? It’s here for the taking. If you can find it! That’s a big IF: the library contains a great many hidden gems, but these are lost in a sea of random possibilities. The library contains many conflicting guides to itself (as these are books too), but no guide is authoritative, and so the library is unusable.

Borges’ story takes mere generation to absurd levels, but shows that mere generation is very easy, while selective generation is very hard. Borges’ Library of Babel

Call me Ishmael. Or Jehovah. Or Buddha. Or Siddhartha. Or Rama. Or Larry. Or The Big Kahuna. Or Skinny Pete. Or Badger. Or Heisenberg. Oh, I could go on, and on, … Arthur C. Clarke’s The Nine Billion Names of God We tend not to value the products of mere generation, or to respect the process enough to call any output “creative”. Consider Arthur C. Clarke’s short story The Nine Billion Names of God, in which a sect of monks hires a mainframe and some programmers to enumerate all possible nine billion names of God in their unique mystical alphabet. The monks believe the world will end when all 9 billion permutations have been generated. The programmers scoff at this belief, seeing no good reason why the universe should be responsive to mere generation. The story is humorous because the monks turn out to be right: the stars in the sky begin to disappear when all 9 billion names are generated!

Like Borges’ The Library of Babel, Clarke’s “Nine billion names of God” explores the limits of Mere Generation. It is all too easy to generate everything from a given grammar or alphabet, but will anyone be patient enough (short of the “big G”) to sift through the deluge of outputs to find those that have any value? Clarke’s story is humorous because it defies common sense (we intuitively side with the programmers, not the monks). God does not play random word games with the universe. Do not touch the Robo-Buddha while it is unwinding Do not touch the Robo-Buddha while it is unwinding

The quality of a computer-generated output lies as much in what is not generated, or is not selected, as in what is generated. In truly creative generation, Many Are Called But Few Are Chosen.

You call this a sandwich, @sandwiches_bot ? It’s a f--king sh-tty sandwich, is what it is ! So while @sandwiches_bot is a fun generator It is a mere generator. Fun but not creative Creativity here requires culinary knowledge

So Creative systems must steer a course between Pastiche & Mere Generation

We think of Creativity as a single concept, but it manifests itself in many guises. It can be “implemented” in many complementary ways.

There is no single cognitive mechanism of creativity, no single hammer than can handle every nail. Rather, creativity involves the application of diverse cognitive strategies, individually or in combination, to suit the particular needs of a given scenario. With no one-size-fits-all strategy, and no definitive guide as to which one to use in which context, the best we can offer is a comprehensive checklist of strategies. A creator must understand what options are available, and experiment with these options to find the right mix.

Naturally, there is no definitive checklist for Creativity. Rather, every creator will carve up the world, and their notion of creativity, into different goal-specific concepts and actions. Thus, an artist may have a different checklist of go-to strategies from a theoretical scientist or a commercial inventor. The Belgian surrealist Rene Magritte proposed a checklist of strategies for producing his distinctive brand of visual wordplay. He intended his checklist to be an art manifesto: Les Mots et Les Images

A very famous creativity checklist (for humans) comes from Alex Osborn

Osborn’s checklist, like others, suggests a menu of tools to explore a space

Strategies allow us to explore a conceptual space. Can we be more revealing?

LIVE  LINE  LINT  LENT  LEND  LEAD  DEAD Consider Lewis Carroll’s game Doublets, in which we explore a space of words Lewis Carroll invented a simple word game that illustrates the role of search in problem-solving. The game is called Doublets: Two words of equal length are chosen (the doublets). The goal is to turn one word into the other by changing just one letter at a time. The catch? Each intermediate state must be a word too. The creator of a game instance must search for two related words (like LIVE & DEAD) that can serve as interesting doublets. The player of each game instance must search for the shortest (or most interesting) path between words to link doublets. This exploratory search can be quick & insightful, or slow & plodding.

Game creators & game players alike must each build their own space to explore.

In Exploratory Creativity, a creator explores an established space of ideas A creator searches a space to find a goal-state that is both novel (as yet undiscovered) and valuable (e.g. useful, workable, efficient). One needs a state space to search, and a value metric to guide the search.

Conceptual spaces contain states of diverse value & use An explorer navigates a conceptual space, looking for areas with states of high value An explorer searches a space using a specific value metric for its states as a guide Often, we seek a valuable state. In Doublets, the path itself is the creative product.

Conventional route to established areas of value in the space Novel shortcut, identified by a pioneering explorer If we can find a novel pathway to a valuable state, this solution is doubly creative Future explorers may also use this short-cut, making it conventional

Whoa! Truly pioneering explorers may transform the space itself, thus changing the rules

W.A. Mozart was a virtuosic explorer of the space of classical music. Time and again he identified complex musical states (concertos, symphonies, etc.) of astounding beauty and resonance that have since become landmark states in the classical space. Mozart inspired many others to explore the same space, but he did not change the space. Later musical theorists such as Arnold Schoenberg changed and enlarged the space of Western music in radical ways, much as Albert Einstein transformed science’s Newtonian conception of space and time. Einstein’s transformed space allowed physicists to find a solution to Mercury’s odd orbit, having failed to find a solution in Newton’s space. Is Schoenberg’s music better than Mozart’s? Let’s just agree it’s different.

Conventional established space, as seen & explored by everyone else A larger space, as transformed by a pioneering searcher Why transform a space? To find new areas of production unreachable in others. States of value that are not present or visible in the conventional space

A creative explorer of a search space aims to find states that are both novel and useful/valuable according to some metric. It can be frustrating to get lost in a valueless area of the space, or to realize that one has been mislead by faulty intutions into searching the wrong part of the space. Narrative jokes do precisely this: they exploit (and reinforce) our assumptions so as to fool us into exploring a space of possibilities that are simply not relevant to the understanding of the given narrative. The punchline, when it finally comes, makes no sense at all in the part of the space to which our mistaken System 1 assumptions have lead us. To make sense of the punchline, we call on System 2 to resolve the impasse and take us – chastened but relieved – to a very different part of the space. Regrets? I’ve had a few.

In complex spaces, the long-way around is the intelligent (or only sensible) route to a goal. Only the joker knows! A too obvious short-cut can tempt System 1 to rush to judgment.. Narrative jokes exploit false short-cuts to trick a listener’s System 1 System 2 realizes too late that the short-cut is false, and only a longer route is really meaningful!

How do we know if we have reached our goal as builders of Creative Systems? How can our systems know?

Personal Firsts versus Truly Historical Creativity Creativity theorist Margaret Boden distinguishes two types of creative outcome: P-Creativity and H-Creativity. P identifies an outcome that is original to its creator, H an outcome that is original to society as a whole. A demonstrably P-Creative machine can also be H-Creative in principle.

Claims of H-Creativity require societal evaluation, though a human evaluator is more likely to ascribe creativity to another human than to a machine. So it is tempting to evaluate CC systems by having them pretend to be humans, as part of an elaborate Imitation Game. It is a temptation we should avoid: We are not in the business of building fake humans!

Imitation Games encourage a race to the bottom for CC/AI research, by encouraging researchers to emulate the worst (or most easily) simulated human qualities, such as a lack of engagement, coherence and profundity. Experience with Turing-style tests in AI suggest researchers emphasise an ability to fool evaluators over an ability to impress evaluators.

No, Mr. Bot, I expect you to TRY! Alan Turing imagined a thought experiment in which a human evaluator would have a discussion of real substance – about literature, poetry, art – with a putative machine. A CC system than can discuss its own influences, and share its inspiring examples, motivations, successes and failures, is much more likely to impress as a truly creative producer.

A Parting Note: Creativity is not an objective phenomenon. A computer, like any other creator, can only offer up its outputs in good faith, for the world to evaluate. Computational Creativity is ultimately the study of how a machine might show such good faith.

Introducing Computational Creativity

More Related Content

What's hot

Viewers also liked

Similar to Introducing Computational Creativity

More from Tony Veale

Recently uploaded

Introducing Computational Creativity