Creativity is the most important factor to ascertain a company's future success. Confronted with an increasingly complex world of shifting global power relations, disruptive technologies and growing volumes of data, enterprises will rely on creativity to keep the advantage. That is the outcome of a survey conducted amongst 1500 CEOs operating worldwide published by IBM in 2010.
So what do you do when you're a technology multinational and you want to stay ahead of the curve? You build a creative machine.
The as yet unnamed system combines vast amounts of data with decision-making algorithms to generate truly creative products. The system demonstrates its artificial creativity through the production of culinary recipes but the method can equally be applied to other creative fields.
Computational creativity is the latest feat coming out of the T.J. Watson Research Center, IBM's innovation hub which is also responsible for chess computer Deep Blue and artificial intelligence system Watson. In a recently published paper researcher Lav Varshney and his team explain how the system works.
To be able to build a creative system, you first have to define what creativity is. For their purposes the authors formulate a utilitarian definition of creativity: the generation of a product or service that is novel, useful and valuable to a knowledgeable social group. In the case of the artificial cook, the level of creativity of its recipes is judged by a team of top chefs based on originality and palatability.
Computer generated creative products aren't new, there are programs which spit out poems, musical arrangements or mathematical proofs. However, these AIs are unable to evaluate their own work in order to determine whether it is any good. In all cases a human needs to weed through the output to pick out the valuable pieces.
The novel contribution of IBM's team is the introduction of a selective component into their system. The creative machine can not only generate millions of new recipes, it can also select the ten most tasty ones.
To pull this off the system relies on tonnes of data. Varshney and colleagues fed it tens of thousands of recipes and other data about cooking such styles and preparation methods. They also added content about related but distinctly different knowledge domains because creative cooking is more than a mash up of existing recipes. Novel ideas emerge when tried and tested know-how is exposed to new information.
The artificial cook has to assess its own output to select the most original and best tasting products according to human standards. To give it that ability the IBM team used data-driven assessment to mimic human perception.
The way we experience flavor is largely determined by how food smells. To create a human perception model the team uploaded data about the properties of chemical compounds and information about how their scent is perceived by humans. To evaluate its dishes the creative machine breaks them down into their constituent chemical compounds and rates them according to how much humans like them.
The system also assesses the originality of a dish. Turns out there is a mathematical formula for that. Called Bayesian surprise, it draws on a database of existing recipes to calculate the probability an observer already knows the object.
The creative machine can operate autonomously or in interaction with human beings. In their conclusion the authors point out that creativity is challenging for both humans and machines. People find it difficult to grasp vast amounts of information and therefore have less material to play with. We're also very slow. “A computational creativity system can test quadrillions of ideas at once [...]. Such creations may offer advantages by being completely ‘outside the box’ through large jumps in thought rather than gradual evolutionary changes.”
Image: Italian grilled lobster made by the creative machine, photo courtesy IBM