Creativity and Artificial Intelligence

DILOMITA - AI TECHNOLOGY > Blog > Uncategorized > Creativity and Artificial Intelligence

Artificial intelligence is one of those buzzwords that we see everywhere now. It can be used to detect credit card fraud, play chess and even drive cars. But these are all tasks which we are more or less willing to give up and admit that computers can perform better than us — it seems reasonable that one day or the other, computers will be driving around, coordinating the traffic in our cities.

But what about art, or creativity in general? That has been something which we could always trust on to be better than computers — when we talk to Siri, it is hard to image it will ever be able to paint something like Dali, for instance. However, we used to feel the same way about chess, years ago. It was the pinnacle of the human intelligence, and way too complex for computers to grasp, until IBM’s Deep Blue amazed the world by beating Garry Kasparov, considered the best chess player of all times. Since then, computers have excelled humans not only in chess, but in many other games, including Go, a Chinese board game, way more complicated than chess. Could the same happen with creative arts?

We can see slow but remarkable progress in the field of creativity: there are computers capable of painting, creating songs, and even writing articles. To be honest, they can be a bit simple sometimes, but they get better every year, and harder to distinguish from human work. How is it possible? How can AI be creative?

There are essentially three ways in which AI can be used to create new ideas:

  1. Producing novel combinations
  2. Exploring the potential of conceptual spaces
  3. Making transformations

Currently, computers (and scientists) are more focused on the first two. The third one, however, has the most potential for disruption and will probably be more explored in the years to come.

Producing novel combinations

Producing novel combinations is possibly the easiest of the three to be explored by a computer, and is very close to making analogies or jokes.

“What kind of murderer has fibre? A cereal killer”.

Not very funny, but you can see there is some logic to it, right? In 1996, AI doctorate Kim Binsted created Jape, a computer program capable of generating puns (this was one of them). Although not always funny, we can understand how a computer can make such jokes by assessing phonetic similarities between words (serial and cereal, for instance) and exchanging them. Much of what Jape does, however, does not make sense and has to be pruned later on by a human, to curate what is funny and what isn’t.

Exploring the potential of conceptual spaces

This is where things start to get interesting: imagine feeding an AI with musical “grammar”, the basic rules of music composition, along with with a list of signatures from famous composers such as Beethoven and Stravinsky: their mannerisms, typical harmonies and melodies. That is, the stuff that makes Beethoven sound like Beethoven. Then, ask this AI to write a new song, that sounds like it was written by Beethoven. Well, this is what David Cope did with EMI (Experiments in Musical Intelligence), and you can listen to the result here:

What makes this special is that, given a conceptual space of musical rules and signatures, the computer can find other possible combinations, within this space, that have not been explored yet. Not only it is possible to create high-quality music, it is also possible to make it sound like a specific composer.

Making transformations

An early example of a successful discovery system the Automated Mathematician, a project from the 1970’s, which could generate and transform small bits of code. This is the embryo for the ultimate computer: the one that can program itself. This kind of system is called artifical development, and it is used in some specific fields with some remarkable results.

This is the least explored of the 3 ways in which AI can be creative, but it definitely has great potential.

What comes next?

One of the biggest bottlenecks in AI creativity is the evaluation of new ideas: after exploring and transforming spaces, how can a computer understand and evaluate its results automatically? How can it know, from all the songs it wrote, which ones to keep? Specially for space-transformation applications, this can be specially tricky, but even more relevant.

Recent advances in AI show that computers are able to produce high-level art, often capable of tricking humans into thinking it was made by another human. Will we ever be able to let computers do this on their own, without our intervention? Not in the near future. Will we ever stop consuming human-made art? Probably not. We can, however, start appreciating both.

The framework presented here is relevant not only for understanding and evaluating new discoveries in the AI field, but for framing new problems and coming up with solutions for them.

“The ultimate vindication of AI-creativity would be a program that generated novel ideas which initially perplexed or even repelled us, but which was able to persuade us that they were indeed valuable. We are a very long way from that.” — Margaret A. Boden

Добавить комментарий

Ваш адрес email не будет опубликован. Обязательные поля помечены *