Creativity And AI

In 1997, Deep Blue of IBM defeated Chess Grand Master Gary Kasparov after a Titanic battle. It actually lost to him the previous year, although he admitted that it possessed "a strange kind of intelligence". To play Kasparov, Deep Blue was pre-programmed with complex software, including an extensive playbook with moves for openings, middle games, and endgame.

Twenty years later, in 2017, Google removed AlphaGo Zero, which, unlike Deep Blue, was completely self-taught. It was given the basic rules of the toughest game of Go ever, to study the game without any sample, and to work out millions of times against himself and work out all his strategies from scratch. This freed it to think in its own way.

These are currently the two main types of AI. Symbolic machines like Deep Blue are programmed as the cause of humans, working through a series of logical steps to solve specific problems. An example is a medical diagnosis system in which a machine cuts a patient's disease from data by working through a tree of possibilities.

Artificial neural networks such as AlphaGo Zero are loosely stimulated by the strands of neurons in the human brain and yet require human input. Their strong point is being realized, which they do by breaking heavy measures of information or rules, for example, the standard of chess or go. They have had notable success in recognizing faces and patterns and also driverless cars. The big problem is that scientists don't know why they work.

This is the reason that the two systems which really point out the difference between them is nothing but art, literature and music. Symbolic machines can do highly interesting work, they are given huge amounts of material and programmed to do so. Artificial neural networks are much more exciting that actually teach themselves and can therefore be called truly creative.

Symbolic AI creates art that is recognizable to the human eye as art, but it is art that is already programmed. There are no surprises. Harold Cohen's Aaron's AARON algorithm produces beautiful images that use templates that are programmed into it. Similarly, Simon Colton at the College of Goldsmith's College at the University of London filled painting to create a sitar likeness in a particular style. However, none of these ever retreat from their program.

Artificial neural networks are far more experimental and unpredictable. The machine works without any human intervention. Alexander Mordvintsev rolling the ball with his Deep Dream and its nightmare images were seen from the Conflict Neural Network (Convonets) and it appears almost spring from the machine's unconsciousness. Then with the machine Ian Goodfellow's GAN (Generic Adversarial Network) serves as a judge for his own creations, and Ahmed Elgamal's CAN (Creative Adversarial Network), a style of art never before seen. They all make more challenging and difficult tasks - machine ideas of art, not ours. Instead of being a tool, the machine participates in manufacturing.

Contrast is even more pronounced in AI-produced music. As an aside, we have François Patches Flow Machines, loaded with software to produce superb original tunes, including a well-reviewed album. On the other hand, researchers at Google use artificial neural networks to use music. But at the moment their music loses momentum after only a minute or two.

The literature created by two types of machines, AI, shows that it is the best of all differences in what can be made. Symbol machines are stacked with programming and rules for its use and designed to produce a particular type of content, for example, Reuters news reports and meteorological forecasts. A symbol machine equipped with a database of plays on words and jokes would trade as usual, giving us, for example, a corpus of machine-made thump jokes. However, similar to craftsmanship, their artistic objects are also according to our estimates.

There is no such restriction in artificial neural networks. Ross Goodwin now trained an artificial neural network on Google, a corpus of scripts from science fiction films, then instructed it to make sequences of words. The result was quite a script for his film Sunspring. With such an absence of limitations, the simulated nervous system would be in normal production function that appears to be blind - or would this "test" be a good idea for us? Such a machine deviates from our past sense of language and can open our hymns to domains regularly assigned as hogwash. Allison Parish of NYU, an arranger of PC poetry, examines the line between Earth and Hogwash. Therefore, simulated nervous systems can initiate human invention. They can introduce us to new ideas and enhance our own innovation.

Proponents of symbolic machines argue that the human brain is also full of software, accumulated from the moment we are born, meaning that symbolic machines can also claim to emulate the structure of the brain. However, symbolic machines are argued from the beginning.

In contrast, proponents of artificial neural networks argue that like children, machines also need to learn. Artificial neural networks learn from the data they have been trained on but are unaware that they can only work with the data that they have.

To say it simply, artificial neural networks are designed to make learning and symbolic machines for a reason, but with the proper software they can each make a bit of the other. For example, an artificial neural network that powers a driverless car must have data for every possible contingency so that when it sees a bright light in front of it, it can recognize whether it is a bright sky or a white Vehicle, so that a fatal accident can be avoided.

What is needed is to develop a machine that incorporates the best features of both symbolic machines and artificial neural networks. Some computer scientists are currently moving in that direction, searching for alternatives that provide more comprehensive and flexible intelligence than neural networks by combining them with the key features of symbolic machines.

At Deep Mind in London, scientists are developing a new type of artificial neural network that can learn to relate to raw input data and present it logically in the form of a decision tree as a symbolic machine. In short, they are trying to build in flexible logic. In a purely symbolic machine all of this has to be programmed by hand, while the hybrid artificial neural network does it by itself.

Merging the two systems in this manner can lead to more intelligent solutions and also to forms of art, literature, and music that are experimental, challenging, uncertain, and fun, as well as more accessible to human audiences.

For more related articles visit to InsideAIML .