AI coding ideas

AI coding ideas

Posted On: November 24, 2009
Posted In:
Comments: No Responses
Evolution, genetic programming, neural networks and typography
This project was my first encounter with artificial intelligence. Feeling is really strange after doing some AI. We experience a machine that is capable of learning and resolving hard non-linear problems. For now there is no conciseness nor emotion, just “intelligence”. Strange feeling is when “intelligence” happens, that effect troubles our emotions and our perception of machine as an complex automat. “Look! It’s just like us!!” Machine intelligence can be programmed (general concepts are old more than a half of century) but a real “magic” cannot exists without emotions and conciseness. So is it intelligent or not? Yes, but It doesn’t feel anything, yet.
As human intelligence, artificial intelligence is capable to give each time different solution for given problem, or a way to resolve it. Design is approximative and iterative process of good and less good solutions in creating a product. I tried to simulate a simple case of processus of creation in computer, completely human indipendent. I asked computer to create a new typography…
First, I simulated one neural network inside of my computer. Then I presented a lot of different letters in different typographies (One of the first exercises for AI newbees is to make one simple OCR program – optical character recognition). Neurons adapted their “weights”, and after some training they were pretty sure and also correct about presented letters. They learned to make difference between letters and letters and some other forms. So at this stade computer has an “idea” of what is a form of one letter.
Than i wrote simple generator of noise (random values for pixels). I asked neural network “Is this sample is similar to letter A and which one is more likely letter A?” After that I wrote genetic algorithm (direct analogy of the Darwin’s theory of evolution) to make a new populations from the best samples in one population. Neurons done a natural selection between random noises. After some time noise became letter A.
This example was done using backpropagation algo (learning – Hebb rule) and genetic algo. All was programmed in Processing

Evolution, genetic programming, neural networks and typography

This project was my first encounter with artificial intelligence. Feeling is really strange after doing some AI. We experience a machine that is capable of learning and resolving hard non-linear problems. For now there is no conciseness nor emotion, just “intelligence”. Strange feeling is when “intelligence” happens, that effect troubles our emotions and our perception of machine as an complex automat. “Look! It’s just like us!!” Machine intelligence can be programmed (general concepts are old more than a half of century) but a real “magic” cannot exists without emotions and conciseness. So is it intelligent or not? Yes, but It doesn’t feel anything, yet.

As human intelligence, artificial intelligence is capable to give each time different solution for given problem, or a way to resolve it. Design is approximative and iterative process of good and less good solutions in creating a product. I tried to simulate a simple case of processus of creation in computer, completely human indipendent. I asked computer to create a new typography…

First, I simulated one neural network inside of my computer. Then I presented a lot of different letters in different typographies (One of the first exercises for AI newbees is to make one simple OCR program – optical character recognition). Neurons adapted their “weights”, and after some training they were pretty sure and also correct about presented letters. They learned to make difference between letters and letters and some other forms. So at this stade computer has an “idea” of what is a form of one letter.

Than i wrote simple generator of noise (random values for pixels). I asked neural network “Is this sample is similar to letter A and which one is more likely letter A?” After that I wrote genetic algorithm (direct analogy of the Darwin’s theory of evolution) to make a new populations from the best samples in one population. Neurons done a natural selection between random noises. After some time noise became letter A.

This example was done using backpropagation algo (learning – Hebb rule) and genetic algo. All was programmed in Processing