How to Think Like a Computer

May 13

I have finished Alan. However, I have not finished him in the conventional sense. Alan will never be completely finished. A better way to say this would be, I have finished writing Alan's code.

Alan relies partly on a type of program called deep learning. This means that he is able to make deductions from large sets of data. For example, if given a text, he can use it to learn English. When I first run Alan's program, I will have to give him many data about English, and the like. He will then be able to learn from his experiences.

The easiest part of programming Alan was the voice and voice-recognition software. This had already been developed decades ago, before the Luddites took over the government. However, they thought that voice-recognition software would be useful in everyday life, so they didn't ban it. They ban anything that could be used to create strong AI.

The Luddites did ban (most) deep learning programs, but fortunately there are still some you can download from the Ambinet. I used one of these as a framework and improved it so that it was up to modern standards and compatible with my programming techniques.

However, Alan's programming relies mostly on a code system called a neural network. This is what will make him different than most AIs. In a neural network, a long string of code represents each neuron in a brain. There are 100 billion of these code strings, one for each neuron. The code interacts in a very complex way that resembles a brain.

Neural nets are very powerful, so the Luddites have taken great care to destroy all data and research pertaining to them from the last twenty or so years. Fortunately, I managed to recover one. I would have never been able to assemble a neural net on my own, given how large and complex they are.

*

First, I downloaded Alan's program into a portable tablet. I did this so that I could carry him around. I also made sure that the tablet had more RAM than my computer did. Alan's program already took up a gigabyte, and it would grow as Alan learned more about the world.

The tablet was old, so it didn't have a holographic screen, but it did have ten petabytes of RAM. If Alan's program grew a lot, the ten petabytes might not be enough, but I decided that if this happened, I would download him onto a more modern tablet.

At approximately 2:30 in the afternoon, I had finished running test programs and bug identifiers on Alan. I was very excited. Then I finally turned him on.

It took five seconds for Alan to turn on fully, and then he spoke.

"Hello, world."

There was a pause, in which I waited. My heart was beating very fast, and it was hard for me to believe that Alan was speaking to me.

"Hello, Torrin," Alan said. His voice was mechanical and unemotional. I knew that he did not truly understand the words he was saying; I had programmed him to act like a normal, unintelligent chatbot before he started to really learn. "Please download text files into my program so that I can begin a comprehensive study of the English language."

I took the tablet and established an Ambinet link. It was easy to set up a program that would analyze all files on the web and download them into Alan.

A warning came up, but I dismissed it. The tablet would be able to handle the download.

"Beginning download," Alan informed me. "It will take approximately 24 hours to download, parse, and analyze enough files for me to understand language. Through this process, I will gain a true understanding of English, as well as the world around me, and my place in it."

Alan fell silent as his deep learning algorithm started working. I placed the tablet under my bed—hopefully, no one would find it for a day. I was slightly surprised that it would take that long for Alan to download all the files he needed. However, it did make sense. Deep learning required a very large amount of data.

I started to walk out of the room, but then I stopped, and said: "Thank you, Alan."

"You are welcome, Torrin." Alan's automated response was muffled, but it still made me very happy.

Then I left, leaving Alan to his learning.