Artificial biology – neural networks

TechNews Writer
Pronouns
(He/Him)
Mon Sep 20, 2021

Neurons are the specialized cells that makes up our brain. Their role is just to take a particular piece of information (which here is an electrical signal) from a set of neurons and pass on to next set of neurons after making small changes. These changes could be either increasing or decreasing the signal’s intensity. Likewise, billions of neurons work together and make up our brain. From the smallest action of making our heart beat periodically to recognizing someone’s face or remembering a particular memory, it is all done by these neurons together. In the same way, we were able to artificially simulate the process that these neurons do in the language of 0s and 1s.

These systems which are known as “neural networks” work by simulating the process that the neurons do in our brain, which is that first they take the information from a set of neurons in a form of electrical signals which in neural network’s term is a number ranging from -1 to 1. Then, each input is manipulated by first multiplying it with weights, which factors up or down the value. Then, the values that the neuron calculates are summed up and the bias is added. After adding biases, the neuron than normalizes the summed-up value to the value that falls in the range of -1 to 1, and the process repeats for the other neurons.

This is just one neuron in action, but now imagine what 6 billion neurons can do, which is what the model of neural network known as GPT-J model can do, as developed by Ben Wang and in collaboration with Eleuther AI. This model can do anything from writing you an essay of the length that you specify on the topic that you like to writing you a program in any programming language that you say. The only thing that you have to do is provide a detailed text which describes all the things that you want the neural network to do. You can play with this neural network on “6b.eleuther.ai.”  Like, for example, when I entered the following sentence into the model, “ humans and machine,” it gave me the result as “humans and machine? Is there a need for a 'God?' The debate over the nature of the universe is one of the oldest in the history of philosophy. Today, it is being re-opened by the work of modern physics, especially the quantum theory, a theory which is beginning to bring together what was traditionally seen as the domain of physics and what was traditionally the domain of philosophy.” The result is so similar that it is very difficult to understand if this is written by a machine or human and the time is not so far when these huge neural networks will be common.

 

 

 

Appears in
2021 - Fall - Issue 3
Channel