Original article was published on Artificial Intelligence on Medium
Conway’s Game of Life, Intelligence, and Neural Network
John Horton Conway, the inventor of Game of Life, has died at the age of 82 due to Covid-19 due to which half of the world is still in lockdown. I got this news because I wanted to add tiny neural networks into cells of Game of Life when I was showing his work as a reference to my team a few days back. I guess now I can just dedicate my work to him instead of taking his feedback.
What Game of Life show us is that through some basic elemental rules one can build a complex system with limitless possibilities. The procedure describes a simplest of the environments which is a grid of white[dead] and black[live] cells and just the following four statements as rules for the cells:
- Any live cell with fewer than two live neighbors dies, as if by under-population.
- Any live cell with two or three live neighbors lives on to the next generation.
- Any live cell with more than three live neighbors dies, as if by overpopulation.
- Any dead cell with exactly three live neighbors becomes a live cell, as if by reproduction.
What is fascinating is that the Turing Machine can be built into the Game of Life. But for this to happen naturally from a starting state rather than through a guided process might never happen in a sequence that runs with current resources and with our entire life as runtime. If we do an experiment and see every single living being as a neural network, then taking every single pulse in these networks as our clock, the Game of Life is running since the formation of the first cell from an amino acid in the oceans. If we compare Intelligence in humans with Turing Machine in Game of Life then nature took a smart move using survival of the fittest to drive towards finding the shortest path to Intelligence even though it looks so long. We call this the path of evolution.
What would be required if we attempt to build Intelligence in a Petri dish in a lab, or a supercomputer in some server room? We will require sufficient time, a model for the processor for intelligence inspired by the human brain but better, and a path of evolution. We do have the processor with quantum computing available on AWS and we do have a model with the Artificial Neural Networks. What is missing is the last piece, the path of evolution. We are not able to pinpoint a strategy that is closer to it let alone the one that is better than it. What can be done would be to build a system as simple as possible, give the system a much better start than nature and to let it run before the rules of survival are determined. That is to ignore the strategy and start working with the tools. One noteworthy aspect here is that progress is already made in terms of motor function in this area.
Suppose someone takes up this task and kickstart the system with just the processor and the model. What they will require is constant observation of the system and its development and with multiple eyes. They will require the help of people who will be able to understand the system and who will know enough to be able to play God with the system and modify it as required by changing the rules of the game. What will also be required is reiterations of the simulations multiple times before the final rules of the survival are identified, discussed, and finalized. The system for the intelligence to grow is a solution rich space, but it also has very sparse density. To extract intelligence out of this solution space by letting it grow naturally and without any guidance would be impossible. Even the guided system can potentially take generations.
I would like to point out to the people who are working in this field that they should take inspiration from the Turing Machine that was built in Conway’s Game of Life, and consider Intelligence as an optimized solution after billion of years of evolution. Intelligence should be seen as a result of evolution as it helps one understand what makes it so difficult to find a quick path to developing intelligence in lab with highest levels of expectations.