We are all gamers at heart. On our best days, we do our best to make progress to our individual goals. The paintings found on the walls of caves tell hunting stories where survival depended on winning the battle against less intelligent species. We can thank our distant ancestors for their victories. Natural selection works.
This form of evolution has been applied to creating neural networks that achieve their goals through neural network "breeding." Using Neuro-Evolution of Augmenting Technologies (NEAT) algorithms, a genetic paradigm is used to select AI models that have the best "DNA." The smartest children become the next generation of parents, as low-performing computer programs are discarded. But the success of these genetic models didn't perform well without substantial computing resources.
Before NIVDIA's meteoric stock rise (2019 Update: "and decline") resulting from the widespread use of massively parallel graphical processing units (GPUs), advanced neural network training was out of the reach of many. It was an age of efficiency and clever "shortcuts" that laid the foundation for the amazing applications of AI we are seeing today.
In 2014, a master's candidate at the University of Tennessee published his thesis for overcoming a challenge of combining genetic algorithms with neural networks to accomplish goals . Further work was also presented that year by researchers at Bar-Ilan and Tel Aviv Universities . While this area of research goes back to the 1980s, the processing power using traditional computer programming methods was obviously lacking.
Soon after, Seth Bling employed a genetic algorithm to train a neural network to beat the classic Super Mario World game after only 34 attempts over a 24 hour period. It wasn't told how to play the game. It learned how to play and win all on its own.
The world is full of games. Some say that life itself is winner take all game, made up of a myriad of chaotic games we seek to win. For narrow tasks, forms of automation powered by AI models are taking over many mundane tasks, improving our way of life. It remains to be seen whether AI models will outperform us in broader ways. That's the opportunity that lies right in front of our eyes.
 "Combining Genetic Algorithms and Neural Networks: The Encoding Problem," Philipp Koehn, December 1994, http://homepages.inf.ed.ac.uk/pkoehn/publications/gann94.pdf
 "Genetic Algorithms for Evolving Deep Neural Networks," Eli (Omid) David and Iddo Greental, ACM Genetic and Evolutionary Computation Conference, Vancouver, Canada, July 2014, https://arxiv.org/pdf/1711.07655.pdf
 "MarI/O - Machine Learning for Video Games," Seth Bling, June 2015, https://www.youtube.com/watch?v=qv6UVOQ0F44
Copyright (c) 2018, Jack C Crawford, All rights reserved