Evolution
| Developer: Keiwan | Cost: Free | Play Here: https://keiwan.itch.io/evolution
Who Can Play: Ages 4-99 | Fun Factor: 8/10 | Learning Curve: minimal | Installation: not required | Time Commitment: 10 mins
Evolution is an excellent playground for children and adults to explore how machines can learn in an unsupervised manner. The player simply creates a creature consisting of joints, bones, and muscle. Hit the "Evolve" button and the creature starts experimenting how to most effectively move down a line. As this game applies evolutionary principles to neural network learning, it basically creates (differently configured neural networks) parallel creates and has them all trying to run down the same line. The evolutionary algorithm scores creatures higher that make it further down the line and eliminates the ones that stay too far behind.
Why This Is Interesting
1. See how evolution sometimes comes up with very surprising but effective "designs."
2. Make small or large changes to a creature and observe how this can change its learning process slightly or significantly.
3. Get a feeling for how the selection of the best possible learning models can leave behind promising, but initially unsuccessful "designs."
Evolution is an excellent playground for children and adults to explore how machines can learn in an unsupervised manner. The player simply creates a creature consisting of joints, bones, and muscle. Hit the "Evolve" button and the creature starts experimenting how to most effectively move down a line. As this game applies evolutionary principles to neural network learning, it basically creates (differently configured neural networks) parallel creates and has them all trying to run down the same line. The evolutionary algorithm scores creatures higher that make it further down the line and eliminates the ones that stay too far behind.
Why This Is Interesting
1. See how evolution sometimes comes up with very surprising but effective "designs."
2. Make small or large changes to a creature and observe how this can change its learning process slightly or significantly.
3. Get a feeling for how the selection of the best possible learning models can leave behind promising, but initially unsuccessful "designs."
Beginning (Generation 7): My stickman falls over and moves a lot faster by crawling on all fours.
|
|
Lesson learned: When you look at the video you can see one of the left behind stickmen making good progress learning to walk upright, but is ultimately weeded out for being too far behind the "crawlers." Try it out, it is fascinating.