Page 1 of 1

Game idea: PBEM race neural networks

Posted: Sat Aug 24, 2024 12:47 pm
by togFox
Think of a race track and autonomous robots have to race around a track or run to the end of a maze but the robots have real neural networks and the NN really learn. If there are like 8 robots on the track then there are 8 real NN's learning while on the track.

It's a game where you literally watch robots win or lose and the NN's grow or not. Smart NN's will let those robots win.

The player is a spectator and can buy robots/NN and then watch them race and win. They can trade and sell and offload dumb NN. They can bet on robots they think are pretty smart and will win the race. It will be meta-data heavy. The player has to comb through all the robots and find the diamond in the rough, buy it, race it, train it and let it learn and when it wins a race then there is finally reward. End up buying a lemon then do you choose to train it more or cut your losses and buy something else. Maybe for a fee the player can merge two NN'S together which effectively doubles its knowledge.

NN's/ robots start work no knowledge and won't be fun to watch but analysing the NN meta after each race shows its rate of growth and competitiveness worth other robots.

Play-by-email means all players are offline and place their orders and requests and then exchanges files with the host. The host then uses the program to "crunch" everyone's orders. Race replay's and refreshed meta-data is then emailed back to players where they can then review the replays and decide on next moves. PBEM games can go as fast or as slow as the host likes.

Can you pick a NN that is rough on the outside but has hidden potential? Can you read the meta and the stats and see how your real NN is growing using real machine-learning techniques? Will you be able to recognise when your robot has reached it's limit and it's time to offload it?

Play with your friends and grow NN's together.

=====

That's the pitch. Half educational, half fun and totally nerdy. Races is where each robot performs well or not but every race is a genuine self- teaching event. I'll need to think of things for the player to do between races so the player can actually influence robots and race outcomes. Maybe allow NNs to be tweaked between races so players can learn about NN parameters and learn how to tweak NNs.

Thoughts?