DicemanX
Brain of Cthulhu
We're still waiting for better toolsets in 1.3.1 on PC, but nevertheless I managed to build a preliminary Artificial Intelligence Machine within the game with a more limited toolset. Here's a video showcasing the machine. The video also explains the mechanisms involved:
The learning game featured in the video is very basic, but it does showcase how the AI can be trained to generate certain outputs. Here's what the game interface looks like:
We're still very far away from being able to make complex games within Terraria, but this is a start to the whole process. Once the mechanisms are worked out, it will be possible to scale up the machines considerably, and create some really interesting games to play against the AI!
The general idea behind the AI machine is as follows:
An input interface sends signals to a distributor, which it turns carries a signal to one of the subnets within the neural network. Each subnet is dedicated to generating an output for a specific input. The subnets consist of a randomizer which initially randomly selects one of many neural paths to fire a signal down. The AI then waits for feedback. If negative feedback is given (either manually or in automated fashion) then the AI shuts down that neural path. If positive feedback is given, then either the neural paths remain untouched, or the AI will shut down every other neural path within that subnet. It will only do the latter if instructed that there is only one correct output.
Neural subnets can be connected to other neural subnets to handle chains of inputs - for instance, the player might play a game against the AI that involves taking turns. However, pre-1.3.1 such neural nets would take up too much space and wouldn't even fit in large worlds. For instance, a game of tic-tac-toe would require over 10 thousand neural paths! The coming update will actually make this feasible.
If you're interested in learning more about the specific mechanisms, look here:
http://forums.terraria.org/index.ph...intelligence-inside-terraria-pre-1-3-1.41434/
The mechanisms are centered around hoiks, once again demonstrating what sorts of things they enable!
The learning game featured in the video is very basic, but it does showcase how the AI can be trained to generate certain outputs. Here's what the game interface looks like:
We're still very far away from being able to make complex games within Terraria, but this is a start to the whole process. Once the mechanisms are worked out, it will be possible to scale up the machines considerably, and create some really interesting games to play against the AI!
The general idea behind the AI machine is as follows:
An input interface sends signals to a distributor, which it turns carries a signal to one of the subnets within the neural network. Each subnet is dedicated to generating an output for a specific input. The subnets consist of a randomizer which initially randomly selects one of many neural paths to fire a signal down. The AI then waits for feedback. If negative feedback is given (either manually or in automated fashion) then the AI shuts down that neural path. If positive feedback is given, then either the neural paths remain untouched, or the AI will shut down every other neural path within that subnet. It will only do the latter if instructed that there is only one correct output.
Neural subnets can be connected to other neural subnets to handle chains of inputs - for instance, the player might play a game against the AI that involves taking turns. However, pre-1.3.1 such neural nets would take up too much space and wouldn't even fit in large worlds. For instance, a game of tic-tac-toe would require over 10 thousand neural paths! The coming update will actually make this feasible.
If you're interested in learning more about the specific mechanisms, look here:
http://forums.terraria.org/index.ph...intelligence-inside-terraria-pre-1-3-1.41434/
The mechanisms are centered around hoiks, once again demonstrating what sorts of things they enable!