77 lines
2.4 KiB
Markdown
77 lines
2.4 KiB
Markdown
|
# CX
|
||
|
|
|
||
|
|
This project aims to develop general artificial intelligence.
|
||
|
|
|
||
|
|
This will require a few subprojects for the AI to be able to do
|
||
|
|
taks in.
|
||
|
|
|
||
|
|
## 1. 3D engine
|
||
|
|
|
||
|
|
To be able to teach the AI how to move in 3D space,
|
||
|
|
it needs a controllable character and some interactive
|
||
|
|
environment to move through.
|
||
|
|
|
||
|
|
This engine, being able to simulate newtonian physics,
|
||
|
|
will feed the AI character with visuals, sounds, limb
|
||
|
|
stress feedback, balance information (where is down?)
|
||
|
|
etc.
|
||
|
|
|
||
|
|
## 2. Chess minigame
|
||
|
|
|
||
|
|
We want this AI to be friendly and chill.
|
||
|
|
A sense for game time is essential in this regard.
|
||
|
|
Once fully develop, it will be able to choose what
|
||
|
|
it wants to play on its own. I think this might be
|
||
|
|
a good starting point.
|
||
|
|
|
||
|
|
## 3. Message prompt
|
||
|
|
|
||
|
|
It shouldn't be too hard for this AI to learn basic
|
||
|
|
use of words and associate (I lost the game) these
|
||
|
|
words with certain objects. It should be possible
|
||
|
|
to train the model to grab a pen just using a message
|
||
|
|
prompt and correct feedback from the environment.
|
||
|
|
|
||
|
|
## 4. Neural network framework
|
||
|
|
|
||
|
|
This framework will contain 3 types of neural networks:
|
||
|
|
|
||
|
|
### 4.1. Input neural network
|
||
|
|
|
||
|
|
The role of this network is to gather information from senses,
|
||
|
|
preprocess the information and feed it inward.
|
||
|
|
|
||
|
|
### 4.2. Output neural network
|
||
|
|
|
||
|
|
The role of this network is to gather output from a previous
|
||
|
|
network, postprocess the information and feed it outward.
|
||
|
|
|
||
|
|
### 4.3. Thinking network
|
||
|
|
|
||
|
|
This neural network will be set to a certain tickrate on which
|
||
|
|
it ticks (units to hundreds of ticks per second).
|
||
|
|
|
||
|
|
This will emulate real thinking using backpropagation if information
|
||
|
|
inside the neural network using backward connections.
|
||
|
|
|
||
|
|
Such an approach should allow the neural network to create memory banks on its
|
||
|
|
own and remember crucial data, this would allow context understanding for as
|
||
|
|
long as the neural network deems the information necessary. Similarly to how
|
||
|
|
human brain remembers information, the neural network would create loops of
|
||
|
|
neurons, which when deactivated, would stop holding on to their current state
|
||
|
|
and free those neurons to participate in another task.
|
||
|
|
|
||
|
|
## 5. Bindings to external tools
|
||
|
|
|
||
|
|
If experiments bring promising results, this AI will
|
||
|
|
sure appreciate being able to communicate with more than
|
||
|
|
just its virtual world.
|
||
|
|
|
||
|
|
Allowing the AI to manipulate with computer tools and/or even
|
||
|
|
talk to other people over messaging systems would speed up any
|
||
|
|
further development.
|
||
|
|
|
||
|
|
From learning how to code, to playing videogames, to doing much more.
|
||
|
|
|
||
|
|
|