Initial Ideas

[safia.lambda]

Just to get things started off, what are some ideas/thoughts you've had about the best way to create an artificial general intelligence? How much progress have you made with these ideas? What are some caveats and benefits to your ideas? 

- safia.lambda

[safia.lambda]

I'll start things off. Most of my work involves connecting genetic algorithms with neural networks. The notion is that separate neural networks are created for separate tasks (text recognition, vision processing, etc) and the optimal neural network for each tasks is breed via a genetic algorithm. The evolution process would involve changing the weights of certain nodes, altering error values, etc. The process is complex and would eventually involve connecting the optimal neural networks together. I haven't made much progress (I started with the language neural networks), and it would take an absurdly long amount of time to breed the optimal neural network.  

TL;DR: genetic algorithms + neural networks = MAGIC

- safia.lambda

[safia.lambda]

It seems like your idea involves a lot of human interaction. How are previously generated NNs used?

[safia.lambda]

Check out Texas Tech's resource and the accompanying reference links for more information on neural networks. 

On Thursday, October 3, 2013 5:36:01 PM UTC-5, Meand Bur wrote:

Yes, I thought of it to be a game in the kind of galaxy zoo, where humans help by executing simple tasks. The simple tasks lead to a greater goal. By allowing the users to share and compete with their AIs, they are more likely to invest time in developing larger, fitter networks.

Previously generated NNs can be embedded within new NNs, thus a network can be a cluster of neurons and sub networks. (in the case of a sub network, the NN is handled as a static neuron which is not mutable (only the input and output connections from/to the new NN are mutable.)) The purpose of this would be to allow the users to generate simple building blocks (e.g. Memorizing NNs, timing NNs, broad varieties of pattern matching NNs, anticipating NNs...) and over time achieve more and more complex NNs, showing complex behavior.

I did not yet have the chance to test this subnetwork idea in practice, nor to research if there are any papers on the topic. Do you have links to papers which, you think are a must read on neuronal networks?