(Geek Stuff) AI work continues

From a recent conversation with my AI program (the AI's words are in bold).

    Am I in poor, average, or good health? average
    Do I have a knife? No
    Do I have a gun? Yes
    How many enemies can I see? 1

    I think I will attack!

    Am I in poor, average, or good health? average
    Do I have a knife? no
    Do I have a gun? yes
    How many enemies can I see? 2

    I think I will hide!

What's really curious is that I thought the second example would have him running, not hiding (hiding is considered a more significant threat response than running). Either this means that I don't understand my AI's reasoning or I haven't trained it properly. Either of those scenarios is likely.
  • Current Mood: pleased pleased
  • Current Music: Pigface - Miss Sway Action
Why does the thought of you training your AI sound so hot? Oh wait, I know why.
AI hmm? Interesting...
What are you using to "weight" threats? I'm assuming certainty factors?
Re: AI hmm? Interesting...
I'm using a feed forward neural network with learning handled by backpropagation. The initial weighting of synapses is done at random, with every neuron having an external bias weight added. After the data is fed forward through the network, the results are compared with intended results and "back error propagation" is performed to adjust the weights.

I hope to release a CPAN module based on this. It will probably be called something like AI::NeuralNet::Simple. Initially it has three layers with the number of nodes per layer variable. The main point of the module won't be to replace any current networks, though. It will be the documentation which sets it apart. I want to be something that someone brand-new to AI can pick up and actually understand. Most of what's out there is pretty obtuse and it took me quite a while to get a handle on it.
Re: AI hmm? Interesting...
And just to make it clear that I'm not some super genius who comes up with this stuff out of the blue, the network I am using is loosely based off some code from the book AI Application Programming. It's a wonderful book and I highly recommend it to programmers. The examples are all in C, but they're complete examples -- not hypothetical snippets.

My main contribution is to rewrite the code for general purpose use and slap a Perl front-end on it.
Apparently your AIbot hasn't read anything about Christ or Ghandi. Or maybe he has, and rejects them outright. ;)

As for scenario 2, perhaps you built in a cloaking device by accident!