Okay, I was testing this earlier, but it seems to be working well now. Let me know if you can't see the image below.



This is a neural network with 8 artificial neurons. There are three input neurons (light blue), four hidden units (green), and one output unit (purple). Positive connection weights are in black, and negative (or inhibitory) weights are in blue.

This network doesn't actually do anything. We haven't evolved functionally-meaningful nets yet. It's just a visual sample, for your enjoyment.

We have two layout views for each network. This one uses a spiral layout algorithm, drawing a node in the middle and spiraling outward.

Here's the same network, in a sequential, layer-based view:



Here's a larger sample network, in spiral layout:



And the same network in layer layout:



Sorry, the end of this one's a little chopped off, but you get the idea. In this last pic, forward connections are black, and recurrent connections (or ones that flow backwards) are in red.

The networks we hope to evolve to play Tic-Tac-Toe should be comparable to these last images...but we're still a ways from there.