Conclusion

• Neurons are leaky integrators
• Synaptic weights change and this is how we learn
• Connectionist models are inspired by this: simple processors and lots of them can do lots of things
• Learning is essential to the process
• There are a range of neural nets
• They do lots of things like associative memory, learn functions, and content addressable memory.
• They are less brittle than symbolic systems.
• They may be a good way of getting real AI, though we're not really close right now.
• Try Igor Aleksander's book An Introduction to Neural Computing. It's a bit old, but it is really solid.