Hi,
I just published a new post on my blog about the new direction that the project is heading to.
Looks pretty cool he?

Awesome Jan!

I am very interested in reading all about this ...

Thank you!

Looks interesting! I’m going to have to wait to play the videos—but I’m excited to take a look.

thanks all.

Jan,

Please tell me more about how you are using neurons.

I’ve done this type of job before.  It’s very interesting.

Let me start by saying it’s not a traditional 3 layered neural network as you would have learned from books or school. This is a free form resonating network. What this means is that there are no restrictions in links (in fact the {} operator is implemented with loopbacks) + there are no ‘weight’ functions + there are different types of neurons + links have a direction + there a also parent/child relationships.
Basically, whenever you declare a pattern like ‘My name is $name’, it is transformed into a neural path (a bit like how you ‘compile’ a regular expression into it’s tree structure). When input is sent to the network, it tries to find a path that leads to an endpoint (walking the neural network: find all outgoing links, match with filters and stuff, split,...)

I’ve made a new screencast of the second character, looks pretty cool he.
I’ve also written a new post with some more details of the <a href=“http://www.chatbots.org/chatbot” class=“term”>chatbot designer</a>.

I’ve made 2 new videos. the first shows an initial AI feature: complete the sequence, which is all done using a single rule and some thesaurus look-ups, more info on my homepage. (probably best viewed in full screen, hd, so you can see the letters).
http://www.youtube.com/watch?v=fG9MfdKOhes&hd=1&fs=1
The second is a small debug session that shows how the network performs this little trick (neural code). this video is narrated (my terrible voice).
http://www.youtube.com/watch?v=JKx3JWLj3S4&hd=1&fs=1

Very interesting presentation Jan, thanks for posting the videos.

Pretty cool he. The complete-sequence function is currently limited to sequences of the form f=na2 + nb + c, but I can add others as well.

If I understood your posts and the videos correctly, you wrote the software for the user interface that is featured in the videos too didn’t you? Kudos to you if you did. I haven’t seen a graphical interface quite like that before. How did you implement it?

Andrew Smith - Aug 22, 2011:

If I understood your posts and the videos correctly, you wrote the software for the user interface that is featured in the videos too didn’t you? Kudos to you if you did. I haven’t seen a graphical interface quite like that before. How did you implement it?

Thanks, Yes, I did make it all myself. That’s the original neural network designer application. I also designed the language and the database that it runs on myself. It was lots of work, but it puts a huge smile on my face whenever I work with it. The ‘split’ instruction that is used in the second demo, is something you simply can’t do in a normal programming environment. Also, the multi threaded debugging part, works much better compared to Visual studio for instance. 
The interface is written in WPF. The code editor which is shown in the demo, is a set of controls that I designed myself (well, 2 or 3 times).

I’ve made a final pre-release video on how to call .net functions from within the chatbot (how to extend it)
http://www.youtube.com/watch?v=Hp51R0xJFL4
Also check out my homepage.

Well, looks like I finally got to a first beta release. There are still some issues here and there, but I think most is starting to work and I managed to include an example for many features. The current release only contains the basic and the pro versions. The designer (which allows for full debugging of the network and things like extending, replacing the pattern matching algorithm).
I’ve also included the start of a general purpose, reusable pattern library (included in a demo project), which I’m hoping is useful as the basis for different types of bots, including ‘watson-like’ jeopardy bots.

From a pattern matcher point-of-view, I think there are a couple of features that make it interesting, and which definitely need further exploration.
I’m thinking of things like the thesaurus variables, sub-topics (which enables a technique that I’d like to call ‘topic-inheritance’) and context (in the chatbot properties). The ‘all’ demo (which contains this common lib, or the start at least) contains many implementations of these features.
There’s more info on my homepage.
Documentation and stuff will come, starting from next week.

I’ve put up an article on short/mid and long term memory usage in my chatbot designer. Could be an interesting (and eye-opening) read for some.

There’s a new article available about how to implement why, because, have, like and inverting the person. Starts out simple, ends a tat more advanced.