Victor Shulist - Aug 7, 2010:

Yes, I’m looking forward to trying out your bot.  Mainly I’m interested in how you provide these ‘flows’.  I’m guessing you are not providing it formal grammar rules but simply letting it see many many examples of which words appear together, and based on statistical analysis, determine the correct parsing?

If you are interested in the flows, it’s best to install the designer, cause the Aici client doesn’t show anything on that. In the project tab, select ‘flows’ and next ‘scanner’ to see how it recognises words, ‘English grammar’ for the other stuff.
The actual code that uses these flow definitions to transform the input is found in ‘code/Flow recognition’. It’s not a big algorithm, but without a doubt, the most difficult I have written to date. It took me 6 months to get it right (and there’s still a threading issue in there, which I have found but not yet fixed). I had a lot of ‘nearly correct’ algorithms, which produced dyslectic or autistic behaviour.
That said, the same algorithm should be able to parse the visual input as well (very cool, I think).

Chuck Bolin - Aug 8, 2010:

Jan,
Your work is very interesting. I read your white paper tonight referenced in your first post.  I got to the “Putting it All Together”...and it didn’t come together for me. =)  Do you have another reference that puts it together in the form of a drawing?  I saw your screen shot of the N²D program and saw the graphics…but it showed only a small diagram.

Welcome to the forums. I hope to follow your project in the coming months.

Regards,
Chuck

Hello Chuck, thanks.
Yes, I have other material as well. The paper is a bit sketchy at times indeed (my first paper + wrote it at the beginning, didn’t yet know it all very well, still don’t )
Reading the blog backwards is probably the best way (as I wrote about ever more detailed stuff). The echo demo is probably interesting, also, designing, debugging, tips and tricks for code + splitpaths. The scanner demo is also worth mentioning as it introduces a bit about the flow recognition algorithm. And I’ve also made the very first entry public again, it sort of explains how I got started with this.

Hello guys,
I’ve just done an update on all the software. So there’s a new Aici version available for those who would like to try it out.
More interestingly (I think), I’ve also (finally) made a web-based version of aici available. It’s in silverlight, so I’m really interested if any of the linux people can get it to work.
One small warning for the brave ones out there: the web-based version is slow. I’m not certain yet why. When testing locally on my machine, it’s also slower compared to the client, so their might be something in the Asp.net code (or somewhere else) I don’t know. What I do know: the shared hosting service (which probably uses some kind of virtualization), does not appear to be the best option for such a resource hog like Aici.
Also: Aici doesn’t yet know many words, relationships or sentence structures, it’s pre-infant stage, so it’s not yet very smart.

Jan Bogaerts - Nov 1, 2010:

It’s in silverlight, so I’m really interested if any of the linux people can get it to work.

I tried installing Moonlight (the Linux version of Silverlight) and it seems to work ok with other Silverlight pages on my Chrome browser, but at the moment when I follow the link that you provided I get a brief loading animation and then a blank screen. Waiting for a while didn’t seem to make any difference either.

Andrew Smith - Nov 1, 2010:

I tried installing Moonlight (the Linux version of Silverlight) and it seems to work ok with other Silverlight pages on my Chrome browser, but at the moment when I follow the link that you provided I get a brief loading animation and then a blank screen. Waiting for a while didn’t seem to make any difference either.

Bugger, it can never be easy, can it. I did the web demo in silverlight cause I saw this cool tutorial on how to do full duplex communication with it, which was exactly what I needed. That way, I could reuse most of my code.  Of course, when I finally get to following the tutorial, it turns out to be for Silverlight 4 (or something), the latest and very fresh release of the platform, which can only be developed using their latest tools. It took me more time installing everything then building the demo. I could have guessed moonlight wouldn’t support this yet.
Back to the asp mess then. That’ll be for the next release maybe.
Anyway, thanks for trying Andrew.

I also tried it. On both Firefox as Chrome, the Silverlight Plugin crashed.

(running Vista; last versions on the mentioned browsers).

Would love to see it working.

Erwin Van Lun - Nov 2, 2010:

I also tried it. On both Firefox as Chrome, the Silverlight Plugin crashed.

(running Vista; last versions on the mentioned browsers).

Would love to see it working.

That should not happen (it works for other people on vista and firefox besides me). Do you also have the latest version of silverlight plugin?

I was running the latest version but apparentely it was corrupt. Now it works. Tx jan.

I’ve tested it, with a few or the suggestions, and it works! But a lot of confusion confusions!!

Why did you build it in a heavy application like SilverLight (which also requires a browser login), instead of a little bit of Ajax?

What’s the next step?

I built it in silverlight because:
1: I’m lazy: I didn’t have to write much, silverlight is very similar to xaml, so I could take the client code and use it for the web with minimal changes.
2: I’m lazy: I don’t really do internet dev (when I can, that is), I’ve never done ajax (I think), php, asp.net (ok, a very very little asp.net since I had to),... And if it’s only for a small demo like this, I don’t really feel an urge to learn something I will probably never use again  
3: They were doing some very cool full duplex communication with it in a demo I saw, using simple ‘events’.  This meant I didn’t have to figure out how to get data back from the server to the web-page: instead I could use a simple event like in the client app.

I guess the next step is to get of my lazy ass and change the web demo to something that’s not so tied into 1 technology. Before that though, there’s still a multi threading bug that needs fixing (I think I’m finally on it’s trial). This is partly to blame for some ‘confusion confusions”.  And there’s also the frame recognition algorithm that needs an update so that it can do complex sentences. I’m hoping that will also fix some ‘confusion confusions’.
In general, the ‘confusion confusions’ is a bit equivalent to a complete black out like with an exam. The network knows something came in, but along the way, something went wrong, and it’s mind went completely blank, so it says ‘confusion’.

thanks man, look forward to the next steps!

Jan, I’m no “web guru” (though you can’t convince my family of that), but I’ve got a little bit of practical experience with PHP, JavaScript, and AJAX, so if you would like some assistance with creating a web interface without using Silverlight, feel free to send me an email, and I’ll try to help out however I can. Since there seems to be no “visual component” to the page (e.g. animations, avatar images, etc.),  it should be relatively simple to use AJAX to achieve similar results to what you have already. And depending on the back-end architecture, you may (or may not) even see a performance increase.

I’ve had a look at the demo, and it seems to work well, although a good deal of patience is required. The average response time seems to be in excess of 30 seconds, and one response took several minutes to receive.

Dave Morton - Nov 3, 2010:

Jan, I’m no “web guru” (though you can’t convince my family of that), but I’ve got a little bit of practical experience with PHP, JavaScript, and AJAX, so if you would like some assistance with creating a web interface without using Silverlight, feel free to send me an email, and I’ll try to help out however I can. Since there seems to be no “visual component” to the page (e.g. animations, avatar images, etc.),  it should be relatively simple to use AJAX to achieve similar results to what you have already.

Thanks, I’ll definitely keep that in mind. Yes, I’m not yet planning to do anything fancy, just a simple page with some info, an inputbox + send button and a response view (the tricky bit).

And depending on the back-end architecture, you may (or may not) even see a performance increase.

That’s interesting. When debugging the web-demo on my local machine, it was also a lot slower compared to the client, so there might be something to gain there. In truth, I simply followed the tutorial, which said: create a WCF service, so that’s what I did.

I’ve had a look at the demo, and it seems to work well, although a good deal of patience is required. The average response time seems to be in excess of 30 seconds, and one response took several minutes to receive.

Thanks. Yep, the web site is currently +30 times slower as the client. .

I figured out why the silverlight client was so slow: it was set to a ridiculously long delay. I’ve also improved the processing speed somewhat and I found the threading bug (which turned out to be unrelated to threading )
So response is a lot faster now. The first input statement is still a bit slower (because of a type of jit that I am using, nothing to be done about that), but after that, it’s very usable at the moment, I think.
There should also be a lot less ‘confusion confusion’. Of course, there’s still nothing in place to learn new sentence structures, so that can still cause the ‘confusion’ thingy,  and the vocabulary is also still very limited.
So I guess the ASP version will still be coming (other platforms is still an issue), but progress will probably be a little slower.

Jan, the web demo shows a remarkable improvement in performance. Well done. I confused it, though, by asking it how many arms four octopus have (an octopus has 8 arms - ok - how many arms does four octopus have). I did the 2 hands/5 fingers per hand test, and that worked well, even switching numbers (2,5, etc.) for words (two, five, etc.), but again confused it when I told it that I lost 2 fingers in an accident. This isn’t a complaint, mind you; Considering that AICI is generating it’s responses, rather than applying a template response to a received input, this is to be expected at this stage of development. I’m actually quite jealous, not to mention impressed, by your progress. Keep up the great work.

Yes, the speed improvement surprised me as well. I had expected some more problems on the shared server. I must say though, that on my desktop, speed has also increased considerably. Still, yesterday’s chat-log has already made it clear that I overstretched it a bit: apparently windows doesn’t like to wait for more then 64 manualResetEvents at a single time, so I had to resort to my own wait objects once again, which has given some more speed improvements I’ll probably update that tomorrow or so.
I must admit that I hadn’t tested the ‘have’ verb handlers fully again after updating a major part of the internals. Aici doesn’t know what octopus’ is, so instead of saying ‘ok’ he should have asked what it was and store what you replied. I’ll have to look into that, cause something like that has already worked. Thanks.
About loosing your fingers, firstly I hope you’ve already found them again cause they’re handy while typing   Secondly: it’s indeed still one step to far at the moment: ‘to loose’ is not known and location info (like ‘in an accident’) is simply not yet processed. What’s worse, the counting algorithm doesn’t handle ‘-’ values correctly yet (I think). That’s also on the to-do list.

Thanks by the way.