Network chatbots to use instead of local Program AB

Hi guys!
The AIML based Program AB seems to be quite old and obsolete technology nowadays.
Training is difficult, and AIML base like ALICE is hard to find in any lang except english.
Building up from scratch is crazy.
Is there plans to add to MRL support of modern AI APIs eg (DialogFlow) or similar?

GroG's picture

Yes, there are already

Yes, there are already services started to begin working with Amazon Rekognition, Lexa, Google Api, certainly it would take minimal effort to get MRL to interface with DialogFlow.

But these are all cloud services with closed source cores.

Program AB is not obsolete technology.  It's still being actively developed, and although it may be more difficult than the other technologies to configure or run, it has one very big advantage.  It's open source.

Our priority is to focus on open source projects rather than interfacing with cloud services. Why ? Because we are given the opportunity to learn and improve, rather than just consume.  

Perhaps there would be interest in developing a UI similar to DialogFlow for ProgramAB or so other AI (like mycroft) .

kwatters's picture

Offline vs Online


  We're happy to look at any open source project you can think of.  It can be integrated.  Reality is, of all the chat bot open source projects that I've found out there, they are all really basically an AIML interpreter.  Where you define a bunch of patterns and their associated responses in config files.  

  If there is truly some new ground breaking technology that you know of,  share a link.

  That being said, we could integrate other chatbots from the web, but I don't think you'll have the ability to customize them as much, and you'll probably find that they're not much better than Alice anyway.

  Just because an algorithm or peice of code is old, doesn't mean it doesn't work :)


Konstantin's picture

Absoluttely true. I should

Absoluttely true.

I should have considered the open source point.

What I rather meant though is a natural restrictions of off-line technology, that became obvious these days.

I was thinking of machine learning setup e.g TenzorFlow (btw it's open) and extended AIML, where neuronetwork would reassign the weights of predefined answers addind more intelligence to conversation. It can't be run on Inmoov locally -  since it's heavy and it needs lots of data to get trained. 

But it's quite a lot of development - and you're right, I could'nt find any open source projects on this. All are plain expert systems, AIML - like.

So.. it was a moment of weakness, that's all)

Tomorrow will start building the AIML base for Russian.  Volunteers are welcome)


kwatters's picture


sorry,  I don't really speak any russian...  but.. i know the InMoov already speaks a little..

Have a look here, the russian version of the AIML has been started.

Tensorflow is very cool.   It's a deeplearning framework.  We have deeplearning support in MyRobotLab with the Deeplearning4j service.  It allows us to create neural networks and train them, and then to use those neural networks to evaluate new data that it hasn't seen before.  Very cool stuff!

I'm not sure how a technology mashup of deeplearning & aiml would look.  There are some very interesting NLP (natural language processing) algorithms that deeplearning frameworks (like tensorflow & deeplearning4j) implement.  Things like sentiment analysis and word2vec are some very cool alrogithms that are implemented by some deeplearning frameworks. 

If you've got some ideas around how this should work, I'm happy to help with the implementation..






Konstantin's picture


Thanks Kevin!

Yes, I'll dig a bit into this and hopefully get back with something. The first task would be to try building a reccurent neural network that would use AIML base (with some additinal tags) and dialogs as a data source for training


BTW - is XMPP client is deprecated? Would be convinient to use for chatbot))

Info page does not help really - any examples or code please? 




kwatters's picture

lstm rnn based replacement for mr t.

so..  very cool..  i have recently learned about lstm based recurrent neural networks and some f their applications in text generation.   i think this is a good area to explore a bit....

GroG's picture

Xmpp client "is not"

Xmpp client "is not" deprecated.

However, it needs some serious "Work & Polishing".

Locally, I've updated to smack 4.2.2 library, but they have drastically changed their authentication & interfaces.

It would be nice to find a stable public xmpp server to use, similar to tcp:// is to Mqtt.

I've run OpenFire xmpp server on server, but we are already over-taxed with traffic of the website itself.

Konstantin's picture

XMPP public server

I use

Old school, free and reliable. is also OK