I’ve often wondered how a chatbot would evolve if a bunch of people worked on it.  I have played with a few models, and I keep coming back to the one problem - to make a good chatbot, you need knowledge… and lots of it…

So with that, I decided to launch a new chatbot.  The bot will be hosted on Github, and open to the public to add and update it’s knowledge.

This is where I would like to ask your help.

The first phase of my project is to define some standards for the coding and rule structure of the bot.  I’ve had some experience with RiveScript, so I have started defining some rules I’d like to follow during the development of Tinman.  This is where I’d like to ask your help.  With the experience you as a community has had in RiveScript development, I’d like to ask for your input in how you think a RiveScript bot should be constructed.  I am sure there will be various point of view, and I welcome that.  In the end, I am looking for a few core guidelines we can follow in adding new rules to the bot.

The next phase would be to add knowledge to the chatbot following the guidelines, and then work as a project team to build one, Turing-test-worthy, Lubner-prize-winning chatbot is capable of beating the best of them, with the goal of being able to hold a conversation, and to be able to pass as an actual human being.

If you’d like to participate, do let me know.  I’m looking forward to working with you, to start the first phase of defining the rules and coding standards.

GitHub - https://github.com/massyn/tinman-rive
Wiki - https://github.com/massyn/tinman-rive/wiki
Chat with him - https://www.massyn.net/files/chatbot/?chatbot=Tinman
(github will be refreshed to the bot every couple of days)

I have long thought that the best path to creating a truly revolutionary chatbot capability is to involve many different developers in its creation and maintenance - sort of a Wikipedia approach. There would be a single core chatbot that provides top level NLP processing, general knowledge and standard ways to handle common concepts such as time, space, math, etc. This core chatbot would be similar in concept to the idea of an upper ontology in the Semantic Web research community. Many additional chatbots, that extend the core chatbot in specific areas of specific domains of interest, would then be developed by the chatbot community at large. The domain specific chatbots would be relatively narrow in scope, such as “medieval Italian painting”, although even that might be too broad to support the level of detail and depth that will be needed to make the system knowledge truly robust. API’s would be published to ensure seamless interaction between the core chatbot and the domain specific chatbots. An incentive development model could be put in place to allow the developers of the domain specific chatbots to publish their chatbots for sale on a common web site, such as is done for Web apps. As more and more domain specific chatbots are developed the overall knowledge of the system will continue to grow and far surpass the capabilities that could be created by any one chatbot developer. This approach still begs the issue of which technical development approach would be best to support the overall system. However, if the API’s are well though out it might be possible to allow the developers of the domain specific chatbots to use whatever basic development approach they desire, within reason.

John Flynn
http://semanitcsimulations.com

A great (and very ambitious) idea! Hopefully Noah Petherbridge will weigh in on this project as he knows the in and outs of RiveScript better than anyone.
He recently addressed my inquiry regarding RiveScript’s ability to retain “learned” information which I think would help greatly with the overall personalization of the bot and make it more believable during conversations.

Of course there are the “giants”, Google, IBM, Microsoft, but none seem involved with the Turing, Loebner and other Chatbot tests so the field might well be wide open!

Here’s hoping!

> freehal.github.io

FreeHAL was something like a “community based chatbot” project, but was discontinued after 6 years.  The interesting thing about FreeHAL@home was that it tried to leverage large-scale distributed computing, like well known SETI@home.

> semanticweb.org/wiki/Project_Halo

>>Vulcan has selected ontoprise GmbH (headquartered in Karlsruhe, Germany) as its prime contractor to develop an extension to the Semantic MediaWiki for Project Halo; the extension is named halo extension and is part of ontoprise’s commercial semantic enterprise wiki distribution named SMW+ (ref. SMW+). SMW+ is serving as a web-based collaboration tool for the user-friendly and low-cost acquisition of scientific knowledge for Project Halo.<<

It seems “Project Halo” is now Allen Institute for Artificial Intelligence “Project Aristo”, and that “ontoprise GmbH” has gone out of business.  However, the point here is that (open) knowledgebase-wikification is another avenue of community participation in chatbot development, not unlike Folksonomy a form of Collective intelligence.

are you planning to extend rivescript code, or just a shared content repo?

CleverBot and that genre use their inputs to add to the content, tho the conversations are closed source. But the interface for adding content (chat!) is much more friendly than github…

A couple more apps in the cleverbot genre include these, which I think are from the same team
http://developer.simsimi.com/
http://rebot.me/dashboard

rebot has a very easy to use dashboard for knowledge management (but no API afai can see)

I guess they’re using the string similarity approach as the learning loop seems very simple. If you repeat the same thing to the bot a few times it will parrot it back.

You can also use tools like wit.ai to provide an interface to the learning of the material, and then link to that from your Rivescript bot. wit has a more sophisticated markup UI to separate intents and other components of an input.

Then you still have the problem of generating responses. WIT etc only help with the input analysis.
You still need to have some nice scripted responses/NLG tools to create the output.

FWIW, I just checked on tinman and the URL is 404
https://www.massyn.net/files/chatbot/?chatbot=Tinman

Probably because the thread is 9 months old?