Review of AI projects - Wit

[neilnelson]

Review of AI projects - Wit

The idea here is to get a handle on the capabilities of similar AI projects for inclusion into Melissa. At this point we are not saying that these capability ideas will or will not work for Melissa and only that this is the landscape we can see.

I found Melissa in the comments area of Five Simple Ways to Build Artificial Intelligence in 2016.

Here is the first one.

Wit

I just bypassed the page requesting an email address and went to the Documentation. It looks like you accept Wit's terms-of-service, which we do not want to do, if you provide an email address.

• 'Build your first app' page

  • Wit appears to be web-page, text based.
  • The idea of collecting user responses of particular information to build a user profile that enables tailored responses to questions such as "What is the weather?" is a good one.
  • What is a Wit 'intent', 'entity', 'intent attribute'? Is this Wit language or is there a recognized language for this AI area?
  • A Wit 'story' is a path of interaction between the user and Wit.

• 'Recipes for conversational apps' page

  • Duckling - "Duckling is a Clojure library that parses text into structured data." Duckling has a BSD license that can be used without issue.
  • There is a lot of good information on this page showing how to use Duckling and matching words and sentences on queries, some of which Melissa does.
  • The implications of getting information such as motion picture theater movie times and airplane scheduling is something to think about.

• HTTP API Reference

  • This page provides web-page interface to Wit's server.
  • Speech-recognition capability is shown on this page with a transfer of an audio file to the Wit server. This would not appear to have the immediate response that Melissa can provide.
  • I am getting the sense that the collected user information and some portion of the user assembled application is maintained on Wit's server but could easily be maintained on the web-site provider's server.

• Node.js.

  • Wit javascript and jason at GitHub to be used for web-pages. Uses Wit's restrictive license.

• Python

  • "pywit is the Python SDK for Wit.ai." At GitHub. Uses Wit's restrictive license.

• Ruby

  • "wit-ruby is the Ruby SDK for Wit.ai." At GitHub. Uses Wit's restrictive license.

• Wit conclusion.

  • There are a number of useful ideas that can be obtained from a study of Wit such as the use of Duckling, a web-page interface, server-based services, and structure and design of an AI API with others not listed here.
  • A web-page interface is something I have thought of here for my LAMP computer/server. And this direction suggests that Wit is targeting developers who write web-pages for their sites that will interface to Wit's server. For example, getting airplane tickets, restaurant locations, event schedules and other commercial connectivity has been promoted by Apple's Siri. This then becomes a way to encourage interest to the developer's site with possible revenues from routing customers to businesses, similar to web advertising revenues.

  I am not against this model which may become useful for Melissa but I am seeing these server-based services as part of the many available that would be centered on a user app that can maximize any useful connectivity without capture by any particular outside interest. The web-page is controlled by the web-page provider that in this case is tightly integrated with Wit through Wit's API and restrictive license.

  The request on Melissa for a Python GUI could substitute for and take ideas from Wit's web-page API.

[tanay1337]

I wish to avoid at the use of Wit.ai for brain.py at this point as I am working on a homebrew solution for improving the check_message() function. We may revisit Wit.ai at a later point.