AI Zone Admin Forum Add your forum

NEWS: survey on 3000 US and UK consumers shows it is time for chatbot integration in customer service!read more..

RICH - Followup

This is actually a follow up to Donalds post under Contests. The examples I shared there were fairly good examples of what might be called learning successes, I thought it would be only fair to include some of the learning “fails” and what Ive learned so far

from this approach. Taken from yesterdays log

What’s your name: whats up vince anything good
Vince: Hello , nothing , excellent , Hey , Yes?

(Thats fairly hysterical)

You can see that the concepts were recognized, but the syntax is atrocious. What Ive noticed is that in a controlled environment, where concepts are introduced in the way you would introduce them to a child, ( simple building blocks that increase in complexity), the learning algorithm produces fairly consistent results.  In the “wild”, where people are using complex syntax and concepts, the responses actually sometimes more resemble someone learning a second language, rather than the target “child”. 



  [ # 1 ]

With the “learning a foreign language” concept in mind, I created a module that performs the same function does not attempt to respond to every perceived concept in a sentence, but rather glean major intent. The same exchange using this module looks like

What’s your name: whats up vince anything good
Vince: Nothin’

It will be interesting to see how the different instances of RICH, lab vs online, begin to develop conversational skills



  login or register to react