<img height="1" width="1" style="display:none;" alt="" src="https://dc.ads.linkedin.com/collect/?pid=902658&amp;fmt=gif">
Schedule a Demo

How Chatbots Work: The 4 Step Process to Maintain Human-Supervised AI

Jun 12, 2019 3:05:30 PM / by Jensen Collins

An Inside Look into Chatbot Maintenance with Human Supervised Artificial Intelligence

When a student starts college, they are often faced with academic and social challenges. At times, they may feel stumped. Well, sometimes our chatbots get stumped, too.  And just like any student, our bots are constantly learning!  Here’s a preview of what we do behind the scenes at Ivy. To explain how it works, we use a university setting as a metaphor, where the bot is the student and our maintenance team are the professors!

 

1. Test Their Knowledge 

Every day your bot is "quizzed" by users asking it questions.  Most often, the bot gives great responses, and our team can simply mark the answers as correct. Occasionally, a user asks a question the bot doesn't know; the bot may venture a guess -or- simply get stumped.  In either case, our maintenance team is there to help by reviewing the bot's response and teaching it to not make the same mistake in the future.  

ivyuniv-post (1)

 

2. Correct Missed Questions

When we review the quiz with your bot, any ‘incorrect’ responses are replaced with the appropriate information from your website. We teach the bot how to relay this information conversationally, so the next time a user asks your bot a similar question, it gives a confident, friendly response.

Sometimes, even our maintenance team (the ‘professors’) gets stumped. We may be expert bot builders, but our partners are the experts in their fields.  Questions that stump our team of bot professors will be forwarded to your bot's Inbox Zero, so you can expand your bot’s knowledge with us. By adding a response, you are essentially tutoring your bot, and our ‘professors’ get to learn more about your school.

 

3. Verify Comprehension

Finally, once we’ve taught your chatbots these new concepts, we verify the new responses. This tells your bot (and our algorithm) that the next time this question (or a similar one) is asked, this is the response to send. The more questions your bot gets, the more questions we verify, and the smarter your AI becomes. That’s the magic of machine learning!

ivyuniv-post-2

 

4. Introduce New Material

The following week, we cover new material! Your bot will continue answering your users’ questions. If familiar questions are asked, your bot now knows the answer. If new questions arise, your bot keeps learning.

 

Just like your students, your bot is learning everyday, and the Ivy team is here to guide it!

Jensen Collins

Written by Jensen Collins

Jensen is an AI Data Researcher at Ivy.ai, responsible for chatbot maintenance.

See Ivy in Action