Creative & Technical Writer, Digital Marketer in Portland, OR
photo-1432821596592-e2c18b78144f_mini.jpg

Blog

Thoughts on PDX tech and cust ops life

 

The Chatbot Uprising

Facebook announced its chatbot API today at its anual F8 developer conference. It's yet another company jumping on the automated messaging technology train, based in artificial intelligence and natural language processing.

I work in technical support - including chat support - and believe it is only a matter of time before I am replaced (at least partially) by chatbots. I image that in 5-10 years, a person in my position would be managing upwards of 10-20 chats at once, acting as a supervisor to specialty A.I. chatbots that successfully answer 80% or more of all customer questions.

Why wait for that inevitability?

Tonight I began training my own chatbot, named SARA: the Software Automated Resource Agent.

SARA is built on Pandorabots, a free, integrated development platform for chatbot programming based on Artificial Intelligence as a Service (AlaaS) API and SDKs.

To supplement SARA's rudimentary social skills (blame her programmer for those), I added an existing opensource personality library from GitHub named Rosie, who is almost too friendly, in my opinion.

SARA is fast, perhaps a little too fast. Many chatbot developers add a bit of time between a human's question and the bot's answer to add a sense of humanity.

SARA is fast, perhaps a little too fast. Many chatbot developers add a bit of time between a human's question and the bot's answer to add a sense of humanity.

In addition to storing customer information (such as a name), SARA currently knows some specifics of the software I support.

The issue of teaching a chatbot is complicated by the fact that it relies on an exact word match to provide specific answers. Because of this, it is necessary to teach SARA that a customer might use "domain" and "url" interchangeably, or call a Grid Gallery Block a "gallery," which should prompt her to ask for clarification.

Since you can teach SARA via programming in XML as well as through interactive chat sessions, I wonder if it would be possible to feed her transcripts of real chats (explaining that she should answer as the agent does) over and over again, to shorten the development time. Ideally, SARA would combine the personality and intelligence of all our company's top chatters. I suppose, then, those folks will be looking for new jobs... and so will I.

For the record: SARA was created on my personal home computer without the aid of internal company documentation. My own knowledge is used to create SARA's understanding of the product. Ahem.

That may sound depressing, but even though chatbots are totally "in" right now, they are still rather clumsy and unnatural in their ability to carry on a conversation.

Most chatbot framework can only retain a few lines of chat as reference before losing all context. Poor SARA can hardly remember what you said one line ago, if it's anything more complicated than a single value (name, age).

And you may recall a recent chatbot failure: Microsoft set its adaptive chatbot AI, Tay, loose on Twitter to learn how humans communicate, and it had to be removed after one day when the bot began spouting racist and offensive tweets.

Though it is a technology that threatens my own job, I am curious to see where it will go. Artificial and automated intelligence has the potential to remove much of the tediousness of common tasks. Heck, even a simple chatbot might reduce chat volume by 5-25% and allow human agents to focus on more complicated problems.

In the meantime, you can chat with SARA by searching for her here! (You must be logged in)