A guide to AI, machine learning and new workflow technologies at HIMSS17: Part 2 : chatbots and workflow

By Charles Webster, MD
07:27 AM
AI machine learning workflow HIMSS17

Conversation! Is there anything more useful than conversation with the right person about the right thing at the right time? Is there anything more entertaining and engaging than conversing with an old and witty friend? No wonder the topic of conversational user interfaces, AKA chatbots, is firing imaginations in health IT. Interacting with a chatbot is like communicating with a human, except the human is replaced by rules and rudimentary artificial intelligence.

What if I told you these conversations are workflows, albeit special kinds of workflow?

Alexa, Amazon’s intelligent speech recognition and generation system for the home, is not a chatbot, yet. However, developing Alexa “skills” is similar to developing simple workflows. Amazon recently announced the $2.5 million Alexa Prize, for creating a chatbot that can converse about popular topics for 20 minutes. Amazon released a preview version of Lex, which uses machine learning (see my previous piece on machine learning and workflow technology), to build conversational chatbots. OhioHealth is using Lex to build a chatbot for patients to book appointments.

[Also: A guide to AI, machine learning and new workflow technologies at HIMSS17 Part 1: Machine learning and workflow]

A recent competitor to Alexa is Google Assistant (GA). I’m exploring its use to create a conversational user interface for Mr. RIMP (@MrRIMP on Twitter), a 3D-printed robot of my own creation. Exciting to me is that GA is based on the linguistic theories of pragmatics and discourse processing I studied in graduate school; I have an “ABD” (All-But-Dissertation) in computational linguistics. Furthermore, “programming” GA is very much like writing workflows for workflow systems! I’ve always understood the important underlying connection between conversation and workflow, and now they are coming together in a big way.

Here is Google’s advice for designing conversations. Chatbot personality should reflect your brand (for example: caring, patient, competent). Don’t try to design conversations word-by-word, but rather at a higher level: start, greeting, elicit request, clarify, repair, satisfy, end, and so on -- a workflow, if you will! Use context: Where are they? What is their mental state (frustrated?)? Assume speech recognition works perfectly (“errors” are your problem, but also opportunities to move conversation forward). Think bigger: give someone access to information they couldn’t access before.

But are conversations really workflows? (Or, “wordflows.” to crack an egregious pun.) A workflow is a series of tasks, consuming resources, achieving goals. Conversation is indeed a series of tasks, achieving goals. I have the goal of finding out the time of my next medical appointment. You have a goal of being cooperative. But I don’t tell you to tell me the time. I ask if you know the time. Your literal answer to my literal question (“Yes.”) is not cooperative. Therefore you need to figure out, and act upon, my actual, but indirectly stated, intent (“Thursday at 1PM). All human language full of this kind of reasoning from indirect evidence toward actual intended goals. And workflow, the steps necessary to achieve a goal, is the key. Representing workflows involves representing goals and steps to achieve those goals. Both workflow systems and human conversation rely on representations of workflow to decide what to do or say next.

The precursor to the chatbot was the despised IVR (Interactive Voice Response) system, allowing humans to interact with computers by telephone keypad. These systems relied on workflow-like representations of call-flow. Today, chatbots are, increasingly, the front end to enterprise workflow systems. These workflow systems interact with databases, transactions, and human customer service workers.

Chatbots are potentially great. But just like human customer service representatives, if the systems they front for, and must interact with, have crappy workflows, the conversations they manage will be unsatisfying. In other words, for healthcare organizations to maximize the benefit of the cool, new artificial intelligence-based communication technologies, they will also need to move toward more sophisticated backend workflow technology.

HIMSS17 exhibitors dealing in chatbot and conversational user interface technology (including frameworks to build them), include Kore (booth 7785), TigerText (1187), Amazon (6969), Orbita Health (7281) and Microsoft (booth 2509, the Microsoft Bot Framework SDK is being used by the Australian Department of Health Services to develop a customer service chatbot). 

HIMSS17 runs from Feb. 19-23, 2017 at the Orange County Convention Center.

This article is part of our ongoing coverage of HIMSS17. Visit Destination HIMSS17 for previews, reporting live from the show floor and after the conference.

Like Healthcare IT News on Facebook and LinkedIn