I could never convince my mother to go see a doctor whenever she felt sick. When we were growing up, my mother would rush us to the hospital if any of us sneezed or coughed in sleep. But when tables turned, she treated herself with potions made of herbs and spices.
A few years back, she was diagnosed with a chronic disease beyond the treatment of home remedies. My siblings and I took turns to convince her to seek professional help. As always, she didn’t budge an inch.
Wait, let me put it this way– she didn’t budge an inch on our advice. But, when friends advised her, she agreed to seek help on one condition- she would talk only when someone would listen to her without judgment. It was at that time that I downloaded an app on her phone. The app was an AI chatbot and I had heard great reviews about it. The chatbot helped her overcome fear regarding treatment. It pointed her to success stories (of recovery) about her disease and helped her open up for a dialogue.
It’s been 2 years since this incident. My mother has regained her health and her magical potion-making skills. The reason why this story is close to my heart is because
a. it’s personal.
b. It rekindles my belief that healthcare products, if done right, can provide human-centered care even with AI-powered chatbots.
In the past decade, I’ve seen a surge in the popularity of chatbots in healthcare products. Healthcare organizations who already have a product want to integrate chatbots in their product expansion strategy. Startup founders who are planning to launch think of chatbots as an obvious feature to start with.
Luckily, I’ve worked with both kinds of partners (we refer to our clients as partners). I’ve helped them evangelize and develop healthcare chatbots that help thousands of patients. The reason why I thought of writing down my thoughts is because a lot of organizations are still doubtful if chatbots are just a passing fad or a necessity.
In a series of posts, I’ll talk about the what, why and how of everything related to chatbots in healthcare. I’ll also talk about the design and development challenges so that you can make an informed decision.
What are chatbots?
A chatbot is an advanced computer program that uses Natural Language Processing (NLP) to understand and answer users’ questions. Through voice or text, a chatbot is capable of simulating human-like conversations and sharing prompt information to end-users.
Earlier, chatbots were used to answer standard FAQs and offer customer support. But now, chatbots are omnipresent. From answering a simple question to indulging in decision-making/complex conversations, chatbots are ruling the world, one response at a time.
What’s behind chatbot’s intelligence?
A human mind.
I read a quote somewhere that a chatbot can be only as intelligent as its creator. And this is true. Chatbots are intelligent because we made them so. An intelligent chatbot learns from its past conversations to improve its future performance. Wondering, how?
Let me break it down to you step by step.
Chatbots thrive on user input. In the chat window, when a user enters his/her question, they take it, analyse it by applying NLP (Natural Language Processing), and match the input text with the intent programmed inside its source code. Based on the code, it delivers the best response that answers the user’s question.
Take the below example of a chatbot that allows users to book doctor’s appointments.
Before answering, the bot reads the questions, runs through the model code and realizes that it needs to ask a few related questions in order to answer.
When the human replies with the dates, the chatbot answers with the list of specialists available in San Jose on those dates, along with a link to book the appointment.
But before the chatbot can answer, the user poses another question- “Can you also share details about a dentist in a nearby area available on the same day?”
An intelligent chatbot (even if it can’t answer in the first instance) will learn from this behavior of users and will be better prepared for location-based answers next time.
What are the different kinds of chatbot technology?
The three most common chatbot development methods are–
- Linguistic (rule-based chatbots)
- Machine learning (AI chatbot)
- Conversational AI (hybrid chatbot)
Rule-based or linguistic chatbots are the one that run on, yes you guessed it right, rules. They use the if-then logic to create conversational flows. Before sharing the answer, the bot matches the input text with the pre-programmed responses, and if it finds a match, it displays it to the user; otherwise, share the generic fall back response.
A simple example is– if the user asks ‘’Book me a morning appointment for a dentist’, then the bot’s question could be- “When are you available this week?” or “Share a time slot that works for you.”
The best part of these kinds of bots is that the rules govern the conversation flow. This makes it possible to code multiple correct answers in advance.
On the other hand, this proves to be a limitation as it means the chatbot can’t learn on its own. Everything has to be told to the chatbot; hardcoded into the program. Due to this highly labor-intensive approach, rule-based chatbots aren’t preferred where intelligent conversations are expected. They are slow to develop and the chatbot developer continuously needs to add more conversational flows to the system to increase its robustness.
What are the most common use cases for rule-based chatbots?
The rule-based chatbots work best for products in which the bot is trying to help users with basic FAQs and customer service. A few examples are-
- Patient registrations
- Insurance qualification & verification
- Doctor’s visit (appointment, cancelation, no-shows)
- Finding a healthcare provider (based on location, symptoms, insurance, disease)
- Healthcare counseling, medication and appointment reminders
What open source platforms are available to develop rule-based chatbots?
There are various open-source platforms available to build the Rule-Based Chatbots without actually writing a single line of code like Janis, Praktice, Wit. There is, however, one drawback- the frequency of updating the utterances in the chatbot administrator is very high.
Machine learning chatbots
As the name suggests, these chatbots are powered by Machine Learning. They are the advanced versions of rule-based chatbots and are better than them in terms of interactions with the end-users. They are more conversational and predictive.
ML chatbots are data-driven, i.e. , they learn from patterns, conversations and previous experiences. So the more data we inject, the more sophisticated conversations they can handle in the future. They learn from past conversations and improve the quality of responses by themselves. So far, machine learning chatbots offer the most positive user engagement as they converse with the users just as a human would do. They are the closest to replicating the human-experience of interaction.
ML chatbots also provide a personalized experience to the user. Since the responses are not rule-based, the chatbot understands the questions and creates a feeling of connection, care, and comfort through its responses. Users find it easier to share their feelings without any inhibitions and after understanding their question, the chatbot responds with the most practical solution with a warm, friendly tone of conversation.
However, there is one limitation on the development side. The ML chatbots need a considerable amount of training data to deliver optimal experience. That means, developing an ML chatbot is not easy. It requires NLP (natural language programming) experts to create and manage them.
What are the most common use cases for ML chatbots?
The ML chatbots work best for products where the bot is trying to help users with empathetic and personalized service. A few examples are-
- Symptoms checking & disease diagnosis
- Medical tagging based on the symptoms and escalation of emergency cases
- More sophisticated patient engagement activities like scheduling an appointment with the healthcare provider in case of immediate care
- On provider-side patient admission
- Patient personal and health record retrieval based on past admission
- Patient engagement (pre-post admission survey, after release care plan)
- Counseling services for students or people with specialized needs
What open source platforms are available to develop ML chatbots?
If you’re looking for OS platforms to develop an ML chatbot, IBM Watson, BotPress, BotKit are your best bet.
Conversational AI chatbots- the hybrid approach
After rule-based and ML chatbots, who would have thought there would be another? Well, we needed another one because we want to have the best of both worlds. The hybrid approach uses both linguistic and machine learning models to create a third solution- conversational AI chatbot.
These chatbots can handle complex conversations by using NLG (Natural Language Generation). The best part of conversational AI chatbots is that they have self learning models, which means no frequent training is required. Developers can create algorithmic models coupled with the linguistic conditioning to deliver smart and complex conversational solutions.
What are the most common use cases for Conversational AI chatbots?
A conversational AI chatbot can be used in the more complex systems such as-
- Changing treatment plans based on patient health conditions like adding new medication if the patient is found allergic to one prescribed by the provider.
- Creating a pre-post surgery checklist for patients based on the patient’s health condition.
What open source platforms are available to develop Conversational AI chatbots?
Botpress, RasaX are the two most common OS used for developing conversational chatbots.
Tech stack and frameworks to build chatbots
Healthcare chatbot solutions are the game changers that are helping bridge the gap between patients and providers. They not only enhance the patient experience, but they also transform clinical care and make healthcare more accessible.
In the next blog, I’ll be talking about building rule-based chatbots. If you read till here and liked reading it, stay tuned for the next.