ChatBots for Digital Interventions

Thinking about using a chatbot in your next intervention? If you've never integrated a chatbot into an intervention before - or if you have but it didn't work out as well as you'd hoped - here are my recommendations.

First, Think Small

Chatbots often do a great job at appearing as if they can do lots of magical things. Maybe it's the suffix "bot" that makes us believe that they are somehow have special powers. As a result, some researchers have unrealistic expectations of what a chatbot can do.

Chatbots are usually not a good example of artificial intelligence or machine learning. They are usually just a friendlier way to gather user input.

Chatbots can be useful for asking for things such as:

• True/false questions
• Short multiple-choice questions
• Likert scales
• Mood assessments

The key here is that the possible human responses are restricted to a predictable, finite set of answers.

Sometimes, free-form text can be allowed, but only when gathering data for later analysis. Expecting a typical chatbot to understand and intelligently reply to arbitrary sentences is a tall order.

That being said...

Then, Think Big (Maybe)

When most of us think of chatbots, we imagine something more conversational. Most commerical chatbot platforms are geared at business solutions, such as customer service, technical support, or sales automation. The goal of these platforms is to offload the most commonly asked questions away from human personnel, freeing them up to focus on the more difficult situations.

To make these advanced platforms work in a research context would require a significant cost in terms of both labor and money. Chatbots only understand natural language input after a significant amount of "training" to understand incoming vocabulary and grammar. And while a few platforms are emerging as low-cost options, such as Google's DialogFlow, they still require integration into a custom software solution. Those integration costs must be borne by your devshop or (more likely) passed on to you.

Recommendations

Before you decide on a chatbot for your intervention, try to answer these questions for yourself:

1. Will my participants really find a chatbot easier to use than a simple survey or questionnaire?
2. What kinds of responses will the chatbot need to generate?
3. Can the chatbot present multiple-choice options, or am I depending on a more complex natural-language processor?
4. What are the costs of various chatbot platforms, especially over the extended length of the study?

Chatbots are finding their niche in more and more places, and undoubtedly can be helpful in many research areas. Before you invest any time or money in a chatbot-based intervention, be sure to think through the details of how to get the best outcome.