Skip to main content
Academy | boost.ai
Building the VA
Improving the VA
Expanding the VAs capabilities
Research
All Topics
  • Get started

How humanlike should your chatbot be?

  • By Nena Van As
  • Published: Mar 10, 2023
  • Duration 2m
  • Difficulty Beginner
  • Rating
    Average rating: 0 No reviews

Whenever you read up about chatbots, human likeness is likely to make an appearance. Both in industry and academia, making our chatbots speak and converse like a human has become such a no-brainer that we hardly stop to think whether this is really necessary - let alone if it should be the goal of conversational AI.

There's more to a bot than meets the human like eye

One thing's for sure: there is a lot of research out there that connects human likeness to trust. But there's one thing that's typically missing from those studies: whether the chatbot understood the user or not. An interesting omission, given that bots, like humans, often fail to understand what's asked of them!

We collaborated with Asbjørn Følstad from SINTEF and Effie Law from Durham University to tackle the myth of human likeness, and introduce conversational breakdown to the mix. Let's cut straight to the results, but if you'd like to read more about the study setup and literature, you can find the paper (published at NordiCHI'22!) below.

ACM Other conferences

Effects of Humanlikeness and Conversational Breakdown on Trust in Chatbots for Customer Service | Nordic Human-Computer Interaction Conference

Trust in chatbots can be shaped by various factors such as humanlikeness in terms of visual appearance and conversational content, and conversational performance in terms of the chatbot's ability to avoid conversational breakdown. The literature is inconclusive concerning the effect of humanlikeness and conversational performance on trust, especially their interaction effect.

Focus on repair

So, first things first: when the chatbot does not understand the user (and repeating the question doesn't help), trust drops radically. And, much to research's surprise, it does not matter if the bot is humanlike or not.

However, if the chatbot is able to understand the user after they rephrase or ask again, something interesting happens: trust doesn't drop that much. In fact, trust is about the same as when the chatbot understood the user immediately.

In other words: if you provide repair, trust is the same as if no breakdown occurred at all. That's extremely powerful! In fact, it's so powerful that the impact of human likeness fades in comparison.

Takeaways

So, practically, what does that mean for you?

  1. Build repair into your conversations.Repair is going to help your chatbot save trust, even when it makes mistakes. At the end of the day, your customers come to your chatbot to get help: so make sure that they get it.

  2. Do not spend time on human likeness at the cost of repair."So we should not care about human likeness at all?" Well, no, that's not what we're saying either. But if you have to choose between spending time on designing human likeness or spending it on building repair, please focus on the repair first. Human likeness may be desirable for other reasons - branding, for example - but know how to prioritise it.

Luckily, we have a great course to help you focus on what's important: building conversational repair into your chatbot's conversations. Check it out here or via the link below!

Skills
100
UX design