Latest news

Rethinking Politeness in Chatbots: Interesting Research Findings

No comments

Your chatbot should be polite to users. That is a no-brainer. No one wants to communicate with an impolite person, so why should a non-human entity be any different? This is especially the case with bots designed to replicate a customer service experience. But, did you know that users will also display politeness to your bot? Yes — it goes both ways.

Humans will (on average) behave in a polite fashion in their interactions with a computer. This article will look at why this occurs and the evidence we have to support this claim, while providing some practical ideas you may want to implement.

(Credit: Nomura, T of Ryukoku University, Japan. NOTE: What is polite is a cultural phenomenon)

Politeness is defined as the practical application of good manners and etiquette. Note, it includes much more than saying ‘please’ and ‘thank you’. Politeness also contains concepts like ‘socially desirable behaviours’. For example, lying is generally considered rude or undesirable, but if your wife asks you how dinner tastes — well, lying may be polite. So politeness provides a social function — a relationship glue — to keep everything working as it should. Humans are polite to other humans.

But humans are also polite to computers! WHAT THE… This is one of the main findings from a body of research called CASA or ‘Computers as Social Actors’. CASA is one of the coolest theories from human-computer interaction and if you are building chatbots, you need to understand it. In summary, CASA states that humans will subconsciously take the same mental models they use in human to human interaction and apply them in their dealings with computers.

This can result in some weird behavior from your users. In future articles, I will cover things like (i) gender stereotyping of bots and (ii) reciprocity. But let’s examine why CASA says that users will act politely towards computers.

How Do We Know This?

In 1999, some researchers from Stanford University’s Communications Department conducted an experiment. The looked at politeness in human to human interviews and appraisals. We know that humans give ‘socially desirable’ responses to each other. Well they wanted to see if this phenomenon held true in human-computer interaction. Here is a summary of the experiment and the findings.

(Findings from: Nass, C., Moon, Y & Carney, P. 1999. Are people polite to computers? Responses to Computer Based Interviewing Systems. Journal of Applied Social Psychology, 29(5). Pg. 1093–1110)

This is fascinating to me. Surely people know that a computer is an inanimate object. Users can’t hurt its feelings — because it doesn’t have any. What is going on here? In short -> humans use ‘heuristics’ which are like mental shortcuts in decision making. Think rule of thumb, guestimates or educated guesses. We run scripts in our heads to make sense of the word. The computer talks to us like a person, so perhaps I could hurt its feelings like a person.

But Ben, maybe:

– The participants in the experiments were inexperienced computer users. They didn’t know the correct way to behave. NOPE. These were college students with extensive computer experience. And even if that was the case, chatbots are also very new and most consumers haven’t yet figured out the correct way to behave.

– OK, maybe the users were not responding to the computer, but actually responding to the human programmer who built the computer. Maybe they didn’t want to offend the programmer. NOPE. Later experiments on something called ‘Source Credibility’ discounted that explanation.

The participants treated the computer as an independent social actor. And chatbot users will probably be doing the same. Recommendations below…

So What? Design Implications:

This has some interesting practical applications for those of us designing bots:

1./ Don’t get your bot to ask for an evaluation: If you want to know how your bot performs, run a focus group and collect feedback in person. Or provide a link to a third-party survey website and collect the user feedback there. If you have the money and time AB split testing is probably best. If user have to provide negative feedback about a bot directly to that bot, they are probably ‘greasing the wheels’ and being polite!

2./ UNLESS you want to pump up your results: Maybe you have to show some positive results to your client or your boss. Well now you know how 😊 Obviously, as a researcher, I cannot condone this lapse in methodological rigour, but hey — we all have bills to pay.

Start thinking about politeness in broader terms. Scripting your bot to say please and thank you is great, but that is just the ‘ticket to ride’ — the bare minimum you need. But hopefully this article inspires you.

Thank you for reading. I really try to keep my writing grounded in evidence! If this article helped in any way, please hit the clap button below. I will be posting summaries and findings of chatbot related research here, so see you again soon. Let’s connect on LinkedIn!


Rethinking Politeness in Chatbots: Interesting Research Findings was originally published in Chatbots Magazine on Medium, where people are continuing the conversation by highlighting and responding to this story.

Source: Chatbots Magazine

magnoliaRethinking Politeness in Chatbots: Interesting Research Findings

Related Posts

Leave a Reply