Anyone who has ever encountered digital customer service probably knows the feeling: the chatbot simply doesn’t understand what you want. You waste your time until your frustration becomes immeasurable. It’s only understandable to insult the virtual assistant. Of course, that doesn’t help, so why are we even arguing with a machine?
They existed before AI, but with the new technology they have become even more widespread: we now encounter virtual assistants on almost every company website and in many hotlines. They are intended to provide quick answers and relieve the burden on customer service.
But that doesn’t always work as smoothly as planned. When a bot gives incorrect or confusing answers, some people react irritably or even aggressively. This is tricky for companies because anger at the bot can have a negative impact on the website or the entire brand.
10 percent of users insult chatbots
A team of researchers led by business IT specialist Professor Alfred Benedikt Brendel from the TU Dresden is investigating the question of why people react angrily to software. A study on this appeared in the Journal of Management Information Systems.
According to Brendel and his colleagues, around ten percent of users behave aggressively towards virtual assistants. This also includes insulting them when there is false information or technical problems.
For their experiments, the researchers compared two chatbot variants. One appears emphatically human. It has a name, gender and profile picture. She also responds in a very friendly manner and underlines her messages with appropriate emojis. The other version, however, remains neutral and forgoes niceties that go beyond the minimum.
In the first round with 175 participants, both variants worked without errors. In this case, the test subjects show neither increased frustration nor aggression. Things only get exciting in the second, crucial experiment with a good 200 people, in which both bots intentionally produce a whole series of errors.
Why friendly chatbots make us angry
The result turns an obvious assumption on its head. One could actually assume that a friendly, human-looking bot would calm people’s minds and be insulted less often. But as the study shows, humanity in AI is a double-edged sword.
The human-programmed bot impacts users both positively and negatively. Indirectly, it increases overall satisfaction with the service, and this satisfaction generally cushions frustration if something does go wrong. But straight away, i.e. the moment the mistake happens, the same humanity increases the anger.
A human-looking counterpart will be much more disappointed if a mistake is made than a neutral machine from which one expects little anyway. Because both effects occur at the same time, humanity can both increase and decrease frustration.
Bots with faces are treated more politely
There is another observation that may be valuable in practice. The degree of humanity of a machine does not only affect the frustration of the user. It also influences how aggression is expressed. Compared to the human-looking bot, angry users are less likely to use truly offensive expressions. If a chatbot has a name and a face, this obviously leads to inhibitions.
The study offers a differentiated result for developers and companies. A human-designed chatbot is not a panacea for angry customers, but on the contrary can actually exacerbate anger over errors. At the same time, however, it counteracts really serious derailments. Alfred Brendel therefore advises companies:
I would recommend that software developers approach human-like design carefully and carefully consider what positive and negative effects additional human-like design elements such as gender, age or certain names can have.
The findings fit into a larger research picture. A study by marketing researcher Cammy Crolic has already come to the conclusion that a human-looking chatbot can reduce general satisfaction, judgment of the company and even purchase intentions among already angry customers. The design of a bot is much more than cosmetic and needs to be carefully considered.
Also interesting:
