When two people talk, each side adapts to the other. Unconsciously, we adapt our choice of words and our syntax to the person we’re talking to and extend our vocabulary when, for example, friends frequently use a certain word. This so-called ’interactive alignment’ is normally a case of give-and-take. In any interaction between humans and machines, however, the conversation remains comparatively one-sided, as linguistic studies have shown. "AI hasn’t progressed as far as the Silicon Valley companies would have us believe," says Dr. Netaya Lotze. For 15 years now, Lotze - a linguist at the Institute of German Studies at the University of Münster - has been systematically analysing how humans communicate with machines.
Her research is based on posts by chatbots, for example from the commentary columns of social media, and they constitute a very promising body of texts. To these can be added data from speech recognition systems such as ’Siri’ or ’Alexa’, which are supposed to make people’s everyday lives easier. Lotze herself won’t have anything to do with such automated data-gatherers. "We have an ’Alexa’ in the office," she concedes, "but it’s only switched on when it is being used for work." For a ’Leonardo di Caprio’ task for example, Lotze’s PhD candidate, Anna Greilich, asks her test persons to do some research and find out facts about the actor - but only using automated speech recognition. To this end, students and volunteers of all ages try out a variety of questioning strategies.
"With our research we want to contribute to AI being discussed in a more neutral and less emotional way," says Lotze. The widespread scepticism expressed in statements such as "machines will be taking over the world soon" is something she doesn’t share. Nor does she believe that in future AI will help to make everything easier, trouble-free and usable intuitively. There is one development that worries her, though. "In our studies with users we observe that an increasing number of people allow themselves to become purely passive receptors after experiencing the same dialogues over and over again and, as a result, they uncritically parrot what the bot tells them."
All in all, it is certainly an advantage, says Netaya Lotze, when language is received in an exchange with the source of its production. This means, she adds, that speaking with computers simply constitutes an additional type of communication, and "most people handle it creatively and confidently." Society needs to agree on common rules, though, to prevent the technologies and the quantities of data collected from being abused for political purposes, for example in order to manipulate election results.
With her working group "Language + AI" - in which humanities scholars from universities in Germany, Austria and Switzerland are involved - Netaya Lotze has already presented her findings at the Massachusetts Institute of Technology (MIT) in the USA.