DIGITAL assistants such as Siri and Cortana are increasingly common on phones and computers. Most are designed to give their users the impression that a humanlike intelligence lies behind the program’s friendly voice. It does not, of course. But dozens of experiments over the years have shown that people readily build strong bonds with computerised helpers which are endowed with anthropomorphic features, whether visual or vocal.

Developing an emotional relationship with a piece of software can, however, cut both ways. As a study published in Psychological Science by Park Daeun, of Chungbuk National University in South Korea, and her colleagues, shows, one emotion sometimes involved in machine-human interaction is embarrassment. This, Dr Park has discovered, makes some users reluctant to ask for help from their artificially intelligent pals. Apparently, they are sheepish about doing so.

Dr Park and her team recruited 187 participants into their study. To start with each was presented with a series of statements on the malleability of intelligence. These included, “you have a certain amount of intelligence, and you can’t really do much to change it”, and “you can always substantially change how intelligent you are”. Participants rated their responses to these statements on a six-point scale, on which one meant “strongly disagree” and six meant “strongly agree”. The reason for this initial test was that Dr Park knew from previous work that, in academic settings, those who believe intelligence to be malleable are comfortable asking for assistance while those who believe it to be fixed often feel ashamed to do so.

The initial test done, the researchers presented their volunteers with a second, which involved looking at 16 sets of three words and trying to think of a fourth word that linked them. For example, when offered “room, blood, salts” a correct answer would be “bath”. Sometimes the first three words were accompanied by an unrequested hint (in the example given, this was “tub”). Sometimes they were not.

Hints appeared as the written form of the word in question, accompanied by a computer-shaped icon. For half of participants this icon had a humanlike face, and the hint was placed inside a speech bubble originating from that face, thus anthropomorphising the presentation to some degree. For the other half the icon lacked a face and there was no speech bubble. After the final set of words had been displayed, participants were asked to agree or disagree with follow-up statements about their experience, such as “it was embarrassing to receive help during the task”, and “others might think I am incompetent because I received help during the task”. This time, they quantified their feelings on a seven-point scale, with higher scores representing greater feelings of unease.

The researchers found that participants who believed intelligence to be unchangeable felt more embarrassed and more incompetent after the tests. Specifically, those whose level of belief that this is true was more than one standard deviation above the mean score on the six-point scale for perceptions of intelligence flexibility (in other words, the top sixth of the sample), averaged 3.2 when measured for feelings of shame and embarrassment if the computer icons they had seen giving the hints had had faces and speech bubbles, but only 2.7 if not. In contrast, people who strongly believed that intelligence could be changed over time (the bottom sixth of the sample) felt the same level of discomfort (around 2.5) whether or not the icons had been anthropomorphised.

A second experiment, in which a different set of participants were allowed to ask for help rather than having it thrust upon them at random had similar results. Dr Park therefore concludes that some people do, indeed, seem to wish to avoid losing face by seeking help from an icon that has, well, a face. And that in turn suggests there are circumstances when the relentless pseudo-humanisation of machine-human interactions could usefully be curbed.