Robotic helpers leave humans switched off

Anyone who's participated in a group project will know that teamwork doesn't always make us more efficient. In fact, can actually lead to so-called social loafing, where our motivation to be productive decreases and we start to rely on others to carry us. Taking our foot off the pedal like this is particularly likely to happen when we don't see the task as having much value, or when a co-worker is performing particularly well. 

Much of the research into social loafing has (understandably) looked at the phenomenon in relation to humans. But our modern era begs the question: do we do the same when working with robots? 

Writing in Frontiers in Robotics and AI, Dietlind Helene Cymek and colleagues from the Technical University of Berlin aimed to explore this mystery. For their study, the team showed 42 German students four images of a circuit board at a time, each of which either contained one, two, or no defects such as cracks, scratches, or soldering faults. Their task was to compare these images to a perfectly in-fact board, and find these imperfections. At first, the images of the boards were blurred, and participants had to reveal parts of it by moving a 'sharpening tool', controlled with their mouse, over the images. 

Some participants were in the team condition, working together with a robot that checked the images and left red marks near potential defects. Others were in the lone condition, and did not see marks left by a robot teammate. After selecting the defects, those in the team condition were asked to 'double check' their work, while those in the single condition were asked to engage in 'quality control'. 

The team was interested in numerous things: how much of the circuit board participants uncovered with the sharpening tool, how long they spent searching for defects, and how successful they were at detecting the defects. Once the task was complete, participants reported how much effort they had put in, how good a job they felt they had done, and how much responsibility they felt they had taken for the task. 

When it came to how much area was uncovered, participants working in a team with the robot checked a slightly smaller proportion of the images than those working alone, but there was not a significant difference. There was also no significant difference in the time participants spent searching for defects. 

However, when it came to successfully detecting defects, differences between the conditions were much more clear. In the lone condition, participants detected around 4 out of 5 defects on average, compared to around 3 out of 5 in the team condition, suggesting that participants did engage in social loafing when paired with a robot teammate. Even so, participants across all conditions strongly agreed that they had put a lot of effort into the search, had done a good job, and felt responsible for their task. 

So, overall, while participants did engage with their task with a reasonably similar amount of effort, performance was much lower. This suggests that participants who worked with robots "looked for defects less attentively than those who worked alone," despite believing they had taken responsibility for their task. 

The team suggests that this could be explained by so-called "inattentive processing", in which they "looked at information, but did not really process it consciously." This has also been seen in studies looking at pilots scanning a display, who nevertheless failed to notice errors. 

Because this experiment was a one-off, rather than a long-term partnership between human and robot, participants would not have felt a close relationship with their synthetic partners. This may have influenced how much participants felt they were working as a team and affected the amount social loafing they engaged in. Other research on loafing has also found that it occurs more when people find tasks of low value – so looking at different kinds of tasks and how people engage with them and their robotic teammates over time may prove informative. 

Read the article in full: https://doi.org/10.3389/frobt.2023.1249252