Simple AI Can Help People Work Together

As director the Human Nature Lab at Yale University in New Haven, CT, Nicholas A. Christakis MD, PhD, has spent decades studying hybrid systems of humans and machines. His research has given cause for optimism about AI’s value in group interactions, with some caveats.

“Our work shows that AI can work to help humans to help themselves; but, of course, the reverse is possible, and AI can be used to harm the group,” he said. 

As an example of the former, Dr. Christakis cited one of his studies in which groups of people attempted to figure out the identity of obscure nouns by making repeated guesses. When the researchers introduced simple bots that secretly passed on the guesses of one group to the other, the groups guessed the nouns more quickly.

“Adding these bots programmed with this very simple artificial intelligence, where they just pass ideas from distant parts of the network, helps the groups of people perform better,” Dr. Christakis said.

Dr. Christakis’ experiments with physical systems further support simple AI’s value in human groups. In one study, he had three people work with a robot to perform a task. When the robot spoke in a neutral way and used the passive voice to identify that a mistake was made, the group’s enthusiasm waned. But when the robot spoke in a more vulnerable way and took responsibility for making mistakes, interaction among the people improved, suggesting that the shifts in robot speech have the power to not only affect how people interact with robots but also with each other. 

“This offers the prospect of modifying social interactions via the introduction of artificial agents into hybrid systems of humans and machines,” Dr. Christakis said. “In fact, how radiologists communicate with each other might depend subtly on how the machines communicate with them.”

Dr. Christakis noted that numerous investigators, including Curtis Langlotz, MD, PhD, who introduced him at the session, are looking at how AI will help radiologists in everything from diagnosis and decision support to communication with patients.

“A key aspect of this will be how the introduction of these forms of artificial intelligence affects how doctors and other health care workers interact with each other and their patients,” he said. “Does AI make our health care systems better able to work together, or worse, and in what ways?”

In response, he likened the introduction of AI to human-human interactions to the way in which adding a few aluminum atoms to silicon changes its properties and color. 

“Analogously, many aspects of the performance of human groups will change as artificial intelligence is added into our social networks and our working groups in hospitals and communities around the country,” he said. “But it’s important to realize what will not change: the humans themselves and the intrinsic properties that make them amenable to such exogenous modifications in the first place.”

Access the presentation, “Social Artificial Intelligence,” (T6-PL05) on demand at RSNA.org/MeetingCentral