Solving the trust equation

Socially intelligent computers can turn difficult online negotiations into win−win situations through tactical information disclosure

Programming fundamental ‘social intelligence’ skills into software agents can make humans substantially more trusting of online negotiations, which can lead to superior outcomes in e-commerce transactions, finds an A*STAR-led team of technology researchers, business experts and cognitive scientists.

Automated software agents that bargain for the best deals on the Internet are widely used for business-to-business sales and processes. However, as people are naturally skeptical of negotiations lacking face-to-face contact, engineers are seeking ways to make such software less intimidating.

Yinping Yang from the A*STAR Institute of High Performance Computing explains that it is challenging to create a computerized negotiator with enough social skills to put people at ease. “These agents have to elicit cooperative behavior such as making concessions while maintaining the negotiation goals,” says Yang. “This requires transdisciplinary knowledge of business and social communications as well as careful computational coding of social-psychological rules.”

Yang and her collaborators from academia and industry realized that one way for computers to gain the trust of human negotiators was to proactively share certain information. For example, the software agent could express that its priority is distribution and offer one price for immediate delivery of merchandise and a lower one for delivery in two weeks — a flexibility that signals a willingness to search for mutual benefits.

To test their theory, the team gave 54 MBA students the opportunity to bargain with software designed to simulate the real-world purchasing of laptop computers. They instructed the students to negotiate with an online agent over four key factors — price per unit, quantity, service level and delivery terms — while keeping in mind that their top priority should be obtaining a low unit price.

After an initial round of bargaining, the researchers’ proactive agent offered to find a joint solution by sharing that if the participant ordered in large quantities, it was willing to make concessions in other areas. The agent then invited the student to reciprocate by divulging their priority. As a control, some students negotiated with a non-proactive agent that simply presented counteroffers without offering additional information.

The results were striking: 80 per cent of the negotiations using the proactive agent were successful, whereas only half of the control group had agreeable outcomes. Surprisingly, even students whose personalities tested positive for cynical, ‘Machiavellian’ traits felt more trusting about the online negotiations. According to Yang, this finding suggests that even self-interested individuals can be steered toward cooperative approaches with the right social clues.

The A*STAR-affiliated researchers contributing to this research are from the Institute of High Performance Computing. For more information about the team’s research, please visit the Social and Cognitive Computing Department webpage.


Yang, Y., Falcao, H., Delicado, N. & Ortony, A. Reducing mistrust in agent-human negotiations. IEEE Intelligent Systems 29, 36–43 (2014).