By: Rawpixel
The date was January 20, 1961. President John F. Kennedy, during his inaugural speech, said, “let us never negotiate out of fear. But let us never fear to negotiate.”
At the time, JFK was speaking about the need for a balanced approach to negotiations between the U.S. and the Soviet Union – one that is neither overly aggressive nor timid.
As the former president alluded to in this now-famous quote, human emotions play a critical role in negotiations. It’s a key component that USC researchers are delving deeper into by gaining a better understanding of how AI agents that are aware of emotions can successfully negotiate.
Thanks to a study by Kushal Chawla, a Ph.D. computer science student at the USC Viterbi School of Engineering and researcher at the USC Institute for Creative Technologies (ICT), we can now add “negotiate with emotional intelligence” to the growing list of things AI will soon be capable of.
Chawla’s research was recently accepted in IEEE’s Transactions on Affective Computing as part of a special issue on the “Best of ACII.”
“This paper essentially establishes the value of capturing emotional attributes when teaching AI how to successfully negotiate,” Chawla explained. “A natural next step would be to incorporate these variables in a dialog system, which we’re working on now.”
If you’ve ever had to negotiate a salary or make a deal when buying a new car, you may have first-hand knowledge of how negotiation can be a complex social interaction that encapsulates emotional encounters in human decision-making.
Motivational tensions often arise when negotiations pit aspirations for individual accomplishments against the demands of sustaining social connections. This leaves difficult decisions for the negotiators about working towards their own self-oriented outcomes or making sacrifices for others. Therefore, predicting the partner’s satisfaction and perception (also known as subjective outcomes) in advance can be crucial for an AI assistant that aims to negotiate with its users.
“This paper specifically focuses on the prediction of those subjective outcomes in a negotiation,” said Chawla. “We use machine learning models to capture the emotion from different dialogues. Once we have those emotional variables, we try to predict the subjective outcomes of the conversations. We showed that by using those emotional attributes seen in conversations, you can better predict the outcomes in the negotiation versus when you don’t use them,” said Chawla.
Researchers studied a linguistically rich dataset derived from CaSiNo, a scenario-based collection of dialogues used for teaching negotiation skills to human participants through role-playing. CaSiNo, revolves around realistic negotiation scenarios that take place between two neighbors at a campsite who aim to obtain extra camping essentials such as food, water, and firewood.
This exercise leads to the collection of conversations full of language used in negotiations. In this study, the focus is on analyzing the subjective measures of negotiation performance, which include the participants’ reported satisfaction with the outcome and their perception of their negotiation partner. Using recent advancements in deep learning, Chawla’s team developed three degrees of emotion recognition techniques: emoticons, lexical, and contextual emotion.
Earlier research used the personal characteristics of individuals to understand how they negotiate and what the results of those negotiations are. However, relying only on these variables misses out on all the other available information, like the affective attributes used in the negotiation itself, which may further help in predicting the outcomes.
“We argue that if you capture emotional expression in the conversation, it helps above and beyond just looking at the demographics and personality of the participants in predicting the subjective outcomes of the negotiation,” said Chawla.
Chawla’s co-authors in this research include Rene Clever and Jaysa Ramirez from USC’s ICT, Gale M. Lucas, a USC ICT researcher and research assistant professor at USC Viterbi, along with Jonathan Gratch, director for virtual human research at USC’s ICT and research professor of computer science and psychology at USC Viterbi.
The insights gained from this team’s work will be helpful in designing adaptive negotiation agents that interact through realistic communication interfaces. Virtual negotiation agents have broad applications in advancing conversational AI like the Google Duplex agent that books a haircut appointment over the phone.
“In the future users could also interact with AI negotiation agents to improve their own social skills,” said Chawla.
As JFK once pointed out, fear, aggression, or timidity are common emotions seen in negotiations. This is a fact that artificial intelligence is now considering. Insight, that will no doubt be valuable in developing virtual negotiation agents that can interact with humans in a more realistic and effective manner.
Published on April 12th, 2023
Last updated on May 16th, 2024
