Artificial agents that interact with humans may have a wide range of valuable applications, for example, helping humans improve their negotiation skills in various fields. To spur the development of these agents, researchers at the University of Southern California (USC) recently created CaSiNo, a dataset that includes realistic interaction dialogues in a camping scenario.

“Our work reflects our ongoing efforts to create automated negotiation systems,” Kushal Chawla and Gayle Lucas, two researchers studying, told TechExplore via email. “Studying how humans interact has been an active area of ​​research in economics, psychology, and affective computing for decades. It’s an interesting playground for multi-disciplinary research revolving around human decision making. ”

In recent years, many researchers worldwide have begun exploring the potential of automated systems that can interact directly with humans. They found that these systems could be particularly helpful in training people on specific social skills (for example, teaching professional students to negotiate successful deals or lawyers to assess settlement rates during legal proceedings ).

“There is already evidence that negotiation skills are also important to advance the capabilities of existing AI assistants,” said Chawla and Lucas. “For example, Google duplex prototype is engaged in a simple form of conversation to book haircut appointments over the phone.”

Most automated negotiation systems developed so far are based on restrictive menu-driven communication interfaces. For example, systems based on the IAGO platform, including the framework previously developed by Chawla and Lucas, require human users to click on specific buttons to communicate with the agent.

“These systems require sharing personal preferences or clicking a button to roll out offers,” said Chawla and Lucas. “While this restriction provides brevity, it comes at a cost. More specifically, it hinders the analysis of many aspects of real-world negotiations, such as persuading a negotiation partner or expressing feelings. Alternative systems Those that enable more realistic styles of communication (i.e., via text or video) can be highly desirable. ”

To overcome system limitations with menu-based interfaces, some research teams have recently been trying to develop chat-based negotiation systems that allow users to communicate more freely by typing or speaking in a human language like English. Allows to do. However, developing and training these systems is far more challenging than creating menu-driven systems.

Chawla and Lucas stated, “The creation of a system that can interact with human partners in a given language requires the creation of a dataset of interactions on which machine learning models can be trained.” “Earlier efforts aimed at developing such datasets have either focused on game settings that are too restrictive that they impede individual interactions, or are too open that they evaluate interaction performance Hurt both from the perspective of downstream applications. ”

In their recent paper, Chawla and Lucas introduced a dataset containing more than a thousand realistic, linguistically rich, and personal conversational dialogues within a clearly delineated environment, namely a camp site. This dataset is called CaSiNo, which means “camp site negotiation.”

“In each interaction, two participants play the role of campsite neighbors and interact for additional essential items (ie, food, water, and firewood),” Chawla said. “Each participant has a predetermined preference toward these items and may or may not require them to have their own justification (eg, increased water for bonfires with friends or more water supplies for firewood is).”

In addition to dialogues, the CaSiNo dataset includes relevant information about each participant, such as which items they most need or are expected to obtain through interaction. During each talk, the two participants talk to each other to decide how to divide the nine packages: food, three water and three firewood in three.

“Participants’ negotiation performance is evaluated in three ways: (1) their final score, depending on what items they were able to negotiate for, (2) how satisfied they were with their performance And (3) how much they like their opponents, ”Chawla and Lucas explained. “All these metrics are important in the context of real-world interactions. Especially in cases where participants engage in repeated interactions with each other, maintaining their relationships may be just as important as their own performance. is.”

Researchers specify approximately 40% of the dialogues in the CaSiNo dataset, specifying the persuasion strategies used by the negotiating parties.

Leave a Reply

Your email address will not be published. Required fields are marked *