Model negotiations An EU team considered the design of negotiation software systems that interact with humans. Researchers used game theory to model negotiation behaviour, including cultural differences, yielding a system at least as proficient as human negotiators. Digital Economy © Thinkstock Computers today often assist humans with business and other negotiations, or otherwise conduct the negotiations by themselves, sometimes against other computers. Yet, little attention has been given to design elements for computers that negotiate with people. The EU-funded IEPHCN (Information exchange policies for human-computer negotiation) project aimed to develop appropriate computational strategies. The intention was to model human-computer negotiations, and to derive ways of creating systems able to successfully conduct negotiations. Other research aspects involved modelling cultural considerations, negotiation tactics, trustworthiness indicators and commitment behaviour. The four-year project concluded in September 2014. Objectives were met via a series of models created using game theoretic reasoning, decision theoretic models and machine learning. Researchers created settings whereby people's behaviour could be studied, leading to three main results. Firstly, the team designed novel settings for studying argumentation and negotiation. The settings included three types of game involving information revelation, contract bidding and a contract game where players had to reach agreement. A second result was the construction of decision-making and information exchange models for computer-human strategic negotiation scenarios. Lastly, all agent designs were cross-culturally evaluated, using hundreds of subjects of various types, from China, Israel and the United States. Researchers concluded that, on average, the rule-based computer agent was able to negotiate as well as people from all test countries. The agent outperformed people only in the United States, while in Lebanon people performed better. Learned based models outperformed people from all three countries, while also showing adaptability. Concerning the contract game, the computer acting as customer outperformed people playing the same role from all countries. Chinese people exceeded provider agents. Results of the corruption game corresponded to a perceived corruption index: considerably higher in China, intermediate in the United States, and slightly lower in Israel. General results of the IEPHCN project yielded new understanding of human-computer strategic decision making. The resulting system designs sometimes outperformed people, while also facilitating their decision making. Keywords Negotiation, negotiation software, game theory, information exchange, human-computer negotiation