Market Watch: The Negotiation Algorithm

Since humans developed their own languages, negotiation became an inseparable part of human society, soon becoming one of the bedrocks of society's functioning. Negotiation exists anywhere from diplomatic missions between governments to the bargaining between street vendors and their customers. Negotiation can either be compromising or an attempt to get the most for yourself. Most of us had attempted to negotiate with someone in some point of our lives for our own goods. For experienced diplomats, negotiators, lawyers, or bargaining shoppers, they must maintain their heuristic method to complete their jobs and achieve their goals. However, are we ready turn these heuristic methods into algorithm; would it be possible for a machine to match the skill of negotiation that humans had refined throughout the entirety of their existence?

Modern high tech companies like Microsoft and Apple have developed speech recognition functions in their AI products like Cortana and Siri. However, they are only enough to understand a specific language and some specific types of questions. Such simple conversations can be done by decision trees since there aren’t many possible outcomes for the users at this point. 

Unlike simple conversations, negotiations are usually able to bring a variety of answers to the same question depending on the goal. Recently, Facebook and the Technology Institute of Georgia revealed themselves to be planning a negotiating AI. The project name is called “Deal or No Deal? End to End Learning for Negotiation Dialogues.” Facebook is going to use game theories and deep learnings to formulate negotiation techniques. As for what the developers described in their abstract, negotiation require complex communication and reasoning skills. AI will need to have a much larger decision tree to prepare for different scenarios. 

As one of the largest social media platforms in the world, Facebook had gathered a dataset of 5808 dialogues between humans in a negotiation task. Their AI will need learn the languages of these negations in order to win said negotiations while trying to do so as a human. Sometimes in negotiation, humans present anger, pleasantries, confusion, boredom and other emotions. In addition, humans are more willing to speak with someone that sounds like them. However, this research won’t include the heuristic approaches of negotiation masters. At this moment, they are only simple negotiators that mostly to compromise any requests, as the research report points out: 

We first train recurrent neural networks to imitate human actions. We find that models trained to maximize the likelihood of human utterances can generate fluent language, but make comparatively poor negotiators, which are overly willing to compromise. We therefore explore two methods for improving the model’s strategic reasoning skills— both of which attempt to optimize for the agent’s goals, rather than simply imitating humans. 

At the same time, by looking at the past speed of skills AI have learned from humans, robots have been able to defeat grandmasters in complex games such as chess. Chess also possesses multiple possible moves, while a multitude of different outcomes and scenarios exist within the game. Nevertheless, AI are able to predict these grandmasters’ moves and defeat them at their own craft. After a number of years, these AI had learned many chess strategies and are able to adapt to the moves of their human counterparts. It may take a few years for these AI to master the art of negotiation verbally, but the result may become just as intricate.

Intelligent agents often need to cooperate with others who have different goals, and typically use natural language to agree on decisions. Negotiation is simultaneously a linguistic and a reasoning problem, in which an intent must be formulated and then verbally realized. – Introduction of the “Deal or No Deal?” research.

According to the researchers Michael Lewis and Dhruv Batra, this algorithm will serve as an assistant to prevent their human masters from making bad decisions in general instead of making the best decision for them. The relationship between AI and user will be like a coach and athlete. The coach can give advice and trainings but it will be up to the athlete to perform in the game. 

Negotiation could be seen as a fight where both parties are trying to get the most from the deal. Usually, winning a negotiation is when your own goal is satisfied. For example, it is easy to tell the result between an unbalanced fight between a weak, unarmed person against a trained swordsman. The swordsman can defeat the unarmed person quickly. In negotiation, the side with more bargaining power is always more likely to win. At the same time, the weaker side still can turn the tide, either due to experience or circumstance. However, when both sides are more balanced in strength or bargaining power, they will then use their negotiation techniques extensively to win.   

In addition, the goal of action will be in consideration. Do you want a win-win or a win-lose result? This will depend on the situation and resources available. 

Whenever this AI becomes implemented, it can be supportively used on diplomatic missions, or perhaps even hostage rescues, public relations, or business agreements - activities that require precise communication. Meanwhile, like any other innovation in AI, the downside is that is can potentially take over certain jobs, leading to a new group of professionals needing to look for new work. Inevitably, this will also lead to the discussion of whether the future will be a utopia with super high tech assistants for future generations of lax humans, or a technologically advanced dystopia with a few powerful companies that utilize machines like these to hold their clear monopoly of power. Like the machines themselves refining their talents, only time may tell.