Chatbots negotiating: humans might never need to barter again

Based on recent findings, chatbots could soon be negotiating on our behalf to arrange holidays, buy cars and even organize our next date.

Every revolution needs a face and the upcoming artificial intelligence powered one is certainly no exception. Enter negotiating chatbots. Indeed, the development of virtual assistants is a useful standard bearer in which to gauge the speed at which we are veering towards high-level machine intelligence (HLMI).

The right chatbots with robust natural language processing are already more than capable of performing incredibly useful tasks for humans. Be it answering FAQs or aiding customers in the purchasing process, bots have evolved significantly for e-commerce businesses in order to increase revenue and enrich the customer experience, all while cutting costs.

But chatbots will be able to do so much more. In fact, recent weeks have given an indication of what we can expect from them in the coming years.

Chatbots: the new wheeler dealers

In the future, when we look to purchase a car, book a holiday or even arrange dinner with friends, chatbots could be the ones negotiating on our behalf. In fact, Facebook recently demonstrated virtual assistants which could, in theory, do just that.

Facebook trained their bots to take part in a game in which players have to decide how to divvy up hats, balls and books. Each of the items was assigned different points for each player, the winner being the one with the most at the end of the round.

The chatbots were able to use negotiating tactics similar to such concepts as game theory by incorporating the “prisoner’s dilemma”. In fact, they were even able to deceive one another during the games by aggressively pursuing items they didn’t care about and conceding at the last moment in order to foster an appearance of compromising.

What makes this even more surprising is that the machines did this without any explicit human design, they were simply trying to achieve their goals.

The key innovation provided by the researchers was the introduction of “dialogue rollouts” which ensured the bots planned several steps ahead. In order for this to happen, the machines simulated completed dialogues during decoding to estimate the reward of each move. Facebook says that concentrating on the reward rather than likelihood function improved the performance against both humans and machines.

Still a long way to go

Great strides have already been made but there are still significant areas of improvement needed before chatbots are unleashed to haggle with second-hand car dealers.

The bots were only able to use basic sentences to negotiate within the game, an example of which is below:

Agent 1: I want the books and the hats, you get the ball

Agent 2: Give me a book too and we have a deal

Agent 1: Ok, deal

While prose good enough to threaten the betting odds for the next Man Booker Prize is not the priority, it is necessary in order to be able to adapt depending on the context and situation. Using these basic sentences will not work if you are applying for a mortgage for example.

As outlined here, we are still a long way off from natural language sentences rather than those that can be understood.

Why chatbots could be organizing our next date

Rather than simply haggling on purchases, it is conceivable that chatbots will be arranging our work/social schedules and even steering negotiations that we are involved in ourselves.

As outlined by Microsoft CEO Satya Nadella, it is likely that chatbots will eventually be with us across all forms of negotiation and will be able to pick up on our personal preferences. In addition, unlike humans, a chatbot would be able to mediate with several entities at once, for different items. We could soon be living in a world where we are able to book a holiday, arrange a date and buy a house all at the same time.

There are already multiple examples of chatbots negotiating on our behalves. Trim offers a chatbot to barter with Comcast over your cable bill while the Facebook Messenger bot DoNotPay is able to challenge parking tickets.

Oxford economists Carl Frey and Michael Osborne had identified three types of intelligence which pose a challenge to AI. Social intelligence was considered the one least likely to be reached in the near future thanks to the need for quintessentially human traits such as “negotiation, persuasion and care.” We might have underestimated the power of AI. In the future when humans ask prospective boyfriends/girlfriends on a date, they might have to go through their virtual assistant first.

Inbenta utilizes its patented natural language processing and +11 years of research & development to create interactive chatbots with an industry leading +90% self-service rate.

Companies around the world including Ticketmaster UK utilize the InbentaBot to maintain a personal service for their customers while reducing support tickets.

Interested in finding out more? Our team of experts is at your service to design a custom proposal for you.

Let’s get in touch