Facebook researchers progress on teaching AI chatbots to negotiate
Facebook researchers say they have made progress in training chatbots to negotiate.
The social networking giant's Facebook AI Research (FAIR) group published a paper to show how bots can be used to plan ahead in a conversation and adapt negotiation strategies.
Facebook's overall goal is to create chatbots that can reason, converse, and negotiate so its personal assistant can compete with rivals such as Google (Google Assistant), Amazon (Alexa), Apple (Siri) and Microsoft (Cortana) to name just a few.
In a blog post, Facebook noted that chatbots can hold short conversations and carry out simple tasks. But meaningful conversations is a challenge because the chatbot has to combine its knowledge of the world while understanding the conversation.
Facebook, which open sourced its code and published a paper, is hoping the research leads to better negotiations, intelligent compromises and novel sentences. Facebook, which in April rolled out a limited version of its bot, noted:
In order to train negotiation agents and conduct large-scale quantitative evaluations, the FAIR team crowdsourced a collection of negotiations between pairs of people. The individuals were shown a collection of objects and a value for each, and asked to agree how to divide the objects between them. The researchers then trained a recurrent neural network to negotiate by teaching it to imitate people's actions. At any point in a dialog, the model tries to guess what a human would say in that situation.
Unlike previous work on goal-orientated dialog, the models were trained "end to end" purely from the language and decisions that humans made, meaning that the approach can easily be adapted to other tasks.
Read also: MasterCard sees Facebook Messenger as a commerce, bot vehicle for Masterpass
Facebook's chatbot paper lands just as some analysts are questioning whether the company's AI efforts are actually making it into products. For instance, Edison Investment Research analyst Richard Windsor said in a research note that Facebook's digital assistant lacks AI in production. Windsor said:
Facebook M has been in beta for over 18 months and has comprised of a combination of automated responses and human interactions where the vast majority of the tasks have been carried out by humans. The problem with using humans of Digital Life services is that it is very expensive to scale the service for 2 billion users especially when the service will be funded by advertising. This why Facebook is working as quickly as it can to develop its in house expertise and while it remains a laggard in AI, it has shown some progress.
For example, at its developer conference, it showed some good progress on machine vision enabling its apps to recognize the world they can see through the smartphone camera. It also made Facebook M available to US users and most recently in Spanish to users in Mexico and US.
However, what has gone live is only a small part of the grand plans that were announced in 2015 which had an always on, all knowing bot with which the user could do almost anything.
In a nutshell, Facebook's research team is working toward that do-it-all bot, but it has some work to do.