When will negotiation agents be able to represent us?
At the Twenty-Sixth International Joint Conference on Artificial Intelligence (IJCAI-17) in 2017, Tim Baarslag and team presented research that outlined how computers that negotiate on our behalf held great promise for the future and would even become indispensable in emerging application domains such as the smart grid and the Internet of Things. However, at that time, truly autonomous negotiators had rarely been deployed in real-world applications.
Fast forward to the 11th Annual Automated Negotiation Agent Competition
Earlier this month, the Olympics for hagglebots was held: the 11th annual competition for artificial intelligence (AI) that has been trained to negotiate. Called the Automated Negotiating Agent Competition, it pits more than 100 participants from Japan, France, Israel, Turkey and the United States against one another in five leagues. Imagine if it was 1,000 participants from across the planet!
If everything is known, “negotiations should not have to happen at all. You need zero rounds, and the deal is struck immediately,” says Tim Baarslag, co-founder of the competition. Machine learning now helps AI predict the other side’s preferences based on a few observations, plus lots of experience in prior negotiations. Once you know these things, even a negotiation involving thousands of side deals “just becomes a big calculation, something computers are amazingly good at”.
- AI Negotiation Platforms: “We now see that some vendors prefer talking to a bot. If there are tens of thousands of vendors, it’s hard to get human attention sometimes,” says Martin Rand, CEO of Pactum, who are working with Walmart.
- “We had a problem at BP. It was taking us 90 to 120 days to get a contract done,” says Michael O’Brien, who managed the oil company’s $2bn (£1.5bn) budget for IT services from 2012 to 2020. So he made a machine-learning tool together with AI developer App Orchid that could learn what BP actually agreed to in past contracts.
- ML Negotiation Applications: “Machine learning helps us learn to recognise when negotiations are going well or badly”, says Jared Curhan, MIT.
- AI Negotiation Advisors: An AI listening by microphone to the first five minutes of a negotiation can predict 30% of the variation in its eventual outcome, just from negotiators’ voices. This research could produce AIs “as an adviser on your shoulder, whispering in your ear. ‘I think they’re lying – you should push harder,’” says Johnathan Mell, University of Central Florida.
“Humans often fail to reach the best agreements where computers can. On top of that, computer negotiation can be done very fast.” Tim Baarslag, 2017