Artificial Intelligence that collaborates: what we can expect
We’re used to comparing the human potential with the abilities of a machine. The most classical example is when a chess champion challenges an artificial intelligence! During these years many tests verified the strength and weakness of both parties. But what about the collaboration capability of machines? Can they work through compromises instead of competition?
The experiment on collaboration
Professors Jacob Crandall and Michael Goodrich from Brigham Young University, in collaboration with colleagues from MIT and other international universities, realized an innovative algorithm. This program allows machines to cooperate with each other with a better efficiency than what happens between a group of people!
To prove the effectiveness of the invention, the scientists set an experiment. The goal was to give artificial intelligence the tool to learn and apply schemes of interaction between humans. The algorithm called S# was installed on some machines, that ran through many 2-players games. The tests included human-human, machine-machine and human-machine interactions.
Researched also inserted into artificial intelligence a range of “cheap talk” phrases. During interactions with a human, they could react to positive actions of their counterpart with sentences like “I accept your proposal” or “Sweet, we are getting rich”; in case of negative actions, the human could be met with a “Curse you” or “You will pay for that”.
The results of the study
In almost every test, machines implemented with algorithm S# outperformed humans in finding compromises and useful solution for both parties. Crandall underlines that half of the people lied at a certain point of the experiment, decreasing the value of collaboration! The honesty of artificial intelligence represented their real winning trait.
The use of “cheap talk” had great results too. In fact, it doubled the cooperation in every test where it was included. Furthermore, in most cases, humans involved in these particular instances weren’t able to tell if they were playing with a machine or another person!
Real-world applications of the algorithm
This academic experiment opens the doors to new applications for artificial intelligence. We’re used to applying machines to repetitive tasks o complex analysis: now we can have virtual assistants capable of collaborating with us! The chance to be helped and receive feedbacks will make us live interactions always more like those with another person.
We can expect in the next future artificial intelligence that will learn to work with us knowing our strong and weak points. And we’ll receive the help we need to complete our tasks in the best possible way! Without speaking of the way machines will be able to collaborate with each other, reaching new levels of integration.
Are you thrilled for the news about technological innovations? Continue following our blog to keep yourself up-to-date!