Researchers at Yale University have developed an analysis software in which robots know how to distinguish between the tools they own and those owned by other people or other robots.
How smart and self-aware should robots be? At a time when advances are being made in the field of robotics, the fear of making robots and artificial intelligence smarter than usual has caused many to remain sceptic on the issue. Yet, with a future where robots will be part of our daily personal and professional lives, some experts believe it necessary to inculcate social conventions in our algorithmic overlords.
Indeed, this is the process undertaken by a team from Yale University (USA) who developed a software that allows a robot to learn to recognize and respect the property of others. The program combines two machine learning algorithms, one using explicit rules, the other using Bayesian inference to infer the property based solely on the qualities of the object being observed.
In their article published by arXiv, the researchers explain that associating these two approaches has allowed them to reproduce the way humans learn to know and respect property by using both explicit rules and empirical learning. “Understanding object ownership, permissions, and customs is one of those topics that has not really received much attention, but that will be critical to the way machines work in our homes, schools, and ofices,” says Brian Scassellati, one of the researchers behind this project.
The learning software platform has been tested on a Rethink Robotics Baxter robot that performs tasks according to property rules. But its designers say it could work with other models of robots.