Latest in Gear

Image credit: Honda Research Institute

Honda teams up with MIT and others to develop curious AI

It wants to design a system that can learn how to learn.
366 Shares
Share
Tweet
Share
Save
Honda Research Institute

Honda is teaming up with three universities on a project aimed at developing curious artificial intelligence. The new three-year initiative, dubbed the Curious Minded Machine, will work towards an intelligent system that can learn continuously, much like a human, and can actually "learn to learn," as children do. "Our ultimate goal is to create new types of machines that can acquire an interest in learning and knowledge, and the ability to interact with the world and others," Soshi Iba, a principal scientist at the Honda Research Institute, said in a statement. "We want to develop Curious Minded Machines that use curiosity to serve the common good by understanding people's needs, empowering human capability and ultimately addressing complex societal issues."

As an example of how a future Curious Minded Machine could work, Honda explains that an intelligent system could be able to observe human interactions and tasks, and then pinpoint better, more efficient ways for people to complete those tasks.

Each of the three universities partnering with Honda will take on a different problem. The University of Washington will work towards developing a mathematical model of curiosity, using the way human children learn through exploration as a guide. MIT's CSAIL program is focusing on designing systems that better predict phenomena by noting what it can't predict well and finding a way to obtain information that could help improve its prediction abilities. Lastly, the University of Pennsylvania aims to mimic biological learning in AI systems by applying "an embodied, active and efficient approach to acquiring representations of the surrounding world and actions."

These three groups and the Honda Research Institute will collaborate as they work together and the university teams will produce bi-annual reports as they go. Data sets, open source software and other results from their work might also be released at the end of the project.

From around the web

ear iconeye icontext filevr