Robots teaching other robots – it’s no longer science fiction

In a world where everyone holds their unique area of expertise, the exchange of knowledge can be an enlightening experience. This got researchers pondering — can the same principle be applied to machines, with robots teaching other robots?

Take for instance a poker pro, a connoisseur of French cuisine, and a Mozart aficionado. When these three combine their knowledge, they learn something new from each other and walk away richer in wisdom than before.

Can robots, much like their human counterparts, expand their understanding by exchanging information with one another?

Under the guidance of Laurent Itti, a Computer Science Professor, and his Ph.D. student Yunhao Ge, a team of researchers embarked on a mission to find an answer to this intriguing question.

Their extensive study, published in the May 2023 issue of the ‘Transactions on Machine Learning Research’ journal, has led them to a startling discovery – the answer is yes!

Their research paper, titled ‘Lightweight Learner for Shared Knowledge Lifelong Learning‘, introduced a new perspective to the growing field of Lifelong Learning (LL) within machine learning.

LL explores the concept of AI agents continually expanding their knowledge as they come across new tasks, while also retaining their prior learning.

“It’s like each robot is teaching a class on its specialty, and all the other robots are attentive students,” Ge playfully illustrated.

Robots teaching robots using SKILL

The research duo went on to introduce their creation – SKILL (an acronym for Shared Knowledge Lifelong Learning). SKILL employs a unique method allowing AI to master 102 separate tasks.

These include categorizing tens of thousands of images of cars by model, flowers by species, or chest X-rays by diseases. Upon learning these tasks, the AI agents shared their newfound knowledge over a decentralized network, eventually becoming proficient in all 102 tasks.

Itti and Ge believe their research presents a new direction in Lifelong Learning. Current research mainly focuses on a single AI agent learning tasks in sequence, a method that they believe is inherently slow.

Their SKILL tool offers a set of algorithms that accelerates the learning process by enabling multiple agents to learn simultaneously. They found that when 102 agents each learn a task and then share, the total learning time reduces by a factor of 101.5, even after considering the time needed for communication and knowledge consolidation among agents.

“Traditionally, you first collect all the data you want your AI to learn, then you train the AI to learn it. But just like people, we’re trying to create AI agents that can keep learning after they discover new things,” explained Itti, the driving force behind SKILL, a project that has received funding from the Defense Advanced Research Project Agency (DARPA).

Itti and Ge believe this is only the beginning, and they have yet to test this novel approach to learning on such a vast number of natural tasks.

“We believe this research, in the future, can be scaled up to thousands or millions of tasks,” projected Itti. He anticipates that in a few years, Lifelong Learning could revolutionize many aspects of our lives, potentially creating a truly connected, intelligent, and efficient global community.

Moving beyond basic recognition

From revolutionizing healthcare to enhancing our digital tour guide experiences, the applications of SKILL are endless. In healthcare, for example, different AI systems could become specialists in different illnesses, treatments, patient care techniques, and the latest research.

Once they consolidate their knowledge, these AIs could function as a comprehensive medical assistant. They could provide doctors with up-to-date, accurate information across all medical fields.

“In essence, any profession requiring vast, diverse knowledge or dealing with complex systems could significantly benefit from this SKILL technology,” Ge noted. While the current version of SKILL focuses on image recognition, future iterations will look to tackle more sophisticated tasks.

The researchers draw parallels between their work and the concept of crowdsourcing, where many people each contribute a piece of the solution to a larger problem.

“In crowdsourcing, many people tackle a piece of a problem and when the knowledge is shared, you have a solution. Now we can do the same thing with AI agents,” Itti stated.

“What if you, as a single person, had to relearn all of human knowledge?” Itti posed, underlining the enormity of such a task. “Humans have the means of sharing information. We are now pushing that idea into the AI domain.”

This study is a significant step forward in realizing that vision, potentially setting a new paradigm for AI learning.