There’s a poker wizard. A friend who knows all about French cuisine. Another friend who is a Mozart expert.
As the three gather, an enriching exchange of knowledge takes place, with each individual sharing insights from their respective areas of expertise. In this dynamic interaction, valuable lessons are gained by each participant, as they leave the meeting having learned something new from the other two. The collaborative atmosphere fosters a fruitful learning experience that leaves everyone enriched with fresh perspectives and insights.
Sharing and exchanging information is a fundamental way through which people acquire knowledge. Similarly, in the realm of artificial intelligence, computers and robots can also partake in this process, facilitating collaborative learning known as “machine learning” or “robot learning.” The question arises: can computers and robots effectively teach one another to learn by sharing their knowledge and experiences? This intriguing concept explores the potential for machines to engage in knowledge transfer and mutual learning, similar to how humans share information to enhance their understanding.
In May 2023, the journal Transactions on Machine Learning Research featured a paper tackling this question, authored by a team of researchers led by computer science Professor Laurent Itti and one of his Ph.D. students, Yunhao Ge.
They come up with a resounding answer:
Yes.
In their paper titled “Lightweight Learner for Shared Knowledge Lifelong Learning,” the researchers present an innovative approach to the burgeoning field of machine learning (ML) research known as Lifelong Learning (LL). This approach allows AI agents to continuously learn while facing new tasks, all the while retaining knowledge gained from previous tasks.
“It’s like each robot is teaching a class on its specialty, and all the other robots are attentive students.” Yunhao Ge.
In their paper, Itti and Ge introduce their creation, SKILL (Shared Knowledge Lifelong Learning), a powerful tool designed to enable AI systems to learn and master 102 diverse tasks. These tasks encompass a wide range of challenges, such as classifying tens of thousands of car images according to their models (Ferrari, Jeep, Cadillac), identifying various flower species, or diagnosing diseases from chest X-rays.
The AIs then shared their knowledge over a decentralised communication network and eventually mastered knowledge of all 102 tasks.
Ge described the collaborative learning process among the robots, likening it to individual robots acting as teachers conducting classes on their respective specialties, while other robots serve as eager and attentive students. Through a digital network connecting them, akin to their private internet, they actively share knowledge, fostering a dynamic and interconnected learning environment.
Itti and Ge called their work a new direction in LL research.
According to their explanation, the majority of current Lifelong Learning (LL) research revolves around a single AI agent learning tasks sequentially, a method that inherently presents slow progress.
As per their statements, the SKILL tool incorporates a collection of algorithms that significantly accelerate the learning process by enabling agents to learn in parallel simultaneously. Their research demonstrated that when 102 agents each tackle a distinct task and subsequently share their knowledge, the time required for completion is reduced by a factor of 101.5, accounting for essential communications and knowledge consolidation among the agents.
“Just like people, we’re trying to create AI agents that can keep learning after they discover new things.” Laurent Itti.
“It’s been the traditional approach,” Itti explained, “where you collect all the data you want the AI to learn, and then you train it accordingly. However, our goal is to develop AI agents that, much like humans, have the ability to continue learning even after they encounter new information.”
Scaling up
Itti views SKILL as a promising stepping stone for advancements in the field of Lifelong Learning (LL), with its development being supported, in part, by funding from the Defense Advanced Research Projects Agency (DARPA).
No prior research has involved so many natural tasks, Itti and Ge said. And this is just the start.
“We believe this research, in the future, can be scaled up to thousands or millions of tasks,” Itti said.
According to Itti’s estimation, this transformation could take place in just a few years. He believes that with the advancements in Lifelong Learning (LL), we could witness a significant impact on various aspects of our lives, bringing us closer to the realisation of a “truly connected, intelligent, and efficient global community.”
Itti provided an example of how Lifelong Learning (LL) could revolutionise the medical field. He explained that distinct AI systems could specialise in learning about various illnesses, treatments, patient care techniques, and the latest research. This collaborative approach among AI agents would potentially lead to more comprehensive and up-to-date medical knowledge, ultimately benefiting patients and healthcare providers alike.
“We believe this research, in the future, can be scaled up to thousands or millions of tasks.” Laurent Itti.
Following the consolidation of knowledge, Itti and Ge elaborated on how these AI systems could act as comprehensive medical assistants. By assimilating the collective expertise of specialised AI agents, they would be equipped to offer doctors the most up-to-date and accurate information across all facets of medicine. This enhanced medical assistance could prove invaluable in improving healthcare outcomes and decision-making for medical professionals.
The potential of Lifelong Learning (LL) extends beyond the medical field. An exciting scenario emerges where every smartphone user becomes a local tour guide while exploring a new city. Each user can contribute by taking photos and sharing valuable details about significant landmarks, stores, products, and local cuisine. Through collaborative knowledge sharing, this wealth of information can create an enriched and dynamic tourist experience, empowering travellers to make well-informed decisions and delve deeper into the culture and offerings of the places they visit.
Once this data is shared across a network, every user would have an advanced digital tour guide in his or her pocket.
Ge highlighted that the SKILL technology holds immense potential for any profession that necessitates vast and diverse knowledge or involves complex systems. In essence, industries ranging from medicine and engineering to finance and beyond could experience significant benefits through the application of SKILL, revolutionising how professionals approach their respective fields and enhancing their problem-solving capabilities.
Beyond recognition
The SKILL tool examined the ability of AIs to simply recognize what is in an image, Ge noted.
“Humans have the means of sharing information. We are now pushing that idea into the AI domain.” Laurent Itti.
“It’s a good starting point,” he remarked, referring to recognition tasks. However, he emphasised that future research will focus on deploying AI systems to undertake more sophisticated and complex tasks. As the technology evolves, AI agents are expected to progress beyond recognition capabilities and venture into realms that require higher levels of intelligence and problem-solving abilities, unlocking new possibilities and potential applications.
Itti and Ge said the concept of crowdsourcing – for example, online reviews of restaurants – is comparable to the idea described in their paper.
“In crowdsourcing,” Itti said, “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 emphasised the immense challenge of a single person attempting to relearn all of human knowledge, a task that would be insurmountable given its sheer magnitude. However, he pointed out that humans possess the ability to share information, and now, they are extending this idea into the realm of artificial intelligence. Through the collaborative and interconnected approach of AI agents, the vision is to replicate the human capacity for knowledge sharing, paving the way for unprecedented advancements and achievements in the field of AI.