Creating microprocessors capable of replicating biological learning systems is the new frontier in artificial intelligence (AI) research, as scientists strive to make AI more adaptable, efficient, and environmentally sustainable. Coordinated by the Neuromorphic AI Lab (NUAI Lab) at the University of Texas at San Antonio (UTSA), an international group of researchers is taking on this challenge. Vincenzo Lomonaco, a leading expert in Continual Learning from the University of Pisa, Italy, is part of this team. In a recently published article titled Design principles for lifelong learning AI accelerators in the esteemed scientific journal Nature Electronics, Lomonaco emphasizes the need for a paradigm shift in AI.
Lomonaco highlights the current limitations of AI, stating, The fallibility of Artificial Intelligence is still too high, and this is because AI, as we know it today, is based on non-adaptable machine learning systems, which make it incapable of dealing with new conditions not previously encountered during the training process. He further explains that the existing AI models require complete retraining when confronted with new information, making them inefficient, costly, and environmentally harmful due to high energy consumption and carbon dioxide emissions.
Upgrading an AI system can incur expenses of several million euros, and a study conducted by the University of Massachusetts reveals that training large AI models emits approximately five times the amount of carbon dioxide produced by an average American car throughout its entire life cycle, including manufacturing.
To address these challenges, Lomonaco and his colleagues at the NUAI Lab propose Continuous Automatic Learning, also known as Continual Learning or Lifelong Learning, as a solution. This approach allows AI systems to assimilate knowledge in a sequential manner without disregarding previously acquired knowledge.
The researchers aim to develop a new learning system by changing current computational paradigms and eliminating infrastructural constraints. Lomonaco explains, That is why, with some colleagues at the NUAI Lab in San Antonio, we laid the groundwork for a new incremental learning system, based on hardware-software co-design. Designing hardware and software components simultaneously, we aim to create a robust and autonomous lifelong learning system for AI. This system will utilize next-generation algorithms inspired by human intelligence, enabling AI to continuously expand its knowledge in a faster, more efficient manner, with energy consumption similar to that of a light bulb.
By achieving this goal, AI could become more adaptable, efficient, and sustainable, paving the way for a future where AI technologies can effortlessly handle new conditions and contribute to solving complex global challenges.
The pursuit of a more human and sustainable future for AI serves as a crucial development in the field. As researchers persist in their efforts to enhance AI’s capabilities, they endeavor to create AI systems that are not bound by their initial training but can learn, adapt, and contribute to the world in a more intelligent and environmentally friendly manner.
