Artificial intelligence (AI) systems have long drawn inspiration from the intricacies of the human brain. Now, a groundbreaking branch of research led by Columbia University in New York seeks to unravel the workings of living brains and enhance their function by leveraging advancements in AI.
Designated by the National Science Foundation as one of seven universities serving as the headquarters for a new national AI research institute, Columbia University received a substantial $20 million grant to bolster the AI Institute for Artificial and Natural Intelligence (ARNI). ARNI is a consortium comprising educational institutions and research groups, with Columbia at the helm. The overarching goal of ARNI is to forge connections between the remarkable progress achieved in AI systems and the ongoing revolution in our understanding of the brain.
Richard Zemel, a professor of computer science at Columbia, explained that the aim is to foster a cross-disciplinary collaboration between leading AI and neuroscience researchers, yielding mutual benefits for AI systems and human beings alike. Zemel emphasized that the exchange of knowledge flows in both directions, with AI systems drawing inspiration from the brain while neural networks in turn bear loose resemblances to its structure.
Traditionally, AI has sought to mimic the brain’s structure in order to create thinking machines. Artificial neural networks, modeled after the brain, consist of millions of processing nodes that facilitate learning when exposed to data. In recent years, the “transformer” neural network has gained traction, aiming to bring AI systems even closer to the human brain. Transformers learn to attend to the most pertinent words or phrases in a question, enabling them to produce appropriate responses. Zemel described this concept as “attention,” drawing an analogy to the “cocktail party effect” where the brain selectively focuses on relevant information amidst a sea of conversations.
The idea of attention has enhanced the usability of generative AI output for users interacting with AI systems. Consequently, researchers are now contemplating whether AI advancements can provide insights into the functioning of the brain. By comprehending complex neural networks, Zemel suggests that researchers may formulate hypotheses and explore new avenues of investigation in neuroscience.
Columbia’s research will address significant questions, including the concept of “robust flexible learning.” While current AI systems excel at specific tasks, they often struggle when presented with new challenges, unlike the adaptable human brain. However, AI has demonstrated rapid language skill development, prompting Zemel to propose that understanding this AI talent may help optimize human brain training methods.
Continual learning is another critical area of investigation, exploring the processes of forgetting and memory recall in both humans and AI systems. Zemel highlights the challenges faced by AI systems, which often suffer from forgetting, and the opportunity to find solutions that can benefit both realms.
Uncertainty poses a third common challenge for both humans and AI systems. Zemel notes that many existing AI systems struggle to recognize when they should exhibit uncertainty, mirroring the limitations observed in human behavior.
The practical applications arising from this cross-pollination of AI and human brain research are already in development. For instance, “brain-machine interfaces” are being utilized to create smarter prosthetic devices, such as mechanical arms, for individuals with impaired motor control. Zemel mentions the development of “AI-assisted prosthetic devices” that combine brain signals with AI interfaces to enable movement.
Columbia University aims to foster further connections between AI and neuroscience experts, facilitating the exchange of ideas and propelling new avenues of exploration. Zemel envisions bringing these specialists together, enabling fruitful collaborations that drive scientific testing and discovery.
By Impact Lab