Artificial intelligence (AI) systems have long been inspired by the intricacies of the human brain. However, a groundbreaking branch of research at Columbia University in New York is delving deeper into the realm of AI, seeking insights that may unravel the mysteries of the brain and enhance its functionality.
Columbia University was selected as one of the National Science Foundation’s designated national AI research institutes. With a generous $20 million grant, the university’s AI Institute for Artificial and Natural Intelligence (ARNI) is poised to advance research that bridges the gap between significant AI advancements and our evolving understanding of the brain.
Richard Zemel, a computer science professor at Columbia, explained that the ultimate goal is to bring together leading AI and neuroscience researchers in a collaborative exercise that mutually benefits AI systems and human cognition.
“It’s a two-way street,” Zemel stated. “AI has drawn inspiration from the brain, and neural networks possess loose connections to it.”
Mimicking the brain’s structure has been a central theme in AI research, with the aspiration of creating machines capable of cognitive thought. Artificial neural networks, composed of millions of processing nodes, enable AI systems to learn from data they are fed.
In recent years, the emergence of transformer neural networks has aimed to mirror the human brain more closely. These networks focus on contextualizing questions to provide more precise answers. Zemel highlighted the significance of “attention” in transformers, drawing parallels with the brain’s ability to select and attend to relevant stimuli in a noisy environment, commonly known as the cocktail party effect.
The concept of “attention” in AI systems has made generative AI outputs increasingly usable for individuals interacting with AI. This progress has led researchers to contemplate whether advancements in AI could yield insights into the functioning of the brain.
Zemel postulated, “By understanding these complex neural networks, can we develop hypotheses and explore new avenues of investigation in the brain?”
Columbia’s research will explore fundamental questions, including the concept of “robust flexible learning.” Many AI systems excel at specific tasks but struggle when presented with new challenges, while the human brain demonstrates adaptability. Leveraging AI’s rapid language acquisition capabilities, Zemel believes that understanding how AI trains efficiently could improve human learning methods.
“Many of these new AI systems excel at acquiring new language tasks quickly, often outperforming humans with just a couple of examples,” Zemel noted. “This prompts us to question how we can adapt human training strategies accordingly.”
Continual learning, the ability to retain and recall information, presents another area of study that intersects both AI systems and human cognition. Zemel explained that both entities face challenges in forgetting, providing fertile ground for investigating ways in which they can mutually assist each other.
The principle of uncertainty emerges as a third common concern shared by both AI systems and humans. Zemel emphasized that current AI systems often struggle to recognize their own uncertainty, mirroring the limitations of human judgment.
Practical applications arising from this cross-training of AI and human brains are already in progress. For example, the development of “brain-machine interfaces” facilitates the creation of advanced prosthetic devices. These interfaces combine brain signals with AI technology to enable precise control over prosthetic limbs, benefiting individuals with impaired motor function.
Zemel expressed his hope that the collaborative efforts between AI and neuroscience experts at Columbia will continue to yield such advancements.
“We are fostering an environment where these brilliant minds can collaborate, exchange ideas, and uncover new avenues for testing and exploration,” Zemel said.
In conclusion, the convergence of AI and neuroscience research at Columbia University holds promise for unraveling the complexities of the human brain. By leveraging the strengths of both fields, scientists aim to unlock transformative insights that will shape the future of AI systems and enhance our understanding of human cognition.