Talking to oneself is a trait that feels inherently human. Our inner monologues help us organize thoughts, make decisions, and understand emotions. However, it’s not just humans who can benefit from such self-talk. Scientists from the Okinawa Institute of Science and Technology (OIST) have demonstrated the potential of inner speech to improve AI learning, showing how AI models can generalize across different tasks more easily when supported by both inner speech and short-term memory. This groundbreaking study was published in the journal Neural Computation.
“This study highlights the importance of self-interactions in how we learn. By structuring training data in a way that teaches our system to talk to itself, we show that learning is shaped not only by the architecture of our AI systems but by the interaction dynamics embedded within our training procedures,” says Dr. Jeffrey Queißer, the first author and Staff Scientist within OIST’s Cognitive Neurorobotics Research Unit.
By combining self-directed ‘mumbling’ with a unique working memory architecture, the researchers improved how their AI models learned, adapted to new situations, and multitasked. This development follows years of interest in brain-inspired modeling for AI learning.
Brain-Inspired Modeling for AI Learning
The team at OIST has long been interested in content-agnostic information processing, which is the ability to perform tasks beyond the exact situations previously encountered by learning general methods and operations. Dr. Queißer notes, “Rapid task switching and solving unfamiliar problems is something we humans do easily every day. But for AI, it’s much more challenging.” This challenge has driven the team to adopt an interdisciplinary approach, blending developmental neuroscience and psychology with machine learning and robotics, among other fields, to find new ways to think about learning and inform the future of AI.
Initially, the researchers focused on the AI models’ memory architecture, examining the importance of working memory for task generalization. Working memory, akin to the short-term ability of a system to retain and use information, plays a critical role in tasks ranging from remembering instructions to quick mental math. By simulating tasks of varying difficulty, they examined the effectiveness of different memory structures. Their findings demonstrated that systems with multiple working memory slots, which act as temporary containers for pieces of information, could generalize better on tasks like reversing the order of and regenerating patterns.
Enhancing AI with Self-Mumbling
Upon adding self-mumbling targets—programming the system to talk to itself a certain number of times—the researchers observed improved performance, particularly when multitasking or completing tasks with many steps. Dr. Queißer emphasizes, “Our combined system is particularly exciting because it can work with sparse data instead of the extensive data sets usually required to train such models for generalization. It provides a complementary, lightweight alternative.”
This approach not only enhances the efficiency of AI learning but also opens up possibilities for developing AI systems that require less data to function effectively, making them more adaptable to real-world applications.
Learning to Learn Better
Looking forward, the researchers plan to introduce more complexity into their models. Dr. Queißer explains, “In the real world, we’re making decisions and solving problems in complex, noisy, dynamic environments. To better mirror human developmental learning, we need to account for these external factors.” This aligns with the team’s overarching goal to understand the neural basis of human learning.
By exploring phenomena like inner speech and understanding the mechanisms behind such processes, the team hopes to gain fundamental new insights into human biology and behavior. Dr. Queißer concludes, “We can also apply this knowledge, for example, in developing household or agricultural robots that can function in our complex, dynamic worlds.”
The implications of this research are vast, potentially revolutionizing how AI systems are trained and deployed across various industries. As the team continues to refine their models, the future of AI learning looks promising, with the potential to create more intelligent, adaptable systems that can seamlessly integrate into human environments.
