Current artificial intelligence technology has hit a wall preventing it from achieving artificial general intelligence. Researchers are now revealing how adding a new dimension of complexity, inspired by the hidden wiring of the human brain, could push AI beyond its current limits toward human-like cognition.
Introduction: when technology hits a dead end
At the heart of what we call artificial intelligence lie artificial neural networks, a biologically inspired innovation designed to mimic the function of neural networks within the brain. Yet despite the incredible progress these networks have made, especially with the emergence of large language models like ChatGPT, they have hit something of a wall. Scaling laws—the idea that simply increasing data size and resources will make models better—no longer work as they once did. It has become clear that new innovations are needed to drive these models forward toward the ultimate goal: artificial general intelligence (AGI), the machine’s ability to think like a human.
The missing dimension: “height”
Some researchers now believe they know how to achieve this: AI design must be elevated to another dimension, literally. In a recent study, two scientists from Rensselaer Polytechnic Institute in New York and the City University of Hong Kong explain that current models already feature “width” (number of nodes in a layer) and “depth” (number of layers) parameters. However, they propose adding a third dimension of structured complexity they call “height”.
Dr. Ji Wang, one of the study’s authors, explains it with a simple example: «Imagine a city: width is the number of buildings on a street, depth is the number of streets you pass through, and height is how tall each building is». When you add height, in the form of internal wiring or shortcuts within a layer, you create «richer interactions between neurons without increasing depth or width. It mirrors how local neural circuits in the brain interact to improve information processing».
Simulating brain wiring
Wang and his colleague achieve this additional dimension by introducing two types of brain-inspired structures:
- Intra-layer links: These resemble lateral connections found in the brain’s cortical column, which is associated with higher cognitive functions. They form connections between neurons in the same layer, allowing them to directly influence one another.
- Feedback loops: Similar to recurring signals in the brain, where output affects input. Wang says this can improve the system’s memory, perception, and insight.
Wang says: «Together, these structures help networks evolve over time and settle into stable, meaningful patterns, much like how your brain can recognize a face even from a blurry photo. They enrich AI’s ability to refine decisions over time, just like the brain’s iterative thinking».
Surpassing the “Transformer” architecture
It was the “Transformer” architecture, introduced by Google researchers in 2017, that fueled the recent AI boom. Yet it has not achieved artificial general intelligence. This new study suggests that integrating intra-layer links and feedback loops can enable models to associate, reason, and refine outputs, making them smarter and more energy efficient.
One fascinating example is how feedback loops can create “phase transitions,” akin to how ice melts into water. Wang says: «For AI, this could mean the system shifts from vague or uncertain outputs to confident, coherent outputs as it gathers more context or feedback. These transitions could represent a point where AI truly “understands” a task or pattern, much like human intuition».
Understanding our brains better
Introducing more brain-like features into AI architecture will not only make it smarter and more transparent, but it could also help us explore the mysterious puzzle of how our human minds work. This application could provide a model for scientists to explore human cognition and investigate neurological disorders, such as Alzheimer’s and epilepsy.
Conclusion: a hybrid future
Although drawing inspiration from human biology offers benefits, Wang believes the future of AI lies in the coexistence of neural architectures and other systems, perhaps even quantum systems. He says: «The optimal solution will likely lie in hybrid designs, borrowing from nature and from our imagination that transcends nature».
