Brain Evolution May Be Driven by Curiosity

New research suggests that curiosity may hold the key to understanding how our brains evolved to become so complex.

Scientists have been studying artificial neural networks to see if they can replicate the intricate processes of the human brain. In a groundbreaking experiment, researchers trained thousands of these networks with different learning styles and found that those which exhibited the most curiosity developed more sophisticated cognitive abilities.

These curious networks not only learned quickly but also spontaneously organized themselves into complex structures that resemble the interconnected modules found in human brains. The networks' neural connections formed densely early on before being pruned away, mirroring the developmental trajectory of human brain development.

Researchers at the University of Cambridge believe that curiosity may be a driving force behind this complexity rather than a consequence of it. To test their hypothesis, they conducted an eye-tracking study with 275 babies between 8 and 15 months old, exposing them to sequences of colourful shapes on a screen that provided varying amounts of information about where the next shape would appear.

The team's findings suggest that curiosity plays a crucial role in driving the development of complex brains. To test this hypothesis, they designed artificial neural networks with varying levels of intrinsic curiosity and trained them on 20 different tasks.

In these experiments, the high-curiosity models were programmed to switch tasks immediately if their learning rate dropped below a certain threshold, whereas the low-curiosity models changed tasks randomly. Despite this difference in approach, both types of networks became experts at the tasks they were trained on.

However, when tested on new and unseen tasks, the curious models outperformed their less-inquisitive counterparts by a significant margin. This suggests that curiosity may be key to unlocking complex cognitive abilities, such as the ability to combine disparate pieces of information in novel ways.

The researchers also observed that the curious networks developed physical characteristics reminiscent of those found in human brains, including efficient wiring patterns and brain-like features. By embedding these networks into a virtual three-dimensional space modelled on the anatomy of the human brain, they were able to simulate some of the key challenges faced by real brains during development.

The neural networks developed a complex structure in response to curiosity, featuring highly interconnected areas that were connected to each other in a hierarchical "modular" way.

This modular design was not the only unusual feature of the curious networks - their density of connections followed a specific trajectory, increasing and then decreasing over time. Researchers found this trend peculiar, describing it as an unexpected pattern.

The development of neural networks in humans also exhibits a similar pattern, with synapses forming rapidly during early childhood before undergoing a pruning process where unnecessary connections are eliminated. This process may be linked to the drive for curiosity, which encourages individuals to seek out tasks that provide valuable information and then eliminate less useful connections.

By comparing the growth patterns of these neural networks to human development, researchers estimated their biological age, finding that the most curious networks were at a more mature stage, equivalent to around 20 years old. In contrast, those without this combination of curiosity and correct wiring remained stuck at an earlier developmental point, similar to a 9-year-old's brain.

The combination of curiosity and specific brain wiring is crucial for driving complex cognitive development in humans. According to researchers, this unique blend enables individuals to learn more efficiently by focusing on tasks that are neither too easy nor too challenging.

Curiosity plays a pivotal role in shaping the architecture of human-like brains, as it allows people to prioritize learning opportunities that maximize their potential for growth and understanding. However, experts caution against assuming that non-curious networks lack complex cognition simply because they learn at a slower pace.

Preliminary studies suggest that even with extended training periods, these non-curious networks may not develop the same level of cognitive complexity as their curious counterparts. This raises questions about the effectiveness of traditional educational approaches, which often prioritize repetition and standardization over individualized learning experiences.

Researchers warn that environments that force children to focus on a single activity for an extended period may be counterproductive, particularly in early childhood development. By emphasizing the importance of curiosity and adaptability, educators can create more effective learning environments that foster cognitive growth and promote the emergence of human-like brain architecture.

Educators can create more effective learning environments by prioritizing activities that foster curiosity and adaptability in children.

This approach is also being explored in the development of artificial intelligence, where researchers are considering whether building curious AIs that learn like humans could lead to more complex and human-like cognition. Currently, most AI research focuses on scaling up systems that consume vast amounts of data and resources, but Poli suggests this approach may be limiting.

The idea is that by introducing elements of curiosity and flexibility into AI design, developers might create systems that are better able to navigate complex tasks and adapt to new situations, much like humans do.