Artificial Neural Networks to Understand the Functioning of the Mind

  • ML cube
  • Insight
  • Artificial Neural Networks to Understand the Functioning of the Mind

Until a few decades ago, the study of the mind and human behavior were essentially the prerogative of psychology and philosophy.

Although it is possible to identify the birth of neuroscience around the 5th century BC, the modern discipline flourished in the 1960s making a great contribution to understanding the biological basis of learning and memory (1).

Without going into detail, we can say that neuroscience, in tracing the general features of the discipline, often relies on terms such as “perception” and “memory” in the analysis of brain functions.

We know for example that the Subjective Visual Vertical (VVS) supports visual perception while the medial temporal lobe is involved in complex interrelated behaviors such as memory and learning (2).

However, using this knowledge alone to describe and categorize neural processes is not enough.

After decades of research, the relationship between the perceptual and mnemonic systems remains an open question.



In their recent publication, PhD student Tyler Bonnen and Professor Daniel Yamins of the Stanford Institute for Human-Centered Artificial Intelligence (HAI), tried to unravel the question using modern AI-based computing tools (3).

The research team proposed a solution based on recent advances in the field of Computer Vision based on computational models capable of predicting the neural response of the visual system of primates, namely the Convolutional Neural Network (CNNs): a type of artificial neural network feed -forward in which the connectivity patterns between neurons are inspired by the organization of the animal visual cortex (4).

In their studies, Bonnen and Yamins initially collected 30 previously published experiments and, using the same stimuli (same images, same compositions, same presentation order, etc.), they determined the degree of accuracy of the model they developed.

Bonnen stated that the framework is able to predict the behavior of subjects with medial temporal lobe injuries while healthy subjects are able to far surpass that model: this demonstrates the clear implication of MTL in those who have been long defined as perceptual behaviors thus unraveling decades of apparent inconsistencies.

Regarding the relationship between MTL and “perception” instead, Bonnen states that “although this interpretation is entirely consistent with our results, we don’t care what words people should use to describe these MTL-dependent abilities. We are more interested in using this modeling approach to understand how MTL supports behaviors ”, underlining how, unlike previous studies, there is not a substantial theoretical advantage, but rather a methodological one.

“We have demonstrated a great proof of principle: these biologically plausible computational methods can help us understand neural systems beyond canonical visual cortices,” concludes the scholar.



To better understand the functioning of the neural networks of our mind, a great contribution comes from the publications of scholars Bonnen and Yamins from which the centrality of the Medial Temporal Lobe (MTL) emerged in all those defined as perceptual behaviors.

Artificial neural networks and in particular Convolutional Neural Networks have therefore allowed a great step forward towards a deeper understanding of the functioning of the neural networks of our mind.