Skip to main content


Brain development could hold lessons for building better artificial neural networks

The Self-Assembling Brain: How Neural Networks Grow Smarter

Peter Robin Hiesinger
Princeton University Press
384 pp.
Purchase this item now

Despite their overlapping interests, it is rare for developmental neuro­biologists to consult artificial intelligence (AI) experts in the course of their research and vice versa. But in his new book, The Self-Assembling Brain, neurobiologist Peter Robin Hiesinger argues that doing so would likely be of great benefit to both parties. In 10 chapters, he describes a series of imagined conversations between four hypothetical individuals—a developmental geneticist, a neuroscientist, a robotics engineer, and an AI researcher—that offer readers insight into the information that is needed both to understand the workings of the brain and to create an artificial system that mimics the brain. These fictional conversations are followed by “seminars” in which the author discusses specific topics in greater detail.

Hiesinger elegantly moves through a variety of topics, ranging from biological development to AI and ending with a discussion of the advances that deep neural networks have brought to the field of brain-machine interfaces. At each stage, he discusses the commonalities and differences that define developmental science and AI research.

I appreciated, in particular, Hiesinger’s efforts to link information on different levels of abstraction, such as when he describes why linking genetic information to human behavior is complicated and reveals the caveats of attempting to do so. He uses this discussion to prompt the reader to ponder at what level we need to work to build an artificial brain, a topic on which he elaborates later.

In the seminar “From Algorithmic Growth to Artificial Intelligence,” Hiesinger describes differences that exist in the training of artificial neural networks (ANNs) and the early development of brains, raising questions about the way we train ANNs. The network architecture and training rules are fixed during ANN training, and the network learns by exposure to large datasets. During human development, however, the brain changes as it grows, and the learning rules change too.

The hypothetical dialogues that open each chapter successfully capture various disciplinary perspectives and illustrate tensions between the fields represented. In one passage, for example, Minda, the developmental geneticist, argues that genes contain all of the information necessary to build an artificial brain, whereas Pramesh, the AI researcher, maintains that this level of description does not translate into a principle he could program into an artificial system.

Such conversations acknowledge the limitations of the respective approaches taken in different fields, identify different naming conventions for related concepts, and convey the enthusiasm many researchers feel when learning about other fields. However, I wish that the views that were expressed had given readers a better sense of how new inter­disciplinary research collaborations might be established.

The book’s last discussion and the accompanying seminar will be of particular interest to researchers who work at the interface between neuroscience and AI. Here, Hiesinger argues that in order to achieve AI comparable to human intelligence, we cannot take any shortcuts and must include information present on every biological level—a position that may be seen as controversial by both neuroscientists and AI researchers. He acknowledges, however, that the path to reaching artificial general intelligence may not necessitate humanlike intelligence.

One shortcoming of this otherwise entertaining book is that it tries to cover many concepts but often fails to provide sufficient detail for the reader to fully grasp a specific issue. However, given that Hiesinger aims to engage a variety of people from different backgrounds, some context must inevitably be sacrificed to keep the narrative moving­­­­. Overall, this book accurately illustrates current debates occurring in this space and is likely to inspire future discussions at the intersection of neuroscience and AI.

About the author

The reviewer is at the Department of Psychology, University of Cambridge, Cambridge CB2 3EB, UK, and the Center for Brains, Minds and Machines, Massachusetts Institute of Technology, Cambridge, MA 02139, USA.