In an unprecedented move this year the Nobel selection committee highlighted the foundational contributions of John Hopfield and Geoffrey Hinton to the development of machine language and artificial intelligence, or AI, for the physics prize. 

If that were not enough, the Nobel committee for chemistry singled out the work of Dennis Hassabis and John H. Jumper for their development of powerful software tools for unravelling the relationship between the linear sequence of amino acids in proteins to their 3D structure with a high degree of confidence.

David Baker developed similar powerful software, which led to the development of potentially useful novel proteins, unknown in nature.

Laureates and others in the field of machine learning devices like to refer to their computers as modelled on the brain.

For example, it’s not uncommon for simplified neural networks modelled after the brain to be illustrated by interconnected nodes, likened to neurons and their neural connections and layers of nodes to similar layering found in the brain. 

The Nobel committee highlighted this similarity between computer networks and the brain’s neural networks in its summaries of the Nobel Prize-winning studies.

The popularity of this analogy is one of the reasons why the term “neural networks” is so often used by experts in the field. However, the analogy leaves me wondering whether it is misleading.

For much beyond those highly simplified illustrations, readers and students are trying to understand just how neural networks actually work. 

Recently a map tracing every nerve cell and fibre in the brain of the house fly was completed leaving anyone looking at such a complex connectome struggling to figure out the functional relationships of all those cells and connections.

And the tiny fly brain turns out to have only 140,000 nerve cells, a drop in the bucket compared to human brains. 

The human brain is incredibly more complex because of the estimated 100 billion neurons, many specific in shape and function, and the vast network of neurons they are connected with, to say nothing of the many trillions of connections in the brain.

Imagine the challenge, then, of mapping every neuron and connection in the human brain and functionally making sense of everything.

That complexity is why most neurophysiologists have stuck to subsystems such as memory, vision, hearing, smell, position sense, the motor system and the spinal cord, which though complex enough, are much simpler to study than trying to unravel the mysteries of consciousness and awareness. 

There’s another issue with using the brain as an analogy for AI and machine language because while many areas of the brain are layered or highly organized in some manner, they usually behave in highly coordinated fashion as happens, for example, with walking.

Walking seems simple enough, but walking involves exquisitely timed coordination at many levels in the brain, from the neocortex to the basal ganglia, cerebellum, brainstem and spinal cord, which systems also incorporate feed forward and backward circuits to fine-tune the activity.

Walking may look simple — but it is not. 

What we can say is that today’s machine language and AI systems are capable of learning from vast databases — well beyond the capacity of any human or group of humans to achieve and identify novel relationships and even develop or acquire algorithms, which offer paths for analyzing the data in ways their human creators did not foresee.

That’s the advantage of modern-day self-learning programs, which are capable of analyzing huge databases and what makes them perfect for analyzing weather and climate data and, in 2021, for translating linear sequences of amino acids characteristic of proteins into 3D constructions of what they actually look like and hints about how they might function. 

That was the triumph of Dennis Hassabis and John H. Jumper’s AlphaFold 2 program for deciphering proteins and the same for David Baker’s RoseTTAFold, which has similar capabilities.

Baker’s team has gone on to create novel proteins not seen in nature, but which have practical applications in health care.

Both groups released their programs for free to scientists around the world and already well over a million scientists employ one or both programs with amazing success. 

Despite their success, these programs aren’t perfect and there are some things they cannot do so far — such as show us how the shape of proteins changes when they interact with other proteins or perhaps viruses.  

Nobel committees don’t always get things right, but this time they made a spectacular, if controversial, choice.

Dr. William Brown is a professor of neurology at McMaster University and co-founder of the InfoHealth series at the Niagara-on-the-Lake Public Library.