How close are we to replicating brain tissue in synthetic form? (Extended abstract)

Dennis Bray, Department of Physiology, Development and Neuroscience, University of Cambridge

Brain vs. Machine

Over the past 50 years, sci-fi writers and other visionaries have been predicting the appearance of humanoid robots able to walk, talk, and act like people. Over the same period of time our knowledge of the brain and understanding of how it acts has been transformed. So, as the singularity approaches, it seems reasonable to ask how close are we to replicating brain tissue in synthetic form? What are the implications for the future?

If we ask: “How much and to what degree can artificial machines perform the same (mental or intellectual) functions as human beings?” there is every reason to be optimistic. The list of what computers can do seems endless and it is far harder to think what they cannot do. Many limitations come under the category of ‘everyday activities’ or ‘common sense’. Understanding speech and maintaining a conversation is still a problem, as is human level vision. It will be some time before a robot plays a good game of tennis or cooks a gourmet dinner from scratch. But given the phenomenal progress to date and the way intelligent machines just get faster and smarter every year, one would be rash to declare that any particular function is permanently beyond reach.

But if we ask: “How closely do the workings (that is, the internal structure and mechanism) of automata resemble those of the human brain?” the response is more problematic. Most robots and intelligent machines are designed with efficiency in mind. They represent the best solution to a problem based on current engineering practice, which is rarely if ever the solution discovered by biology. It is true that bio-inspired computing has produced neural networks, genetic algorithms, cellular automata, and other algorithms that mimic some natural process. But even a cursory examination of any of these applications shows that the resemblance to anything biological is purely superficial.

There remain explicit attempts to reproduce the brain solely from the standpoint of circuitry such as the grand challenge declared by the National Academy of Engineering to ‘reverse engineer the brain’ and Henry Markham’s Blue Brain project that seeks to build a human brain in 10 years. These ventures have received enthusiastic support from physicists and engineers but to many biologists they seem overblown and unrealistic. It seems fatuous to represent a neuronal synapse—a complex, structure containing many hundred different kinds of protein, each a chemical prodigy in its own right—with a single line of code.  Such a model could never incorporate the synthesis and turnover of crucial molecules or the dynamic growth and shrinkage of synaptic structure that accompany learning.  Nor would it include the local synthesis of synaptic proteins, the modulating effect of microRNAs, the influence of glial cells, diffusing hormones, oxygen, and blood flow. Most crucially, no simplistic computer representation could ever to match the anatomy and physiology of a living brain, which requires years of development and learning to mature.

In summary: I agree that intelligent machines could be capable of at least reproducing human performance in any specified intellectual task within a decade (the outstanding exception to this statement being sentience or awareness, which no one has any idea how to install into a machine). But as to the question of reproducing a functional human brain, I think we are still in the dark ages. We probably do not even have suitable tools to replicate biological ‘hardware’, with its multiplicity of forms and continual dynamic turnover. And we certainly have no idea how to connect it up correctly. Until we can install something equivalent to development and learning, then any computer network we build, no matter how large and fast, will be empty of function—just another electric doll.

But perhaps it will not matter. The next fifty years are going to be very interesting and many take the view that evolutionary progress is transitioning from humans to machines, whether we like it or not. Seen from this vantage point, modeling every last detail of our brain’s performance becomes less important—a side issue in the grand sweep of things.