Post Lecture Recap: Future of the Brain

Last night, I headed down to the beautiful Bell House in the Gowanus section of Brooklyn for the latest Secret Science Club lecture, featuring Cognitive Scientist Dr Gary Marcus of NYU.

Dr Marcus began his lecture by displaying President George H. W. Bush's 1990 Proclamation 6158, declaring the "decade of the brain". He then noted that, sadly, little has been done to follow up on this lofty declaration. Twenty-four years after the proclamation of the "decade of the brain", we still can't reliably use brain scans to diagnose such mental illnesses as autism and schizophrenia.

Regarding the biology of the brain, there are many unknowns, the basis of short-term memory, the full role of Broca's area, the mechanisms behind the brain's ability to process sentences- all are unknown. How does the brain "decide" which part performs which functions? What changes to the brain are made in the formation of long-term memories? Is the brain "analog" or "digital"? Can we even abstract brain function?

There is no serious Theory of Brain, despite the fact that a lot of researchers are doing a lot of research on various related topics. Regarding the development of Artificial Intelligence, Dr Marcus quipped that it is always "20 years in the future", whereupon some bastard in the audience likened it to workable fusion power. The core problems of AI remain unsolved: machine reading is stuck at a 4th grade "D" level, sketch understanding is extremely limited, there is no genine language understanding, there is no "common sense" reasoning. Despite the advances that have been made, there is not a lot of real progress.

Regarding the lack of progress in the development of artificial intelligence, Dr Marcus joked that there is a misguided search for "silver bullets", but there are no "three laws to put on a T-shirt". He then catalogued some of of the various models that were put forth- Parallel Distributed Processing, Neural Networks... the model that is currently generating excitement is Deep Learning. Dr Marcus noted that it is easy to fool "deep learning" systems- if an artificial intelligence is trained to recognize one thousand items, item number one thousand and one will stymie it. Such is system is limited, it only functions within a "closed world".

Whole brain emulation, while seemingly a great idea in the long term, is unrealistic in the short term- scientists currently can't model the 302 neurons in a worm's brain, much less the 86 billion neurons in a human brain. The quest for a single "Canonical Cortical Computation" model is in its early stages- a single common principle is hoped for.

Regarding the structure of the brain, even though there are considerable differences in the various parts of the brain, scans of these areas appear to be similar. There is no satisfactory account for what a "canonical circuit" might be- Dr Marcus quoted noted scientist Bono, we "still haven't found what we're looking for". There is no reason to think the brain is simple- complexity is found at every scale.

Dr Marcus prefaced the next part of the lecture with a quote from pioneering neuroscientist Santiago Ramón y Cajal:

Unfortunately, nature seems unaware of our intellectual need for convenience and unity, and very often takes delight in complication and diversity.

He likened evolution to a tinkerer, fiddling around with spare parts- it is difficult to reconcile a canonical circuit model with developmental, molecular, or evolutionary biology... the biology of the brain is haphazard. The traditional view of the brain, proposed by the visual cortex researchers Hubel and Wiesel, is one of a hierarchy of features. In a simple visual system, cells would differentiate between light and dark, with more complexity, a simple line could be distinguished, eventually, a right angle would be perceived. In reality, single neurons can be stimulated by very specific visual input- the infamous Jennifer Aniston neuron.

The lecture then shifted to the topic of making brain maps. Dr Marcus opined that neuroscience has a sorry history of attempting to explain the brain using metaphors drawn from the latest technology. The brain is not a hydraulic system, nor is it a hologram, nor a computer- one cannot download a brain app. The brain performs as many "devices" acting at the same time.

The computation that occurs in the brain is massively parallel- the brain is not subject to the limitations of Von Neumann architecture, it is probably more like a Field Programmable Gate Array in which many blocks can be configured to do different things. Although the structure appears homogenous on a superficial level, it is customizable in order to perform many tasks. While the anatomy of the brain is fairly uniform, the brain is "tuned" by experience- "nurture" is an important factor in brain structure. There is a lot of parallel "architecture" in the brain to integrate the many "computational blocks" needed to perform the brain's myriad functions.

Gene expression differs across the cerebral cortex, but the closer two parts of the brain are to each other, the more closely they configure- there is no grand principle of the brain, there is variation across the brain. Evolution reconfigures its toolkit over time. If synthetic biology progresses to the point in which a synthetic brain can be "wired" according to code, the computational blocks could be configured in customizable ways.

Once again, the Secret Science Club delivered a fantastic lecture. In the Q&A, some bastard in the audience brought up the subject of plasticity in the brain, and Dr Marcus reiterated the importance of experience "tuning" the brain. For a taste of the Secret Science Club experience, here's a video featuring Dr Marcus:





Pour yourself a libation and drink in the SCIENCE!