Computer vision and the bicameral mind
A while back a friend suggested that I read Julian Jaynes’s “The Origin of Consciousness in the Break Down of the Bicameral Mind”. I thought rather condescendingly to myself that while I was at it I might as well read “Zen and the art of motorcycle maintenance”. The Jaynesian offering was put at the bottom of my reading list. However, while going through “The God Delusion” by Richard Dawkins, I came across the statement that it is hard to tell whether or not Julian Jaynes’s hypothesis is utter crap or possibly the most profound idea of our time. Feeling that this was like an ironic sign from above, this passage set me off to find out what all this bicameral business was about.
The book starts off by arguing that much of what we view as our daily mental life has nothing to do with consciousness. Consciousness is not necessary for learning, reasoning, concepts or even thinking. Consciousness is an analog for the real world. It maintains a metaphorical mindscape, it creates an analog for “I”, it constructs a narrative of our lives and it assimilates objects into a familiar hierarchical schema. All these concepts are linguistic in nature and here is the kicker: this implies that language predates consciousness. Jaynes then hypothesizes that shortly after the advent of language our minds were split in two, an executive part called a god and a follower part called a man. The god part provided voices by which the man part is guided. Arguably this allowed for much greater levels of societal organization, where people could be kept on task without the need for direct supervision and control. However these societies eventually crumbled under their own weight and what emerged from the chaos is the consciousness that we know today. In many ways our literature can be viewed as pointing to this series of events. The Iliad – where the gods are the puppet-masters of the Trojan War and the Odyssey – where the wily Odysseus rejects the gods and says that we can think for ourselves. The story of Adam and Eve can be easily interpreted in this light.
Critics have maintained that such transformations would require mental rewiring on a time-scale that evolutionary biology simply cannot support. However one might argue that our brains are so hierarchical in nature that rewiring can be triggered by a variety of mechanisms that result in jumps that appear to be highly non-incremental. We see this time and time again with victims of traumatic brain injuries. Schizophrenia hints at this. One can also argue that humanity, unlike any other species, can undergo forms of hyper-selectivity due to our unique propensity towards genocide on massive scales. Jarred Diamond points to this through his incredible book: “Guns, Germs and Steel”. It has been noted that the average troop of chimpanzees exhibits more genetic diversity than the entire human race. Our genes are not only greedy – they are nasty.
Whatever one thinks of the Bichameral hypothesis, one must accept the view that the minds of today are different from those that existed a thousand years ago and they may be profoundly different from those that were in existence ten thousand years ago. What does this mean for the friendly computer vision researcher? Simply put trying to build, from scratch, mechanisms that attempt to mimic complex evolved systems without knowledge of their intermediary states and catalysts is really really hard. In “Out of Control”, Kevin Kelly documents how attempts to reconstitute stable prairie grasslands invariably end up in parking lots filled with weeds. So what is the way forward? Maybe instead of all of us trying to build the same kind of recognition algorithms all racked and stacked via ROC curves measured against standard datasets, we should each go our own way for the purposes of building a diverse stock of computer vision and AI ideas, with the understanding that only through symbiosis can we escape the trap of incrementalism. While we may no longer need our appendix or the god-part of our minds, if they did not exist at some point in time, we would not exist as we do today. To this end I point to the following video that uses a technology that could be added to the stew. It is based on Yan Tong’s ideas on congealing, where thousands of images can be annotated with landmark points based on a few manual examples.