One component of the singularity is when man humans are able to build computers as powerful as the human brain. Some also think of this point as when we get artificial intelligence (AI) up to the level (and beyond) of the human brain. Many approaches to AI have been largely software driven (top down) with the thought that the we don’t have to understand the biology of the brain. The difference between a rat and us is that we have more of these columns and that the connections between them are more complicated and diverse (see )
Ray Kurzweil, and others, running with Moore’s Law predicted that by 2013 we will reach a point of computer power able to run human brain functional simulation (see here (2001 essay), and his book, The Singularity is near). Further on, by 2025 there will be supercomputers capable of ‘uploading’ human brains (only 17 years from now).
The video features Henry Markram, from the blue brain project, giving a talk at MIND08. What makes this talk interesting is instead of taking a top down approach to obtaining human level intelligence this group is taking a bottom up approach (ok – if the cortex is considered the bottom). They are trying to “create a biologically accurate functional model of the brain”. To accomplish this they are first reverse engineering the brain – database this (500 petabytes) – then forward engineer this into mathematical models where they can run brain simulations on IBM supercomputer hardware. The piece they have started out, which they think is the most strategic, is the neocortical column. These neocortical columns he refers to as our microcircuits that enable us to have our higher cognitive functions. The difference between a rat and us is that we have more of them and that the connections between them are more complicated and diverse (see this piece regarding the small-network properties of the human brain). Each column is made up of 10,000 neurons (400 different ‘types’) with 30 million connections.
They expect to have a full scale human brain model within 10 years. Markram says they will need a computer 20,000 time more powerful (exoflop ability) than any computer that exist currently (2008) along with 200 petabytes of memory. The 2018 date of full scale human brain model puts us about 5 years behind Kurzweil’s 2001 prediction (not sure about his updated 2005 predictions in ‘The singularity in near‘) – but reasonably close. Of course Markham is also only predicting that his team will be able to accomplish this goal within 10 years.
Some of the visuals of this video are quite impressive.
We live in interesting times, and interesting near-future times.