The ultimate programmable computer - the brain. Have researchers achieved the goal of AI and actually created a massively parallel computing device that can compete with the best?
Building a parallel computer based on organic components is a problem that has been solved for many years - just look at any life-form with a brain.
For years however the idea of building something similar starting from scratch and without the involvement of living organisms has been something that seemed both possible and impossibly difficult at the same time.
Now, an international research team from Japan and Michigan Technological University has created a similar process of circuit evolution in an organic molecular layer that can solve complex problems. It is claimed that this first time a brain-like "evolutionary circuit" has been realized and many news channels are reporting this as if we were close to a creating a brain in a box. The work is described in Massively parallel computing on an organic molecular layer, (Bandyopadhyay et al, 2010) published online on April 25 in Nature Physics.
This organic computer is claimed to be "massively parallel" but at the moment only changes corresponding to ~300 bits are performed which still has some way to go. The basic idea is to realise a 2D cellular automaton as a chemical system with varying numbers of connected neighbours.
Currently a chemical with four conductivity states is used in a bi-layer and is manipulated using a scanning tunnelling electron microscope (STM) which sets its conducting state by moving to a location and setting a voltage.Thus the layer can be "programmed" with an input configuration of states.
The molecules then tend to organise themselves into "circuits" according to which state is most common in a given area. Once written the patterns of states evolve according to a range of rules - the system has its own dynamic.
This makes it possible to build logic gates and possibly implement traditional computing methods in a new material. However the layer can also be programmed to simulate differential equations. The examples given are of diffusion and cancer cell development.
Of course all of this sounds promising and there is a great deal of scope for presenting what is going on as more than a chemical reaction that can be controlled by an STM in the manner of a cellular automaton. For example, as the molecules are self-organising based on their state it is tempting to say that the system has "self healing" properties, but this might be far fetched.
In the same way, to say that the layer is "intelligent" in any way is also probably the sort of hype that AI has long been prone to. Equally having the ability to build physical cellular automata doesn't necessarily mean we have any idea what to do with them or how to use them to compute something useful.
Magnetic resonance images of human brain during different functions appear on top. Similar evolving patterns have been generated on the molecular mono-layer one after another (bottom). A snapshot of the evolving pattern for a particular brain function is captured using Scanning Tunneling Microscope (scale bar is 6 nm). The input pattern to mimic particular brain function is distinct, and the dynamics of pattern evolution is also typical for a particular brain operation.
This research is an interesting development - building cellular automata as chemical layers is a valid endeavour - but the over-statement of the case, the use of emotive words such as intelligence and "self-healing", and the comparison of patterns of activity to patterns of real neural activation are the sort of hype that really doesn't help AI and has indeed harmed it greatly in the past.