Alan Turing, who probably got there first no matter how exotic your approach to artificial intelligence, once had the idea of "unorganized machines". He was thinking of possible ways that the initial neural networks might form a newborn baby's brain. One of his ideas was a collection of initially random logic gates that cold self-organize or be trained for particular tasks over time. He saw this as a possible approach to realize intelligent machines. We could implement such an idea today in hardware or software but what about using chemistry? This is exactly what researchers at the Unconventional Computing Group at the University of the West of England are doing. As described in their recent paper, "Toward Turing's A-Type Unorganized Machines in an Unconventional Substrate: A Dynamic Representation in Compartmentalised Excitable Chemical Media" (PFD format):
Collision-based computing exploits the interaction of moving elements and their mutual effects upon each other’s movement wherein the presence or absence of elements at a given point in space and time can be interpreted as computation. Collision-based computing is here envisaged within recurrent networks of BZ vesicles, i.e., based upon the movement and interaction of waves of excitation within and across vesicle membranes ... A-type unorganised machines can therefore be envisaged within networks of BZ vesicles using the three-vesicle construct for the NAND gate nodes, together with chains of vesicles to form the connections between them.
The BZ (Belousov Zhabotinsky) medium is a chemical concoction of sulphuric acid, sodium bromated, cyclohexadione, and a few other chemicals, the result is pictured above. Think of it as a collection of bubbles that form something like neural networks where the signals are waves passing through the points where the bubbles touch, forming logic gates and other types of circuits. Researchers have described lots of common logic components including AND, NAND, NOR XOR, inverters, adders, and more. They've formed memory circuits and other more complex circuits. An interesting overview of the logic gates can be found in a set of slides from the talk, Neural Isomorphisms of Adaptive Belousov Zhabotinsky Encapsulated Vesicles (PFD format). So who knows, instead of robots with positronic brains, we may end up with robots who have chemicals sloshing around in their heads! (and does BZ remind anyone else of the Mathmos from Barbarella?)