Post date: 16-Feb-2012 23:41:51

The Engineering discipline gains tremendous power by taking a reductionist, modular approach. Complex engineering problems are recursively broken down into smaller and smaller component problems with well defined interfaces. Each of the small component problems are solved in turn, and then they are all "bolted" back together into the engineering solution.

However, I think that the traditional Engineering approach is ill suited to constructing an intelligent device. The structure of engineered artifacts are starkly different to the structure of organisms which have grown.

When we use the traditional engineering approach to construct an artefact, the resulting structure emerges in spite of the environment. When an organism grows, the structure emerges as a result of the environment. For example, when we engineers build a bridge, we first construct foundations which act as the interface with the environment. The foundations provide a (hopefully) solid and unchanging base on top of the, typically irregular, surrounding terrain. The construction of the remaining structure then proceeds, hopefully resulting in a bridge which closely reflects the original plan. The environment, the wind, rain, temperature, tides, bacteria, bird migrations etc hopefully have no influence on the outcome. The growth of an organism is encoded in DNA, but dependent on, and influenced by, the environment in all stages of its development.

You could proceed by adding meta-rules ad-infinitum or "... make rule like behavior emerge out of a multi-level bubblingbroth of activity below. This means that you give up the idea of trying to explicitly tell the system as a whole how to run itself."

Hofstedter in Metamagical Themas

It is not at all clear that it is even possible to decompose biological systems in an engineering sense.

There has been a simply tremendous quantity of research done into neuroanatomy and cognition – the architecture and operation of brains. Research is ongoing and with increasingly sophisticated techniques. Overwhelming complexity abounds at every level in a brain; chemistry, cellular, neurological, psychological and social. There is most certainly structure in a brain, but that structure is inextricably intertwined as a consequence of evolution.

By way of example, consider the spiny lobster; a crustacean which has become a very popular experimental invertebrate. Lobsters swallow food whole and then rhythmic contractions of the stomach and gastric teeth macerate the food. These rhythmic contractions are controlled by a group of 11 neurons called the stomatogastric ganglion. These 11 neurons and their connectivity were identified in 1974, but the basic mechanisms by which this ganglion produces the rhythmic action has not been understood until very recently. (Selverston et al, 2009). The behaviour of even this tiny group of neurons has been extremely difficult to unravel.

A human brain contains in the order of 1012 = 1000,000,000,000 neurons and with 1000 times as many connections between neurons. Furthermore, there are many different types of neuron and their connectivity is far from uniform. The chemical and electrical working within a single neuron also has a very high degree of complexity. There has even been a serious suggestion that quantum effects play a critical role in a brain's operation (Penrose 1994) Finally, the connectivity of the entire network is in constant flux as new experience is incorporated.

Perhaps, the most obvious approach is to treat neurons as the modular components. However, as we have seen, there are 1012 neurons in a human brain, each neuron is a complex device in it's own right, and the connectivity between neurons is anything but simple. The dynamics of a network of only 11 lobster neurons would appear to severely challenge the modular approach. Dividing brains up into larger components, reduces the number of components but increases the connections between them.

"The human genome ... does not specify the entire structure of the brain. There are not enough genes available to determine the precise structure of everything in our organisms, least of all the brain, where billions of neurons form their synaptic contacts."

Antonio Damasio in Descartes Error

"In a sense, the morphogenetic field is the nub of the answer to the riddle of brain evolution: complexity is most often derived from simplicity, that is, great diversity and great complexity have arisen because they both are merely the result of a few, simple random mutational events that affect the behaviour of particular morphogenetic fields, the phenotypes of which have been favoured highly by natural selection."

Butler & Hodos in Comparative Vertebrate Neurobiology

If intelligence is necessarily based on structures such as brains, then we need different techniques for intelligence engineering.

An alternate approach may be to seek a bottom-up technique inspired by replicators, DNA, random mutation, sexual crossover, competition and natural selection in an environment. However, the success of these mechanisms has been reliant on a great deal of time and truly astronomical numbers. Consider that the current population of insects on earth has been estimated at more than 1019 and bacteria of more than 1030. These tiny organisms are replicating every few days (insects) or hours (bacteria). Most of these replications place a DNA variant into the environment which may be naturally selected, or not. Bacteria have been operating like this for well over 2 billion years and the ancestors of their replicators began well before that. Furthermore, each organism is a unique result of their particular evolutionary history – rather difficult to replicate!

A naïve bottom-up approach would appear to be well beyond our means. However, we recall that evolution by natural selection is a completely 'blind' process where-by some gene variants replicate successfully and others simply do not. We, on the other hand have a goal – an intelligent device – and successful examples.

It would seem to require a step back to concentrate on specifying the critical junctures and to allow he rest to 'look after itself'. It is a more 'organic' approach focussed on defining the growth process and influencing the growth environment.

Butler, A.B. & Hodos, W. (2005), Comparative Vertebrate Neuroanatomy - Evolution and Adaption. Wiley.

Churchland,P.S. and Sejnowski,T.J. (1992), The Computational Brain. The MIT Press.

Damasio,A.R (1995), Descartes' Error : Emotion, Reason, and the Human Brain. Harper Perennial.

Hofstadter, D. (1996), Metamagical Themas: Questing for the Essence of Mind and Pattern. Basic Books.

Koch,C. (1999), Biophysics of Computation. Oxford University Press.

Penrose,R. (1989), The Emperor's New Mind. Vintage.

Penrose,R. (1994), Shadows of the Mind. Oxford University Press, Oxford.

Selverston,A.I., Szücs,A., Huerta,R., Pinto,R., & Reyes,M. (2009), Neural mechanisms underlying the generation of the lobster gastric mill motor pattern. Frontiers in neural circuits.