Networks and Biology: Wiring ourselves into a bad theory

The one thing that can be said about networks is that they are easy to draw. Anyone who's done "join the dots", or who has looked at a map, or studied physiology or neuroanatomy understands networks in their essence: a set of points joined together with lines. The join-the-dots pattern permeates the natural world like a kind of fractal motif. But what we see and what things actually are, are not the same. How would we know if networks actually exist?  

In order to know whether a network is real, we would have to be able to establish some kind of correlation between our observations of the network's structure (which is "the network"), its behaviour, and any changes we might make to that structure. Obviously, if the network is human-made, then the relationship between an electronic  network's structure, how it behaves, and predictable outcomes in the light of changes to it would seem to be straight-forward. But in complex artificial networks, such as those defined by machine learning models, predictability in the light of network change is elusive. We are strangely unbothered by this, because we see the same type of unpredictability in natural networks. 

We may see in the brain an array of enormously complicated "networks", but beyond some very crude interventions which zap entire sections of "the network", there is little predictability in the effects of these interventions. So when we see little predictability in the AI webs our consciousness has made, we are inclined to imagine ourselves in the image of God, and satisfy ourselves that if fuzziness is good enough for our understanding of nature, it is good enough for our understanding of artificial intelligence. 

But this fuzziness should ring scientific alarm bells. Networks do not just spring from nothing. They emerge in nature from biological processes. To put it more directly, networks emerge from the dynamics of cells. Neurons are cells. Nerves are made from cells. Tree roots, fungal fibres and bacterial colonies are made from cells. The cell is the thing. The network is an epiphenomenon arising from the cell's behaviour.

This point is important when we think about our technology. If we designed our technology from the metaphor of the cell, rather than the metaphor of the network, we would have very different technology. And I am increasingly convinced that if we understood our existing networks with their mystical properties (like machine learning) from the perspective of cells, then their behaviour would be much less mysterious to us.

The main thing a cell must do is maintain a boundary between itself and its environment. It must maintain its internal environment and maintain balance with an ambiguous external environment, and it requires energy to perform these functions. It is through performing these functions that the cell establishes relations with other cells, from which the physical characteristics of a "network" might be seen to emerge. 

However, this mechanism must drive the cell through the processes of self-organisation with its environment. Networks are driven through a process whereby the cell seeks stability in its organisation relative to its environment. This can be achieved through absorbing features of the environment so as to adapt itself and organise itself into increasingly complex life forms - a process Lynne Margulis called "endosymbiosis". Increasingly complex life forms in turn provide the cell with increased adaptability in the face of environmental challenge. These processes of endogenisation and adaptation are the basis for the epigenetic mechanisms which are exciting increasing interest in current empirical biology.

But endogenisation and adaptation mean that history and time is embedded in the structure of the cell, and in the networks it forms. Biological networks - like neural networks - are more like scar tissue, or the scree lines fomed by geological events than they are simple nodes and arcs. At each stage of organisation, the cell must maintain homeostasis and balance with its environment; at each stage it tends towards the conditions of its initial formation - conditions which are historically embedded in its own structure.

This is the "network" science we need. It is not a science of networks at all, but of dynamic processes of maintaining boundaries at all levels of organisation, from the brain and the liver through to consciousness, communication, technology and education. Behind the rigid visualisations of network dynamics on Facebook, and the scree lines and scar tissue of individual biographies, and biological history. 

Looked at this way, the way we think about our networks of human communication are grotesque distortions of nature produced by a bad theory. Instead of cooperating and organising themselves, the bruised egos of individual nodes compete against one another, each node seeing to be the loudest or the best, or "clusters" of damaged souls reinforce pathological and explosive boundaries in politics. 

The basic point is that the homologue of the cell's  boundary wall is not the person's skin; it is a dialogue's boundary. At a human level, we organise ourselves through communication - that's where our boundaries are formed. However, when locked-in to the network technologies of social media, the boundary walls are reinforced against the environment - there is little endogenisation, and hence little growth and development. It is the homologue of the cancer cell. 

There is an urgent question in technical design: whether it is possible to create a dialogical technology which can reproduce the organisational processes of the cell, including its endogenisation of the environment, and its maintenance of self-organisation against an ambiguous environment. To do this requires a much less mystical view of nature and of things like machine learning. Such a view can be found if we jetisson our obsession with the network, and instead think about the commonalities between how we maintain our communicative boundaries, and how a cell does it.