Systems in Nature. Why they have higher plasticity & innovative capacity, why efficiency is an organisational tuning parameter
- Adam Timlett
- May 14
- 5 min read
Stance Sketch, James Robert White www.jamesrobertwhiteart.com
Download timestamped PDF for citation
Managing risk, both opportunity and adversity is about optimising for both for what you can produce now, and what the future may require, and this can look messy.
In biology, this is about producing the optimal mixture of both finished products, and also optimising their parts for use in different future products the organism may expect to need.
In a sense, this is similar to a linear programming problem where we optimise a mixture of products in a factory for maximum profit or for min cost, like how many denim jeans we make versus denim shorts and jackets. In an organism it might be the use of certain enzymes to produce products that go into different kinds of nutrients that the organism can synthesise.
The difference is that this is explicitly about risk, not just optimisation, as here, more uncertainty about what product to make favours a higher proportion of investment in parts with multiple uses, over finished products. This openness to new uses and 'affordances' of parts is the hallmark of a more adaptive system. This can be important in developmental processes where similar processes can lead to different results and organisation, such as cellular differentiation and different gene expression in different parts of the body, as the organism develops.
But Nature goes further, and also produces unfinished 'parts of parts', or bits of parts, (note the use of the term 'bits' here just means below 'parts' level, so is dependent on perspective of a model that 'sees' certain things as useful parts). This means Nature makes even fewer assumptions than we generally make in software architecture about what we expect to be useful parts in future products. The finished products that result are more down to chance and exogenous processes that act like natural selection to remove some "bit results" like damaged cells, or cells that are incorrectly differentiated, and positively select others, like useful antibodies or cells that are usefully differentiated. An obvious example of the value of this variation is the production of a huge variety of antibodies in the immune system given uncertainty in what pathogens are in the body, but this is actually a pervasive process, also important in developmental biology. Mark Isalan (reference below) argues that instability, which he describes in his paper in his discussions of the liar paradoxes in gene regulatory networks is very productive, and this is a variation of this argument focused on networks, but he argues it applies not just to gene regulation, but also can be useful in other key processes.
So, we can summarise these different levels of production and variation in terms more familiar to computer scientists and architects.
A system that optimises at the top level, just products we need now, is mono-modular and doesn't optimise for modular parts. It's a monolithic system. In the worst case it can result in a 'ball of mud' which is hard to adapt, because the parts are not pairwise disjoint sets. It is only good as the current superset, the current product, parts are hard to re-use.
A system that optimises at the top two levels, both for products and useful parts, is multi-modular. It parts are pairwise disjoint sets which can make both the current superset and other sets of these sets by re-using the parts in different ways. The investment in reusable parts is often partly at the expense of the current superset, the current product. This is where architecture comes in to deal with the trade-off.
A system that optimises at all three levels, current products, useful parts and useful bits (bits of parts), is distributed meta-modular. Its parts are not pairwise disjoint sets, but it can be optimised to reuse parts due to the instability and rivalry in use of bits by different parts.
It can produce a far greater variety of parts from overlapping bits. This means it can be more plastic and more agnostic about the future requirements, like the immune system's production of antibodies or instability in gene regulatory networks leading to more variety and potential organisational forms in development.
Optimisation for bits of parts entails type instability and bit rivalry. It means parts are no longer pairwise disjoint sets.
Changes lead nor just to novel parts reuse, but breaks apart parts to recover bits from liquidated parts. Implementing it tends to mean a generalisation of Natural Selection is occurring. It means an exogenous system is selecting or removing bits forming new potential products that are produced in a partial non-deterministic way by the instability in parts/types. The obvious example is the immune system.
This production of variation in parts is efficiently achieved by over-compression of models leading to instability in types and bit use partly described by Mark Isalan, (again reference below) in his discussions of the liar paradoxes in gene regulatory networks.
It also occurs in the 'liminal' zone at the limits of the control of sub-systems, again, favouring production in this liminal zone, over just production in the zone of control. This favours over-compressed models that lack control. Perhaps that's why artists favour being a little out of their depth in order to be more creative.
This selection and pruning of such unstable results is partly achieved in biological systems by cell recycling, i.e. apoptosis (programmed cell death), and damage and repair systems. Selection is achieved by local fixing effects by exogenous systems within the organism. These effectively create a generalisation of Natural Selection within the organism.
A high level of variety in parts is produced without complete control, and fixation occurs of viable results, e.g. the developmental processes described by Turing Patterns as seen in the coat of tortoiseshell cats in which rival alleles compete and there are different winners in different regions, creating the tortoiseshell pattern. Also seen generally in developmental biology which leverages this instability as Isalan describes.
The whole mixture of finished products, parts, and unfinished novel potential products is tuned by the level of efficiency, precision and optimisation of the sub-systems.
The resulting levels of compression and optimisation in sub-systems manage both emergent organisation and the mixture of finished and unfinished products and novel potential products by sub systems, via levels of bit rivalry permitted by compression demands and constraints.
Efficiency is like a volume control for a music system. Max is not best for the sub-systems because we are managing both risk in terms of opportunities and risk in terms of adversity via the mixture of bit rivalries that we allow. If we max efficiency we just exclude useful instability and bit rivalry from the emergent organisation. It's like going to the shops without doing the recycling as part of the trip. The recycling is like a rival model or task to the 'bits' allocated to going to the shop. Going to the shop with maximum efficiency also changes the emergent organisation. Limits to efficiency are that you are changing the organisation that emerges.
A notable tuning strategy is over-compression of sub-systems for production of useful instability.
References
Mark Isalan. Isalan M. Gene networks and liar paradoxes. Bioessays. 2009 Oct;31(10):1110-5. doi: 10.1002/bies.200900072. PMID: 19722183; PMCID: PMC3130135.
