The 95/5 Rule: Most Change Looks Evolutionary
What follows is a major insight into how complex systems manage their predictable and unpredictable processes in service to adaptation. Nearly all, or roughly 95%, to a first approximation, of the pathways, decisions or events that occur in complex adaptive systems of all types are bottom-up, evolutionary processes that, though they may probabilistically predictable with respect to the next path, decision, or event, are so contingent and creative that they rapidly become unpredictable in the medium and longer term. Yet a critical subset, on the order of 5% or less, are top-down, hierarchical, developmental processes. These globally coordinated processes, by contrast, are convergent, conservative, and intrinsically future predictable, even over the long term. We can call this a 95/5 (or 97/3, 99/1, or other similar) rule, for different processes. We will stick with the name 95/5 rule for now, until evo devo processes have been much better measured and quantified in a variety of complex systems.
Why the 95/5 rule exists is obvious, if you think about how evolution and development work as we have defined them. Evolutionary processes constantly branch, or fan out, in a “bottom-up” way, as each species/system tries to adapt by trying lots of local creative experiments, which are often generated based on their local contexts. Think of all the things you do in a typical day that happen only because of the environment you find yourself in, or the information you’ve just seen or heard. Only a few of these evolutionary experiments are destined fall into funnels, special configurations that will predictably reach higher levels of complexity. These special branches are guided “top-down” to these higher levels due to the special initial conditions and laws of the universal environment, and the way those conditions and laws have been tuned for future-specific form and function via many previous cyclings of the system (in living systems and the universe as a system).
A good way to remember this rule is the Foresight Iceberg schematic, pictured at right. In evo devo systems, only a small percent of change can be long-term “visible” to foresight professionals, in that it is intrinsically predictable, convergent, and constrained or driven top-down by a few fundamental rules and actors. The vast majority of change is submerged, and significantly harder to predict, except in the short term, or in broad statistical ways, because it happens bottom-up and collectively via the creative action of a vast number of local actors, in interacting in contingent and increasingly divergent ways.
We can see this 95/5 rule operating in complex systems of all types.
A few examples:
- Almost all the genes in an organism are evolutionary. They recombine unpredictably in offspring to make new variety (phenotypes). But approximately, 3-5% of the genes are instead developmental (the developmental-genetic toolkit). They funnel the organism to one predictable phenotype. To be more precise, roughly 50% of human genes are expressed at some point during complex activities like organogenesis (Yi et al. 2010). But less than 10% of our genome is highly regulated during organogenesis, and only 5% of our DNA is ‘ultra-conserved’ across complex species (eg. human, mouse, rat) (Wagman and Stephens 2004). Preliminary work has shown that this ultra-conserved DNA can no longer undergo change without terminating or seriously damaging developmental processes. It is the core set of “top down” instructions for creating the organism. I would also bet that of the 10% of our genome that is highly regulated during organogenesis, more than half of these genes are evolutionary (still capable of creative, bottom-up change without deleterious effects to the organism). The rest are part of the core “developmental toolkit”. They conserve and funnel biological events into critical and predictable top-down choices and processes in the emergence of modularity, hierarchy, and life cycle for the organism.
- Almost all intelligence is built bottom-up. In “Modular Cognition,” (Aeon, 2022), Levin and Yuste, in a lovely article that considers the evo-devo intelligence in a single cell, describe intelligence as a network process in which pattern completion, by a few actors (the 5%), activates an entire circuit of interdependent actors (biochemical networks, gene-protein regulatory networks, neural networks). Modular hierarchies of such circuits, built bottom up, under replication, variation, and selection, keep the entire system within predictable developmental parameters. In other words, contingencies of behavior abound, for all organisms in a stochastic world (the 95%). But these ever growing networks guide complex systems, top-down, to an expanding set of predictable adaptive responses.
- Almost all of the cells in organisms are guided via bottom-up, contingent, local interactions, to form stochastic structure and function. Their positions and fates aren’t predictable in advance, as they differentiate from totipotency based on contingent local cues. A small subset are fated to spatiotemporally predictable structural and functional outcomes, once they emerge. They form a top-down “environment” and “scaffolds” which constrain the way the other cells emerge. Example: Radial glial cells, which generate and direct the formation of neural architecture.
- Almost all species on Earth evolve “randomly,” with some increasing in complexity-intelligence, and some decreasing (parasites, cul-de-sacs, etc.) as they adapt. Only a small fraction of species (~1%?), those with social brains, language, and the ability to use tools to create niches, are caught in the “funneling” path to greater world- and self-modeling, environmental control, and intelligence development as they adapt. For more, see Conway Morris, Life’s Solution (2003).
- Almost all the thoughts in a mind are unconscious, guiding the organism bottom-up, in contingent, unpredictable ways. A small fraction (1-5%?) are conscious, guiding the animal top-down as planned, convergent, optimization-seeking behavior. See Mlodinow, Subliminal (2013).
- Almost all actions in an organization happen by workers acting in bottom-up, contingent, local, unpredictable ways. Management sets top-down policies, plans, and objectives for action, but management’s actions (1-5% of org. action?) create only the critical framework, or envelope, of acceptable behavior. For more, see Kelly, Out of Control (1995).
- Almost all change in society is driven bottom-up by individuals acting in divergent, contingent, local, experimental ways (markets, democratic politics, other bottom-up behavior). A critical subset (1-5%?) of social change is top-down, planned, optimization-seeking policy set by societal leaders (govt, institutions, corps, church, etc.). Whenever societies get too centrally planned, autocratic, or plutocratic, they lose power, productivity, and collective intelligence to more bottom-up ones. Whenever societies have insufficient top-down rules, planning, and institutions, they become chaotic and unstable (think of the historical transitions to anarchy), and again lose power, productivity, and intelligence to ones with stabler environments and smarter rulesets and institutions that allow for both individual initiative and large scale, coordinated collective action. Bottom-up and top-down societal management always seems to be in a rough 95/5 balance. See Acemoglu, Why Nations Fail (2012).
The generic value of a 95/5 rule in building and maintaining intelligent systems, if one exists, would explain why the vast majority of universal change appears to be bottom-up driven, evolutionary and unpredictable in complex systems, what systems theorist Kevin Kelly described as “out of control” in his prescient work Out of Control (1994). The operation of this rule is the reason the evolutionist, chaos, and random accident perspectives have been so dominant in society for so long. Yet a critical subset of events and processes in evo devo systems is top-down/systemically directed, developmental, and intrinsically predictable, if you have the right theory, computational resources, and data. Discovering that developmental subset, and differentiating it from the much larger evolutionary subset, will make our world vastly more understandable, and show how it is constrained to certain future destinies, even as creativity and experimentation keep growing within all the evolutionary domains.
Fortunately, as our science and computing abilities advance, theories of universal evolution and development will be deeply tested, and proven or falsified, by simulation. For now, you must use your own knowledge and intuition to ask whether the 95/5 rule (hypothesis) makes sense. Science is a long way from being able to verify or refute it, or any alternative. We can only cite evidence and argument.
- Wagman and Stephens (2004) Surprising ‘ultra-conserved regions discovered in human genome. UCSC Currents.
- Yi, Hong et al. (2010) Gene expression atlas for human embryogenesis. FASEB Journal 24(9):3341-3350.