I think I have moved forward in the process of revising my manuscript ‘Climbing the right hill – an evolutionary approach to the European market of electricity’, as a resubmission Applied Energy . A little digression: as I provide, each time, a link to the original form of that manuscript, my readers can compare the changes I develop on, in those updates, with the initial flow of logic.
I like discussing important things in reverse order. I like starting from what apparently is the end and the bottom line of thinking. From there, I go forward by going back, sort of. In an article, the end is the conclusion, possibly summarized in 5 ÷ 6 bullet points and optionally coming together with a graphical abstract. I conclude this specific piece of research by claiming that energy-oriented policies, e.g. those oriented on developing renewable sources, could gain in efficiency by being: a) national rather than continental or global b) explicitly oriented on optimizing the country’s terms of trade in global supply chains c) just as explicitly oriented on the development of some specific types of jobs whilst purposefully winding down other types thereof.
I give twofold a base for that claim. Firstly, I have that stylized general observation about energy-oriented policies: globally or continentally calibrated policies, such as, for example, the now famous Paris Climate Agreement, work so slow and with so much friction that they become ineffective for any practical purpose, whilst country-level policies are much more efficient in the sense that one can see a real transition from point A to point B. Secondly, my own research – which I present in this article under revision – brings evidence that national social structures orient themselves on optimizing their terms of trade and their job markets in priority, whilst treating the energy-related issues as instrumental. That specific collective orientation seems, in turn, to have its source in the capacity of human social structures to develop a strongly cyclical, predictable pattern of collective learning precisely in relation to the terms of trade, and to the job market, whilst collective learning oriented on other measurable variables, inclusive of those pertinent to energy management, is much less predictable.
That general conclusion is based on quantitative results of my empirical research, which brings forth 4 quantitative variables – price index in exports (PL_X), average hours worked per person per year (AVH), the share of labour compensation in Gross National Income (LABSH), and the coefficient of human capital (HC – average years of schooling per person) – out of a total scope of 49 observables, as somehow privileged collective outcomes marked with precisely that recurrent, predictable pattern of learning.
The privileged position of those specific variables, against the others, manifests theoretically as their capacity to produce simulated social realities much more similar to the empirically observable state thereof than simulated realities based on other variables, whilst producing a strongly cyclical series of local residual errors in approximating said empirically observable state.
The method which allowed to produce those results generates simulated social realities with the use of artificial neural networks. Two types of networks are used to generate two types of simulation. One is a neural network which optimizes a specific empirical variable as its output, whilst using the remaining empirical variables as instrumental input. I call that network ‘procedure of learning by orientation’. The other type of network uses the same empirical variable as its optimizable output and replaces the vector of other empirical variables with a vector of hypothetical probabilities, corresponding to just as hypothetical social roles, in the presence of a random disturbance factor. I label this network as ‘learning procedure by pacing’.
The procedure of learning by orientation produces as many alternative sets of numerical values as there are variables in the original empirical dataset X used in research. In this case, it was a set made of n = 49 variables, and thus 49 alternative sets Si are created. Each alternative set Si consists of values transformed by the corresponding neural network from the original empirical ones. Both the original dataset X and the n = 49 transformations Si thereof can be characterized, mathematically, with their respective vectors of mean expected values. Euclidean distances between those vectors are informative about the mathematical similarity between the corresponding sets.
Therefore, learning by orientation produces n = 49 simulations Si of the actual social reality represented in the set X, when each such simulation is biased towards optimizing one particular variable ‘i’ from among the n = 49 variables studied, and each such simulation displays a measurable Euclidean similarity to the actual social reality studied. My own experience in applying this specific procedure is that a few simulations Si, namely those pegged on optimizing four variables – price index in exports [Si(PL_X)], average hours worked per person per year [Si[AVH)], the share of labour compensation in Gross National Income [Si(LABSH)], and the coefficient of human capital [Si(HC) – average years of schooling per person] – display much closer Euclidean a distance to the actual reality X than any other simulation. Much closer means closer by orders of magnitude, by the way. The difference is visible.
The procedure of learning by pacing produces n = 49 simulations as well, yet these simulations are not exactly transformations of the original dataset X. In this case, simulated realities are strictly simulated, i.e. they are hypothetical states from the very beginning, and individual variables from the set X serve as the basis for setting a trajectory of transformation for those hypothetical states. Each such hypothetical state is a matrix of probabilities, associated with two sets of social roles: active and dormant. Active social roles are being endorsed by individuals in that hypothetical society and their baseline, initial probabilities are random, non-null values. Dormant social roles are something like a formalized prospect for immediate social change, and their initial probabilities are null.
This specific neural network produces new hypothetical states in two concurrent ways: typical neural activation, and random disturbance. In the logical structure of the network, random disturbance occurs before neural activation, and thus I am describing details of the former in the first place. Random disturbance is a hypothetical variable, separate from probabilities associated with social roles. It is a random value 0 < d < 1, associated with a threshold of significance d*. When d > d*, d becomes something like an additional error, fed forward into the network, i.e. impacting the next experimental round performed in the process of learning.
In the procedure of learning by pacing, neural activation is triggered by aggregating partial probabilities, associated with social roles, and possibly pre-modified by the random disturbance, through the operation of weighed average of the type ∑ fj(pi, X(i,j), dj, ej-1,), where fj is the function of neural activation in the j-th experimental round of learning, pi is the probability associated with the i-th social role, X(i,j) is the random weight of pi in the j-th experimental round, dj stands for random disturbance specific to that experimental round, and ej-1 is residual error fed forward from the previous experimental round j-1.
Now, just to be clear: there is a mathematical difference, in that logical structure, between random disturbance dj, and the random weight X(i,j). The former is specific to a given experimental round, but general across all the component probabilities in that round. If you want, di is like an earthquake, momentarily shaking the entire network, and is supposed to represent the fact that social reality just never behaves as planned. This is the grain of chaos in that mathematical order. On the other hand, X(i,j) is a strictly formal factor in the neural activation function, and its only job is to allow experimenting with data.
Wrapping it partially up, the whole method I use in this article revolves around the working hypothesis that a given set of empirical data, which I am working with, represents collectively intelligent learning, where human social structures collectively experiment with many alternative versions of themselves and select those versions which offer the most desirable states in a few specific variables. I call these variables ‘collective orientations’ and I further develop that hypothesis by claiming that collective orientations have that privileged position because they allow a specific type of collective learning, strongly cyclical, with large amplitude of residual error.
In both procedures of learning, i.e. in orientation, and in pacing, I introduce an additional component, namely that of self-observed internal coherence. The basic idea is that a social structure is a structure because the functional connections between categories of phenomena are partly independent from the exact local content of those categories. People remain in predictable functional connections to their Tesla cars, whatever exact person and exact car we are talking about. In my method, and, as a matter of fact, in any quantitative method, variables are phenomenological categories, whilst the local values of those variables inform about the local content to find in respective categories. My idea is that mathematical distance between values represents temporary coherence between the phenomenological categories behind the corresponding variables. I use the Euclidean distance of the type E = [(a – b)2]0,5 as the most elementary measure of mathematical distance. The exact calculation I do is the average Euclidean distance that each i-th variable in the set of n variables keeps from each l-th variable among the remaining k = n – 1 variables, in the same experimental round j. Mathematically, it goes like: avgE = { ∑[(xi – xl)2]0,5 }/k. When I use avgE as internally generated input in a neural network, I use the information about internal coherence as meta-data in the process of learning.
Of course, someone could ask what the point is of measuring local Euclidean distance between, for example, annual consumption of energy per capita and the average number of hours worked annually per capita, thus between kilograms of oil equivalent and hours. Isn’t it measuring the distance between apples and oranges? Well, yes, it is, and when you run a grocery store, knowing the coherence between your apples and your oranges can come handy, for one. In a neural network, variables are standardized, usually over their respective maximums, and therefore both apples and oranges are measured on the same scale, for two.
The method needs to be rooted in theory, which has two levels: general and subject-specific. At the general level, I need acceptably solid theoretical basis for positing the working hypothesis, as phrased out in the preceding paragraph, to any given set of empirical, socio-economic data. Subject-specific theory is supposed to allow interpreting the results of empirical research as conducted according to the above-discussed method.
General theory revolves around four core concepts, namely those of: intelligent structure, chain of states, collective orientation, and social roles as mirroring phenomena for quantitative socio-economic variables. Subject-specific theory, on the other hand, is pertinent to the general issue of energy-related policies, and to their currently most tangible manifestation, i.e., to environmentally friendly sources of energy.
The theoretical concept of intelligent structure, such as I use it in my research, is mostly based on another concept, known from evolutionary biology, namely that of adaptive walk in rugged landscape, combined with the phenomenon of tacit coordination. We, humans, do things together without being fully aware we are doing them together or even whilst thinking we oppose each other (e.g. Kuroda & Kameda 2019[1]). capacity for social evolutionary tinkering (Jacob 1977[2]) through tacit coordination, such that the given society displays social change akin to an adaptive walk in rugged landscape (Kauffman & Levin 1987[3]; Kauffman 1993[4]; Nahum et al. 2015[5]).
Each distinct state of the given society (e.g. different countries in the same time or different moments in time as regards the same country) is interpreted as a vector of observable properties, and each empirical instance of that vector is a 1-mutation-neighbour to at least one other instance. All the instances form a space of social entities. In the presence of external stressor, each such mutation (each entity) displays a given fitness to achieve the optimal state, regarding the stressor in question, and therefore the whole set of social entities yields a complex vector of fitness to cope with the stressor.
The assumption of collective intelligence means that each social entity is able to observe itself as well as other entities, so as to produce social adaptation for achieving optimal fitness. Social change is an adaptive walk, i.e. a set of local experiments, observable to each other and able to learn from each other’s observed fitness. The resulting path of social change is by definition uneven, whence the expression ‘adaptive walk in rugged landscape’. There is a strong argument that such adaptive walks occur at a pace proportional to the complexity of social entities involved. The greater the number of characteristics involved, the greater the number of epistatic interactions between them, and the more experiments it takes to have everything more or less aligned for coping with a stressor.
Somehow concurrently to the evolutionary theory, another angle of approach seems interesting, for solidifying theoretical grounds to my method: the swarm theory (e.g. Wood & Thompson 2021[6]; Li et al. 2021[7]). Swarm learning consists in shifting between different levels of behavioural coupling between individuals. When we know for sure we have everything nicely figured out, we coordinate, between individuals, by fixed rituals or by strongly correlated mutual reaction. As we have more and more doubts whether the reality which we think we are so well adapted to is the reality actually out there, we start loosening the bonds of behavioural coupling, passing through weakening correlation, and all the way up to random concurrence. That unbundling of social coordination allows incorporating new behavioural patterns into individual social roles, and then learning how to coordinate as regards that new stuff.
As the concept of intelligent structure seems to have a decent theoretical base, the next question is: how the hell can I represent it mathematically? I guess that a structure is a set of connections inside a complex state, where complexity is as a collection of different variables. I think that the best mathematical construct which fits that bill is that of imperfect Markov chains (e.g. Berghout & Verbitskiy 2021[8]): there is a state of reality Xn = {x1, x2, …, xn}, which we cannot observe directly, whilst there is a set of observables {Yn} such that Yn = π (Xn), the π being a coding map of Xn. We can observe through the lens of Yn. That quite contemporary theory by Berghout and Verbitskyi sends to an older one, namely to the theory of g-measures (e.g. Keane 1972[9]), and all that falls into an even broader category of ergodic theory, which is the theory of what happens to complex systems when they are allowed to run for a long time. Yes, when we wonder what kind of adult our kids will grow up into, this is ergodic theory.
The adaptive walk of a human society in the rugged landscape of whatever challenges they face can be represented as a mathematical chain of complex states, and each such state is essentially a matrix: numbers in a structure. In the context of intelligent structures and their adaptive walks, it can be hypothesized that ergodic changes in the long-going, complex stuff about what humans do together happen with a pattern and are far from being random. There is a currently ongoing, conceptual carryover from biology to social sciences, under the general concept of evolutionary trajectory (Turchin et al. 2018[10]; Shafique et al. 2020[11]). That concept of evolutionary trajectory can be combined with the idea that our human culture pays particular attention to phenomena which make valuable outcomes, such as presented, for example, in the Interface Theory of Perception (Hoffman et al. 2015[12], Fields et al. 2018[13]). Those two theories taken together allow hypothesising that, as we collectively learn by experimenting with many alternative versions of our societies, we systematically privilege those particular experiments where specific social outcomes are being optimized. In other words, we can have objectively existing, collective ethical values and collective praxeological goals, without even knowing we pursue them.
The last field of general theory I need to ground in literature is the idea of representing the state of a society as a vector of probabilities associated with social roles. This is probably the wobbliest theoretical boat among all those which I want to have some footing in. Yes, social sciences have developed that strong intuition that humans in society form and endorse social roles, which allows productive coordination. As Max Weber wrote in his book ‘Economy and Society’: “But for the subjective interpretation of action in sociological work these collectivities must be treated as solely the resultants and modes of organization of the particular acts of individual persons, since these alone can be treated as agents in a course of subjectively understandable action”. The same intuition is to find in Talcott Parsons’ ‘Social system’, e.g. in Chapter VI, titled ‘The Learning of Social Role-Expectations and the Mechanisms of 138 Socialization of Motivation’: “An established state of a social system is a process of complementary interaction of two or more individual actors in which each conforms with the expectations of the other(’s) in such a way that alter’s reactions to ego’s actions are positive sanctions which serve to reinforce his given need-dispositions and thus to fulfill his given expectations. This stabilized or equilibrated interaction process is the fundamental point of reference for all dynamic motivational analysis of social process. […] Every society then has the mechanisms which have been called situational specifications of role-orientations and which operate through secondary identifications and imitation. Through them are learned the specific role-values and symbol-systems of that particular society or sub-system of it, the level of expectations which are to be concretely implemented in action in the actual role”.
Those theoretical foundations laid, the further we go, the more emotions awaken as the concept of social role gets included in scientific research. I have encountered views, (e.g. Schneider & Bos 2019[14]) that social roles, whilst being real, are a mechanism of oppression rather than social development. On the other hand, it can be assumed that in the presence of demographic growth, when each consecutive generation brings greater a number of people than the previous one, we need new social roles. That, in turn, allows developing new technologies, instrumental to performing these roles (e.g. Gil-Hernández et al. 2017[15]).
Now, I pass to the subject-specific, theoretical background of my method. I think that the closest cousin to my method, which I can find in recently published literature, is the MuSIASEM framework, where the acronym, deliberately weird, I guess, stands for ‘Multi-scale Integrated Analysis of Societal and Ecosystem Metabolism’. This is a whole stream of research, where human societies are studied as giant organisms, and the ways we, humans, make and use energy, is studied as a metabolic function of those giant bodies. The central assumption of the MuSIASEM methodology is that metabolic systems survive and evolve by maxing out on energy efficiency. The best metabolism for an economic system is the most energy-efficient one, which means the greatest possible amount of real output per unit of energy consumption. In terms of practical metrics, we talk about GDP per kg of oil equivalent in energy, or, conversely, about the kilograms of oil equivalent needed to produce one unit (e.g. $1 bln) of GDP. You can consult Andreoni 2020[16], Al-Tamimi & Al-Ghamdi 2020[17] or Velasco-Fernández et al. 2020[18], as some of the most recent examples of MuSIASEM being applied in empirical research.
This approach is strongly evolutionary. It assumes that any given human society can be in many different, achievable states, each state displaying a different energy efficiency. The specific state which yields the most real output per unit of energy consumed is the most efficient metabolism available to that society at the moment, and, logically, should be the short-term evolutionary target. Here, I dare disagreeing fundamentally. In nature, there is no such thing as evolutionary targets. Evolution happens by successful replication. The catalogue of living organisms which we have around, today, are those which temporarily are the best at replicating themselves, and not necessarily those endowed with the greatest metabolic efficiency. There are many examples of species which, whilst being wonders of nature in terms of biologically termed efficiency, are either endemic or extinct. Feline predators, such as the jaguar or the mountain lion, are wonderfully efficient in biomechanical terms, which translates into their capacity to use energy efficiently. Yet, their capacity to take over available habitats is not really an evolutionary success.
In biological terms, metabolic processes are a balance of flows rather than intelligent strive for maximum efficiency. As Niebel et al. (2019[19]) explain it: ‘The principles governing cellular metabolic operation are poorly understood. Because diverse organisms show similar metabolic flux patterns, we hypothesized that a fundamental thermodynamic constraint might shape cellular metabolism. Here, we develop a constraint-based model for Saccharomyces cerevisiae with a comprehensive description of biochemical thermodynamics including a Gibbs energy balance. Non-linear regression analyses of quantitative metabolome and physiology data reveal the existence of an upper rate limit for cellular Gibbs energy dissipation. By applying this limit in flux balance analyses with growth maximization as the objective function, our model correctly predicts the physiology and intracellular metabolic fluxes for different glucose uptake rates as well as the maximal growth rate. We find that cells arrange their intracellular metabolic fluxes in such a way that, with increasing glucose uptake rates, they can accomplish optimal growth rates but stay below the critical rate limit on Gibbs energy dissipation. Once all possibilities for intracellular flux redistribution are exhausted, cells reach their maximal growth rate. This principle also holds for Escherichia coli and different carbon sources. Our work proposes that metabolic reaction stoichiometry, a limit on the cellular Gibbs energy dissipation rate, and the objective of growth maximization shape metabolism across organisms and conditions’.
Therefore, if we translate the principles of biological metabolism into those of economics and energy management, the energy-efficiency of any given society is a temporary balance achieved under constraint. Whilst those states of society which clearly favour excessive dissipation of energy are not tolerable on the long run, energy efficiency is a by-product of the strive to survive and replicate, rather than an optimizable target state. Human societies are far from being optimally energy efficient for the simple reason that we have plenty of energy around, and, with the advent of renewable sources, we have even less constraint to optimize energy-efficiency.
We, humans, survive and thrive by doing things together. The kind of efficiency that allows maxing out on our own replication is efficiency in coordination. This is why we have all that stuff of social roles, markets, institutions, laws and whatnot. These are our evolutionary orientations, because we can see immediate results thereof in terms of new humans being around. A stable legal system, with a solid centre of political power in the middle of it, is a well-tested way of minimizing human losses due to haphazard violence. Once a society achieves that state, it can even move from place to place, as local resources get depleted.
I think I have just nailed down one of my core theoretical contentions. The originality of my method is that it allows studying social change as collectively intelligent learning, whilst remaining very open as for what this learning is exactly about. My method is essentially evolutionary, whilst avoiding the traps of evolutionary metaphysics, such as hypothetical evolutionary targets. I can present my method and my findings as a constructive theoretical polemic with the MuSIASEM framework.
[1] Kuroda, K., & Kameda, T. (2019). You watch my back, I’ll watch yours: Emergence of collective risk monitoring through tacit coordination in human social foraging. Evolution and Human Behavior, 40(5), 427-435. https://doi.org/10.1016/j.evolhumbehav.2019.05.004
[2] Jacob, F. (1977). Evolution and tinkering. Science, 196(4295), 1161-1166
[3] Kauffman, S., & Levin, S. (1987). Towards a general theory of adaptive walks on rugged landscapes. Journal of theoretical Biology, 128(1), 11-45
[4] Kauffman, S. A. (1993). The origins of order: Self-organization and selection in evolution. Oxford University Press, USA
[5] Nahum, J. R., Godfrey-Smith, P., Harding, B. N., Marcus, J. H., Carlson-Stevermer, J., & Kerr, B. (2015). A tortoise–hare pattern seen in adapting structured and unstructured populations suggests a rugged fitness landscape in bacteria. Proceedings of the National Academy of Sciences, 112(24), 7530-7535, www.pnas.org/cgi/doi/10.1073/pnas.1410631112
[6] Wood, M. A., & Thompson, C. (2021). Crime prevention, swarm intelligence and stigmergy: Understanding the mechanisms of social media-facilitated community crime prevention. The British Journal of Criminology, 61(2), 414-433. https://doi.org/10.1093/bjc/azaa065
[7] Li, M., Porter, A. L., Suominen, A., Burmaoglu, S., & Carley, S. (2021). An exploratory perspective to measure the emergence degree for a specific technology based on the philosophy of swarm intelligence. Technological Forecasting and Social Change, 166, 120621. https://doi.org/10.1016/j.techfore.2021.120621
[8] Berghout, S., & Verbitskiy, E. (2021). On regularity of functions of Markov chains. Stochastic Processes and their Applications, Volume 134, April 2021, Pages 29-54, https://doi.org/10.1016/j.spa.2020.12.006
[9] Keane, M. (1972). Strongly mixingg-measures. Inventiones mathematicae, 16(4), 309-324. DOI
[10] Turchin, P., Currie, T. E., Whitehouse, H., François, P., Feeney, K., Mullins, D., … & Spencer, C. (2018). Quantitative historical analysis uncovers a single dimension of complexity that structures global variation in human social organization. Proceedings of the National Academy of Sciences, 115(2), E144-E151. https://doi.org/10.1073/pnas.1708800115
[11] Shafique, L., Ihsan, A., & Liu, Q. (2020). Evolutionary trajectory for the emergence of novel coronavirus SARS-CoV-2. Pathogens, 9(3), 240. https://doi.org/10.3390/pathogens9030240
[12] Hoffman, D. D., Singh, M., & Prakash, C. (2015). The interface theory of perception. Psychonomic bulletin & review, 22(6), 1480-1506.
[13] Fields, C., Hoffman, D. D., Prakash, C., & Singh, M. (2018). Conscious agent networks: Formal analysis and application to cognition. Cognitive Systems Research, 47, 186-213. https://doi.org/10.1016/j.cogsys.2017.10.003
[14] Schneider, M. C., & Bos, A. L. (2019). The application of social role theory to the study of gender in politics. Political Psychology, 40, 173-213. https://doi.org/10.1111/pops.12573
[15] Gil-Hernández, C. J., Marqués-Perales, I., & Fachelli, S. (2017). Intergenerational social mobility in Spain between 1956 and 2011: The role of educational expansion and economic modernisation in a late industrialised country. Research in social stratification and mobility, 51, 14-27. http://dx.doi.org/10.1016/j.rssm.2017.06.002
[16] Andreoni, V. (2020). The energy metabolism of countries: Energy efficiency and use in the period that followed the global financial crisis. Energy Policy, 139, 111304. https://doi.org/10.1016/j.enpol.2020.111304
[17] Al-Tamimi and Al-Ghamdi (2020), ‘Multiscale integrated analysis of societal and ecosystem metabolism of Qatar’ Energy Reports, 6, 521-527, https://doi.org/10.1016/j.egyr.2019.09.019
[18] Velasco-Fernández, R., Pérez-Sánchez, L., Chen, L., & Giampietro, M. (2020), A becoming China and the assisted maturity of the EU: Assessing the factors determining their energy metabolic patterns. Energy Strategy Reviews, 32, 100562. https://doi.org/10.1016/j.esr.2020.100562
[19] Niebel, B., Leupold, S. & Heinemann, M. An upper limit on Gibbs energy dissipation governs cellular metabolism. Nat Metab 1, 125–132 (2019). https://doi.org/10.1038/s42255-018-0006-7
