Darwin and 'Agentless Agency'
Updated: Feb 7, 2022
One of the truly revolutionary aspects of Darwin's theory of evolution by natural selection has probably been underestimated. In order to produce his theory he observed an example of human agency in the form of the human breeders of domesticated animals who selected which animals to breed and so shaped animals such as dogs to produce endless domestic varieties. Combining this with a lot of other information Darwin then postulated that a very similar process could occur to wild species of animals without any agent making the decisions; Instead 'nature' would 'decide' and select which animals were to survive, thrive and reproduce.
What Darwin's idea means in terms of 'agency' is the following: Darwin has observed a process (domestic breeding) that seems complex and requiring a complicated 'agent' (such as a human being) to assess different animals, compare them, and take this information and decide based on his or her objectives which are likely to lead to the desired qualities if bred together. The human agent must then also successfully engineer the process by which the selected animals (and only them) then reproduce. Such a process could easily be envisaged as the work of some greater 'designer' of the natural world. But Darwin's insight was to replace all that complexity with an agentless (so far simpler), open-ended, in some sense 'emergent' process. In theorising that evolution occurs under 'natural selection' Darwin has pioneered the power of 'agentless agency'; Processes that seem complex that can be far simpler with less agency (i.e. complexity) required.
Defining Agency in this context
So by 'agency' here what I refer to is this: Any process or mechanism by which some entity or entities, exercise a capacity for action and control by taking information about different options, processing it in some way, and then selecting a course of action or option which favours the outcome desired by the same 'agent' or 'agents'. By this definition many things could be agents by virtue of possessing such complexity. Agents range from simple organisms, to animals of almost any kind, especially those possessing central nervous systems, but not only such animals (the octopus has no central nervous system, but is clearly an agent by this definition). Further, I would be happy to include systems of self-organised or otherwise organised collectives or communities of organisms, especially if some self-preserving principles can be applied, or otherwise objectives can be discerned. Such communities range from slime moulds, ant colonies, bee hives, and symbiotic organisms to human organisations such as companies set up to pursue profits or other objectives or armies.
Artwork by James Robert White
The complexity of agency
What Darwin realised about agency is that it is expensive and complicated. By positing less agency at work in evolution than his rival theorists he was able to discern the real phenomena at work. Those skeptical of Darwin's theory argued that complex objects such as an 'eye' could not be the work of 'blind selection' but must have some 'designer' at work in their intricate design. However, the proof of Darwin's theory is ubiquitous in Nature and given enough time and a mechanism of inheritance in the form of genes, 'blind' selection has led to the huge variety of life and organisation that we see on earth, including eyes, and would probably work on other planets given the right conditions.
By focusing on the agentless agency aspect of Darwin's theory it is possible to expand on the intuition that what seems to require the complexity of agency may only seem to do so. Examples are complex environments which seem to require complex agents to navigate them. Given that Darwin is right about natural selection, it may also be the case that, given enough complexity and uncertainty then less agency, rather than more, might be the more workable solution to a range of problems, both in biology and economics. The theory I propose goes along the lines that whereas agency tends to have to get more and more complex and so expensive and unreliable as a solution as the environment gets more complex, so it is that above a certain level of envioronmental complexity, simpler may actually be best, or at least, the only workable solution.
Why the Evolution of the Eye depends on your point of view, hunter or hunted
To return to the example of the evolution of an eye, consider the eyes of an eagle, or any other predator. A predator animal's eyes tend to be on the front of its head, the better to see over longer distances for prey, and to use binocular vision to increase the accuracy and resolution of distant prey. This means that to look around and hunt the predator must exert agency to decide where to look and to search for prey by constantly moving its head around, or otherwise scanning by eye movements around the visual field. This necessity for agency to control attention, by taking information from the environment and deciding where next to look is a function of the coordination demands on the agent in terms of hunting.
For a predated animal the situation is very different. A preyed upon animal like a sheep or a pigeon will have evolved eyes that are suited to spot danger from any unexpected direction, and so the eyes are on either side of the head instead of facing outwards from the front. The focal range of a pigeon, the angles from which it can simultaneously see, is fully 340 degrees. A human's binocular focal range is only 120 degrees by comparison. So, when considering the role of agency versus agentless agency in animal vision, one can analyse and compare an agent-oriented solution with a less agent-oriented solution. Then one finds that when the uncertainty (a form of complexity) in the environment is very high in terms of the coordination demands created by the environment, such as the potential direction of attack of a predator for a pigeon to evade, then past a certain range of uncertainty in such coordination demand, the pigeon or any predator animal subject to the same range of uncertainty will tend to evolve a solution that involves less agency rather than more. That means it requires less control and decision making about where to look. The hypothesis is that this is because only a very complex form of agency would be able to manage such uncertainty effectively, and this is presumably not efficient or workable.
In contrast, the predator animal with no animals predating it, can adopt an agency-oriented eye design in which coordination demands are driven by the agent itself (hunting) rather than being driven by essential uncertainty of demand (being potential prey).
This approach to explaining how uncertainty and complexity in coordination demands from the environment and other animals in the environment drive different solutions that emerge again and again in Nature is an approach that highlights the differences between the suitability of an agent-oriented approach versus a less agent-oriented approach. It is expected that when coordination demands are highly uncertain then at a certain point a simpler (less agency oriented solution) will be favoured.
This type of pattern is likely to be true not just in Natural systems but in economics too, and points towards a convergence of research in to agentless agency (or less agent-oriented solutions) in economics as well as Nature. Another key example of how this can work is in the concept of searching: When there is a very high degree of uncertainty and limited information from the environment about where to look less agency-oriented solutions may be best. To back up this supposition, researchers have identified evidence of a type of random search known as a Levy Flight being used by certain animals that appear to have very limited information about where their next meal is likely to come from. This includes various animals from simple fruit fly larvae to sharks who appear to be searching without using the more complex parts of their nervous system to evaluate their options. The idea is that when the problem is complex enough in terms of uncertainty, less agency, in terms of less decision-making, less memory, etc, is actually better than more agency.
AI and Innovation
This is a strategy that I am also researching in terms of the potential for such type of agentless (or non-agent oriented) searching to explain how innovation works in economics as well as in microbiology. In doing so I have company in the form of AI researchers such as Kenneth Stanley from OpenAI and Joel Lehman who are also questioning the value of agency in performing complex processes of learning and adaptation for AI systems. In doing so, they question the value of agents pursuing objectives in such hopelessly complex environments where there is no clear path to success.