A RESEARCH STORY
THE ORIGINS OF GOAL-DIRECTEDNESS
Unravelling the origin of goal-directedness could open up a whole new perspective on reality,
cosmic evolution, consciousness and the meaning of life.
GOAL-DIRECTEDNESS IS EVERYWHERE BUT ELUSIVE
The world is full of living and non-living systems, from minds to organisms and social systems, that are able to achieve goals in an autonomous way, even when faced with perturbations. In other words, all these systems show goal-directed or agentic, self-correcting behaviour. Survival - also called "self-maintenance" - is the primary, universal goal, but there are numerous other subordinate goals in naturecultures like chasing invaders (macrophages), pumping around oxygen (hearts) or keeping the room temperature constant (thermostats).
Many goal-directed systems even seem to exist only to achieve their goal. For example, we can’t meaningfully describe what macrophages do without referring to their purpose or function. Their whole being seems to amount to conducting their "seek and destroy" manoeuvres. We also accept that most features of biological, mental or social systems exist "for" the achievement of the system’s overall goal. Roots support the growth of a tree by absorbing water and nutrients. The future state of killing shapes the anatomy of claws and teeth in a cheetah embryo. Molecular biologists, neuroscientists and geneticists can barely do their work without employing this "goal-directed" mode of thinking.
Goal-directedness is everywhere, but we either ignore it or don’t truly grasp it. Within the sciences, it remains a controversial concept that is frequently misunderstood (PAPER 2023). We tend to assume that goal-directed behaviour requires human consciousness and that nonhumans are only "apparently" goal-directed, that they are only executing - instinctively or mechanically - what they are genetically or culturally programmed to do. Many scientists also talk around goal-directedness by using the notion of "teleonomy" to explain biological features. The main idea is that nonhuman beings can’t control their trajectory actively and independently from internal (genetic, instinctive) or external (designed, environmental) programs or forces. Their behaviour is fully determined.
Another research area that often takes goal-directedness for granted, but would benefit from exploring its origin, is intelligence research. Definitions of intelligence generally refer to the ability to solve problems, which assumes that intelligence and “goal-directedness toward solving problems” are inseparably linked. Modelling how the underlying “goal to solve problems” has emerged, could shed more light on the origin of intelligence and boost the development of AI systems that can determine their own goals (and values), instead of inserting these by design or letting the learning algorithms obscure them. It’s worth exploring if such “agentic” AI systems could bridge the gap between human understanding and machine intelligence, and lead to more responsible, ethical and flexible AI systems.
HOW DID GOAL-DIRECTEDNESS EMERGE?
Unfortunately, ignoring or ambiguously explaining away goal-directedness leaves many questions unanswered. What is the real mechanism behind target-oriented behaviour? Is goal-directedness rooted in consciousness, or is it the other way around? Can you attribute goal-directed behaviour to genes only? In other words, can a preset repertoire of actions at the genetic level account for all the possible environmental challenges? This seems unlikely, as organisms need an exponentially larger repertoire of actions to overcome unexpected perturbations, crucial for their survival. If nonhuman living beings would only follow the limited genetic repertoire of actions, how can we explain their ability to choose the right action from this much larger repertoire?
And in ultimate terms, what is the origin of goal-directedness? Evolution theory explains the emergence of subordinate goals that support the primary goal of self-maintenance. Through variation and selection, every system finetunes its self-maintenance strategies, becoming increasingly adaptive for survival. What evolution theory doesn’t explain, however, is how the primary goal of survival emerged, and thus the origin of "goal-directedness" itself. It assumes a lowest level of goal-directedness that evolution can build on, leaving the primary level unexplained. How did goal-directedness emerge? Was there a time when nothing on Earth was goal-directed? How did mechanical, unstructured, chaotic, physico-chemical processes evolve entirely spontaneously, without intelligent design, into complex systems that autonomously pursue and achieve goals like self-maintenance? How did something that was initially only determined by external causes acquire the autonomy to pursue goals and achieve a form of independence? How did abiotic components (molecules, membranes) develop purposeful action? And how does the origin of goal-directedness relate to the origin of life, mind and behaviour?
All these questions indicate that exploring the origin of goal-directedness may teach us something fundamental about consciousness and the origin, nature and meaning of life.
A NEW SCENARIO-MODEL FOR GOAL-DIRECTEDNESS
Standard scientific models make it nearly impossible to explain the transition from purposelessness to purpose, which is probably why goal-directedness has been disregarded for so long. To unravel this major transition, we decided to integrate relevant ideas and approaches advanced by Cybernetics, the Dynamical Systems Theory (DST) and the new formalism of the Chemical Organization Theory (COT). This allows us to build a more elegant, transdisciplinary scenario-model for the origin and evolution of goal-directedness (PAPERS 2015, 2017, 2021, 2023).
Important, we are studying goal-directedness in abstract terms, not in a specific biological or (bio)chemical setting (PAPER 2017). Our model is independent of technicalities like the biochemistry of the first cells. It doesn’t specify concrete molecules (such as RNA) or biochemical pathways (such as the Krebs cycle) that may have played a role in the origin of life on Earth. Instead, it describes the conditions and processes at a higher level of abstraction by using the mathematical language of (chemical) reaction networks proposed by the Chemical Organization Theory, which relates to all kinds of process-based ontologies (not only "chemical" ones, the term "COT" is coined historically). This approach allows us to specify functional roles and structural features in a way that can be applied to a wide variety of cases (PAPER 2017). Our scenario-model should be able to explain the emergence of both biological and artificial processes like computer programs, together with their more complex and subtle aspects, but also (as yet unorganized) social systems, and even the origin of goal-directedness on other planets or stars.
OUR ASSUMPTIONS
Existing "origin of life" scenarios have difficulty clarifying how the first living systems could become goal-directed at survival. Some try to explain the origin of goal-directedness starting from autocatalytic sets, but these tend to be too unstable and dependent on their environment to maintain themselves, failing to produce effective goal-directed systems.
We propose a different hypothesis (PAPER 2023). Because goal-directedness is an abstract property, we had to find a way to observe (and explain) it by looking for a related, less abstract and better observable property: resilience. Our assumption is that goal-directed behaviour emerges in reaction networks that have self-organized and evolved into more stable or resilient "chemical organizations". We define (chemical) organizations as networks of reactions that have managed to self-organize autonomously by becoming operationally closed and self-producing or "autopoietic". All the resources consumed in such a network are produced again by that same reaction network in sufficient amounts (self-producing) and only resources are produced that were also in the initial set (closed). Take for example convection cells, which are circular flows of liquid that emerge through self-organization when a liquid is heated from below.
Next, we assume that most spontaneously emerged, self-producing systems are intrinsically fragile. Therefore, most organizations will only be able to maintain their goal of self-maintenance in ideal circumstances. In a more realistic, unpredictable environment, an organization’s constituent processes will be more easily disrupted by perturbations in a way it can’t bounce back to a self-maintaining regime. Such a fragile organization is unable to maintain its "qualitative identity" and will quickly disappear. For example, a small reduction in temperature difference may be sufficient to destroy a convection cell. We do not consider fragile organizations such as convection cells to be effectively goal-directed. In other words - and linguistically this is a bit confusing - just being "directed" at a goal, acting to reach it and even achieving this goal in the absence of perturbations (ideal circumstances), is not enough to be "effectively" goal-directed.
We consider an organization only to be goal-directed when it needs to act to reach its goal of self-maintenance and succeeds at reaching this goal both in the absence and presence of perturbations. Such an organization is called "resilient": it is able to survive by actively bypassing perturbations that make it deviate from a far-from-equilibrium goal, or it can survive by evolving into a new organization with a new network structure. Resilience makes organizations to some degree autonomous: they can recover on their own from perturbations and follow their own course, not allowing the environment to make them deviate from that course. In order for goal-directedness to function not just in principle but also in practice, the range of perturbations that a resilient organization must be able to deal with should be as large as possible. All this implies that resilient organizations exhibit "agency": instead of being passively subjected to external forces, they act so as to counter or exploit these forces in the service of their goals of survival and growth.
To conclude, in our definition only "resilient organizations" are goal-directed. You could say that through their mere existence, they "bring along" the concept of goal-directedness, together with the primary goal of self-maintenance or survival. In other words, wherever we observe the emergence of resilient organizations, features commonly understood as "footprints of goal-directedness" emerge as well. To develop a scenario for the origin and evolution of goal-directedness we, therefore, need to explore the origin and evolution of self-organization and, especially, resilience.
Understanding which features increase the resilience of organizations, and how these features emerge and evolve, is thus the main focus of our project. To investigate this, we combine a bottom-up with a top-down analysis of resilience. Bottom-up implies triggering the emergence of organizations, confronting them with perturbations, selecting the most resilient ones and analysing their features. A top-down analysis implies designing resilient organizations by providing them with known resilience features, adding perturbations and testing for resilience.
OUR APPROACH AND FIRST RESULTS
The first step was developing COT software to investigate the conditions under which reaction networks start to self-organize in “ideal circumstances”. What we expected from theory, and tested with COT software, is that when reaction networks are sufficiently large and complex, various organizations spontaneously emerge. We also confirmed that, in the long run, and leaving technicalities aside, attractors of the reaction system dynamics correspond to organizations (PAPERS 2007, 2012, 2021, 2022). This is fairly easy to model both mathematically and computationally, but we don’t know the exact network size and degree of complexity yet needed for organizations to emerge, nor the exact role of catalysts in this process. Refining our models will give us more insights.
The next step was testing how resilience emerges in these computer-generated organizations by exposing them to various perturbations and observing which features characterize the most resilient ones (PAPER 2022, 2022-2). Because our COT models don’t prescribe resilience features and can’t simulate perturbations - or changes in the environment of an organization - we had to refine our models by using Cybernetics and the Dynamical Systems Theory (DST).
The Dynamical Systems Theory helps us to simulate perturbations by injecting or removing resources and reactions from an organization’s reaction network. With COT software, we can monitor whether these perturbed organizations survive, collapse or evolve into a new organization. We decided to quantify the resilience of organizations in terms of their ability to resist perturbations in the short term (local resilience) and in terms of how many other perturbed organizations they have evolved from (global resilience). We also extended the notions of “local” and “global” resilience to the notion of “collective resilience”, which measures how groups of organizations tend to form cycles transforming into one another when perturbed. The discovery that collectives of organizations can be more resilient than their parts is an indication that evolution does not only drive entities (organizations) but also their potentialities (other organizations in the collective) (PAPER 2023). We are also studying how our three notions of resilience - local, global and collective - relate to the resilience features described by DST and Cybernetics.
The theory of Cybernetics specifies six features a system needs to effectively anticipate, regulate or recover from external perturbations, such as negative feedback and buffering. These cybernetic "control" mechanisms help organizations to become resilient and attain goals like continued self-maintenance. The Dynamical Systems Theory describes nine additional features of resilience in dynamical systems, such as a large basin of attraction and abundance. However, Cybernetics and DST don’t clarify how all these resilience features emerge and evolve in organizations. Translating them into COT language allows us to generate and "measure" the presence of resilience features in computer-generated organizations (PAPER 2023). Investigating to what extent organizations possess certain resilience features, allows for developing an evolutionary taxonomy from less to more resilient organizations and the features they (don’t) share.
Our hypothesis is that resilient organizations are characterized by at least the cybernetic "negative feedback" mechanism, which stabilizes an organization’s production and consumption processes (PAPER 2021) (video). As a consequence, resilient organizations are much more stable and evolve more easily than autocatalytic sets, which are characterized by a positive feedback mechanism that quickly exhausts any food source, destroying the whole cycle.
From our simulations, we also learned that organizations become more resilient when they end up in larger "basins of attraction". A large basin ensures sufficient plasticity, persistence, and concerted action which are - next to negative feedback - necessary DST resilience features for organizations to reach their goal of self-maintenance in the face of perturbations (PAPER 2022).
Evolution then favours organizations with the most effective resilience features, such as a larger basin and negative feedback. The evolution of goal-directed organizations seems to follow a trajectory that maximizes the increase of resilience while minimizing the increase of complexity (PAPERS 2023-1, 2023-2). Eventually, a resilient organization will become an integrated, synergetic, dynamic, "intelligent" system that is sensitive to its surroundings.
What we don’t know yet are the precise conditions under which organizations with large basins emerge, nor the exact basin size needed for organizations to become resilient. One hypothesis is that the more perturbations an organization undergoes that don’t make it collapse but evolve into a new organization, the higher the probability that the resulting organization will have a larger basin of attraction, making it even more resilient. An ongoing series of perturbations triggers organizations to jump out of small basins while allowing them to settle into more resilient, large-basin attractors which are resistant to further perturbations.
If we can simulate this process in detail with COT software, we would have a realistic, solid, operational scenario for the origin and evolution of resilient organizations, and thus for the origin and evolution of goal-directedness. The last step necessary to flesh out this formal model and confirm its viability would be to compare it with known evolutionary transitions by applying it to data from existing goal-directed systems including metabolic networks and social systems.
IMPLICATIONS OF DISCOVERING THE ORIGIN OF GOAL-DIRECTEDNESS
What would be the impact of developing a formal scenario that describes the origin and evolution of goal-directedness in an abstract way, based on the origin and evolution of self-organization and resilience?
First, such a scenario could be used to formulate falsifiable predictions about the behaviour of any goal-directed system. It would be applicable to many processes, from biological and artificial life to social systems and even extraterrestrial contexts. This would suggest that goal-directedness tends to self-organize out of any sufficiently rich substrate of chemical, physical, informational, social or other interactions. It would imply that goal-directedness is not a mysterious concept but a universal phenomenon, part of the fabric of the cosmos. Goal-directedness being a natural outcome of the evolution of the universe would open up a whole new perspective on cosmic evolution.
Second, goal-directedness being a fundament of reality would imply that science can no longer reduce everything to inert pieces of matter and purposeless mechanical processes without agency. This would extend the mechanistic scientific worldview (whether random or deterministic) and bring it more in line with humanistic and spiritual philosophies that seek purpose in life (PAPERS 2021, 2023). A formal scenario for the origin of goal-directedness could provide the foundation for a ‘science of purpose’ that is both fully compatible with causality and other laws of physics, and gives a scientific grounding for our intuition that life is intrinsically purposeful, or even meaningful. This could throw new light on what life is, and how it evolved.
Third, such a scenario would also be a good model for the emergence of life on Earth. Because our scenario-model is based on the emergence of resilient organizations from networks, it aligns well with the self-producing dynamics postulated in other theories - like cybernetics, autopoiesis and autocatalytic cycles - as the essence of agency, and thus of life. Detecting the origin of goal-directedness early in evolutionary history would imply that "purpose" is an essential part of what defines all living creatures (PAPER 2023).
Fourth, goal-directedness being a fundament of life implies that later evolved biological features are not needed to display goal-directed behaviour. In other words, a living system doesn’t seem to need (human) intelligence, conscious intentions, deliberate choices, explicit internal (mental) representations of goals (absent in lower animals, plants, and bacteria), or even a nervous system to display goal-directed behaviour. The origin of goal-directedness thus seems to be a much more fundamental question than the origin of consciousness. Consciousness is likely rooted in goal-directedness, instead of the other way around. One hypothesis is that consciousness even evolved in the service of goal-directedness, as an extension or outgrowth of it, enabling greater control and flexibility in the pursuit of goals.
Lastly, a solid scenario for the origin of goal-directedness would help us to better understand humans as autonomous, autopoietic agents who want to maintain and develop their identity within a complex, challenging environment by using various cybernetic and resilience strategies (PAPER 2023). Investigating these strategies can help to understand human behaviour and the origin of cooperative social systems, which can be seen as self-maintaining wholes with a common purpose emerging from the synergy between internal networks of interactions.
Please find more info about the project ’The Origins of Goal-directedness’ here.