Reflections on the Irruption, Computation and Anticipation Workshop

On December 23rd, 2024 I attended the event Irruption, Computation and Anticipation, a discussion led by Professor Tom Froese, who has recently proposed a new framework called Irruption Theory (I got most of the quotes in this review from that paper). From that workshop I was able to understand the basis and implications of this novel paradigm, which I noticed is compatible with multiple existing narratives. But before delving into it, let’s start at the beginning.

The working hypothesis of Irruption Theory is very simple: “the more an agent’s embodied activity is motivated, the less that activity is determined by its material basis”. In order to capture the duality between the motivations and materiality of an agent’s activity, the concept of irruption is introduced. Etymologically, irruption is a descendant of the Latin verb rumpere, which means “to break”. Under this particular context, “irruption is defined as any physiologically unintelligible change in bodily activity associated with a manifestation of agential efficacy, which hence will appear akin to a change in noise correlated with the exertion of volition”. Now we can list the three foundational axioms of Irruption Theory as follows:

1. Motivational Efficacy. An agent’s motivations, as such, make a difference to the material basis of the agent’s behavior.
2. Incomplete Materiality. It is impossible to measure how motivations, as such, make a difference to the material basis of behavior.
3. Underdetermined Materiality. An agent’s behavior is underdetermined by its material basis.

First of all, we must note that there is a certain incompatibility between the first two axioms. If, by Axiom 2, it is in principle not possible to measure the impact of the agent’s motivations, then how is it possible for these motivations to have an impact on the agent’s material basis (Axiom 1)? This disparity arises under the premise of determinism, which is not universally accepted in science (chaotic dynamics and quantum mechanics being two clear examples). This observation leads us to Axiom 3, with which the apparent tension between Axiom 1 and 2 is reduced, since “Axiom 2 can no longer be interpreted in the sense of evidence of absence counting against Axiom 1, given that the material basis itself could never amount to a complete determination of behavior either”.

From these three axioms it can be noted that the Theory of Irruption seeks to avoid dualism. Instead of assuming body and mind are governed by laws of different nature, we accept these explanatory gaps at face value and to take them as positive indications of a complex relation: mind and matter are one, but they are not the same. As has already been well pointed out by Turvey and Shaw almost thirty years ago, the most important conceptual step to be taken in the twenty-first century with respect to the foundations of a scientific attack on “knowing about,” is the rejection of any dualism and the acceptance of mutuality. In this case, irruption theory points out a deeper reciprocity between mind and body.

The deeper we get into the world of Irruption Theory, the more questions come to mind. How can an irruption be quantified? How can an irruption make a difference to behavior? How can an irruption lead to appropriate behavior? In response to these questions, irruption theory proposes three theses, each of these draws on existing research programs in embodied and enactive cognitive science.

1. Irruption Thesis. The living body is organized as an incomplete system such that it is open to involvement of motivations via increased material underdetermination.
2. Scalability Thesis. The living body is organized as a poised system such that it amplifies microscopic irruptions to macroscopic fluctuations that can impact behavior.
3. Attunement Thesis. The living body is organized as an attuned system such that it responds to scaled up irruptions in a context-sensitive and adaptive manner.

The first thesis suggests that irruptions can be indirectly quantified in terms of their unpredictability. This allows us to connect irruption theory to information-theoretic measures of entropy for neural, physiological, and behavioral signals. Since there is growing evidence that information-theoretic entropy of brain activity is associated with all kinds of motivated activity, irruption could help us to exclude some of the many neural correlates of consciousness (based on entropy) that exist.

Observe that so far irruption is suggested to occur at small scales. But living organisms evolve on multiscale complexes. Thus, the second thesis suggests the existence of mechanisms operating in the whole organism that prevent their underdetermining effects from being washed out by large-scale material factors; some kind of amplification is needed. As mentioned by Tom, this of course could be connected with the use of 1/f noise in living systems for louden signals and transmitting information. “A fitting account of the origins of such dynamics in biological systems is self-organized criticality, whereby a system organizes itself so as to be poised to respond to perturbations in a scale-free manner”.

Also notice that irruptions can only serve to counteract existing material constraints by increasing their underdetermination. Thus, irruptions cannot directly control a motivated activity’s material basis. What the third thesis suggests is that there are constraints which allow the organism to handle directionality. In fact, these constraints have the same nature as those mentioned by Montévil and Mossio, whose mutual dependence and closure explain the organization observed in biological architectures. Thus, it might be that irruption theory can also give us information about the origin and evolution of life. “Once our bodies have become appropriately attuned, by evolution, development, and/or learning, the unfolding of future behavior can then be largely offloaded into the affordances and constraints of the agent–environment system”.

As rightly pointed out by Hölken, Irruption Theory has some conceptual gaps with respect to other multidisciplinary approaches (e.g. complex systems science). In the same spirit, let us discuss some of the potential connections between Irruption Theory, Computation and Anticipation. In general the conception of irruption could be used to quantify the degree of agency in any living creature. However, what about machines? Could we exploit irruption theory to produce a measure of artificial general intelligence? In general the computational devices that we have developed so far are static and they do not adapt to their environment as life does. Thus, under the classic von Neumann architecture, we will not be able to reach any type of agency if we use the tools that Irruption Theory proposes.

Now let us assume that by computation we mean any procedure by which input information is transformed according to predefined rules and turned into output data. In fact, there is a whole branch of computer science called unconventional computing, which is in charge of developing computational structures which go beyond the von Neumann architecture. Under this broader panorama, we can imagine computational devices capable of adapting to their environment, and thus capable of producing irruption. Unfortunately there is some kind of sequential-centrism in the models of computation explored so far. Nowadays almost all our conceptions of computation are based on Turing machines, a mathematical model of computation describing an abstract machine that manipulates symbols on a strip of tape according to a table of rules.

In my opinion, if we want to reach agency by means of irruption in machines, we should start by exploring non-sequential models of computation, which could allow us to capture the recurrent, stochastic and parallel features that we observe in life. During the workshop I also noticed some comments related to the computability of life. It is true that there is no consensus on the computability of life. Artificial life is grounded on the premise that life is computable, which means there exists an algorithm capable of expressing the nature of life. However, since the last century there have been people who rejected this proposition.

Particularly Robert Rosen points to the distinction between modelling and simulation. On one hand, a model of a system incorporates understanding of how the machine works in a congruent way. On the other hand a simulation just produces some of the same behaviour, without necessarily using any mechanisms that replicate what happens in the real system. As pointed out by Gatherer et al., relational biologists do not deny that complex systems can be simulated, but complex systems can never be computed as functions. The full computational analysis of complex biological systems is beyond our current computing capabilities. Indeed, simulation can be an excellent predictive tool for the natural world, but we cannot reduce the semantic aspects of life to pure syntaxis.

Rosen built then what he called (M, R)-systems, a minimal model of organization which, according to him, captures the most essential feature of life: closure to efficient causation. In biological terms, this means that all enzymes, all catalytic processes that are used within that system, need to be synthesized by the organization per se, without recourse to any external agent. Although many to this day claim that Rosen denied the computability of life, the truth is that he never denied the possibility of simulating life, but he was not truly interested in it. Nowadays we can find many attempts of simulating (M, R)-systems using different models of computation, from lambda calculus to sequential machines. This malleability makes me think (M, R)-systems are a good candidate to explore unconventional computing architectures in order to build agential machines. Particularly the idea of representing (M, R)-systems as a set of machines exchanging information could be connected to the paradigm of algorithmic networks, which suggest that it is possible to go beyond the Turing computational barrier using distributed systems made of sequential machines.

Furthermore, recently it was proposed a temporal parametrization of (M, R)-systems such that it captures all the biological and philosophical intuitions regarding the notion of self-organisation and can implement a computation under an iterative scheme. This is very striking, because with such a temporal parameterization, the author connects (M, R)-systems to an important family of self-organization models that assume the free energy principle through active inference, which means it is possible to get inference from (M, R)-systems. This is consistent with a recent paper arguing that autopoiesis, adaptation and anticipation are present in all life forms at all evolutionary scales.

The same author has shown that if we assume that at least a weak version of the free energy principle is true, then we can merge enactivism and computationalism into a single philosophical scheme, which he baptizes as “Computational Enactivism”, and it turns out to be stronger than either computationalism or enactivism on their own. In my opinion these two ideas proposed by Korbak could be useful for the future extension of irruption theory, and they are worth exploring.