Information Function Biology: The Blog

It’s a happy coincidence when people you know from entirely different contexts, turn out to be friends or working together in a surprise connection. So it is with this important paper from Henry Potter, George Ellis and Kevin Mitchell and I am still marvelling at the connection between the Dublin mind specialists (Henry and Kevin) and the South African cosmologist (George).
Their paper is generally to be welcomed, though I also think it is a shame they had to spell out what is already well established in science (see e.g. Del Santo and Gisin, 2019 and other works they cite). I suppose they had to do this for the benefit of those philosophers writing about free will who had not yet accepted it. To reiterate their point for emphasis -- the universe is not deterministic and physics tells us why it cannot be.

Freedom and information

I was also pleased to see their explicit rejection of randomness as a source of freedom - to be random is to have no influence at all on events and in fact randomness is just as constraining as determinism since it disallows anything other than what is random (and thereby devoid of information). A person who lives by dice has enslaved themselves to chance and has no freedom at all.

The authors focus is on the freedom of free will, so we should not expect them to attend to the ‘will’ of free will on this occasion, but I do have some things to say on that point later. Their main objective was to bring an end to the erroneous idea that physics tells us physical reality is causally predetermined, in the sense that every event has a cause from which it is uniquely specified. We know that is not the case, even though hindsight can make it look that way: there is a directed tree (acyclic graph) of causes that, if traced in reverse, would lead all the way back to the ‘big bang’ [1].  As many would expect, I particularly approve of their information-based argument that the early universe could not have embodied sufficient information to support pre-determination. They point out that information is the difference between maximum potential entropy and the realised entropy of a system (the whole universe in this case) and that has been increasing throughout the history of the universe. Put simply (though they point to more sophisticated physical arguments that extend this to any system at any time), the entropy of the early universe was a lot less than it is now, so the universe of the distant past never had the physical capacity to represent (in any form) the universe of the present. Indeed as the universe expanded, it gained entropy and some of that has been ‘converted’ into information, i.e. random assemblies of matter and energy have been redistributed into organised (spatiotemporally autocorrelated) assemblies, which are  “embodied information” according to Farnsworth, 2022; Farnsworth, 2025a.

Perhaps the most profound of all the physical truths highlighted by the authors is that the physical universe, though obeying the laws of physics, is not determined by them. For example, the laws are time-symmetrical, but that does not impose time-symmetry on what actually happens (only on the way it happens). Here, I think they missed an opportunity to refer to the important insight of Howard Pattee who explicitly separated the general laws of physics from the particular constraints on their application (see e.g. Pattee,1969; 2001). These constraints are the initial and boundary conditions that apply to all physical systems. Of course initial and boundary conditions have always been understood as a necessary and separate part of any description of a particular physical system. Pattee’s insight was to recognise the role of these as intrinsic to the system (not just a description of it), with the role of storing and transferring semiotic content (how a molecule becomes a message). Under the control of such constraints, fundamental forces act in ways that are determined by the geometry of particle locations. Their force fields all emanate from fundamental particles that have specific locations in spacetime, so whilst the forces do not change, their actions do; they depend on where every particle is in relation to every other, at any particular time. This ‘geometric’ particularisation is precisely what is meant by embodied information and for that reason, I have interpreted efficient cause (physical cause) as the action of forces constrained by physical information (Farnsworth, 2022; Farnsworth, 2025a).

This is the same (physical) information as has been growing through the history of the universe. Information is (necessarily) physically embodied by particularising the spatiotemporal arrangement of matter and energy (e.g. the difference between an isotropic gas cloud and a star or galaxy - as referred to in their Figure 1). This sort of information is not just important to free will, it is the source of it and everything else that happens -- read on. 

Whatever happens next depends on exactly where everything is now and that, dear reader, is embodied information.

From freedom to will

Potter et al. recast the problem of free will in a wonderfully succinct question: not where does freedom come from, but, in a naturally underdetermined world, where does the control come from? In their words: “The appropriate focus is not on whether an agent ‘could have done otherwise’ than they did; but, rather, on how they managed to prevent all of the other physically possible ‘otherwises’ from happening, such that they were able to do what they wanted to do or what they chose to do.” This moves the focus away from freedom and towards ‘will’.

In an intrinsically under-determined world, the problem for an agent is what to do with all that freedom; what to do that makes sense for the agent. The authors pose this as the question to be more fruitfully addressed, they do not, so far, offer an answer. We should add the question of what makes sense for the agent (Potter et al said “what it chose to do”), not to restrict it to some sort of rational calculation, but at least to be explicable, else it might as well be random.

This question of an explicable selection from the set of possible actions made available by freedom depends on there being an agent that could, hypothetically at least, account for the choice. The existence of this agent is assumed in the argument of Potter et al. 2026.

But it is the identification of this agent that I think is the key to understanding the ‘will’ part of free will. The question of ‘will’ is essentially a question of attribution and that is the link with accountability, responsibility and therefore the real reason many philosophers show an interest in this subject. To put it starkly, if a person is convicted of an offence and it turns out that science can explain their criminality in terms of genetic predisposition, life events leading up to it, physiological and psychological state just prior, etc., then are they really to blame? This is more or less the argument laid out in ‘Behave’ (2017) by Robert Sapolsky, who concludes as a free will skeptic. But as Potter et al. point out, explicable is different from inevitable. The key question, then, is what is the agent of the crime that is being held to account and even is there an agent at all? Even if underdetermination leaves room for choice of action, is there really anything choosing?

Defining Agency: how things can do it for themselves.

There is a risk of tautology as we try to define ‘agent’. For example the Stanford Encyclopaedia of Philosophy (Schlosser, 2019) defines agent as a ‘being’ with the capacity to act, meaning it is able to exert causation under the control of its intentions. So an agent is something capable of agent causation, which is causation attributable to an agent and there lies the tautology. We need a definition of agent separate from its action, so we can then examine action and intention as additional properties that result in agent causation.

If we consider the metaphor for causation as an acyclic directed graph (classically the forked path, but updated by Potter et al. to a diffusion process), then an agent will be something necessary to account for the particular paths taken. Calling that a ‘being’ has implications, e.g. of life, but this is too vague and seems to introduce assumptions without justification. The branching path of philosophical thinking has a parallel in the theoretical physics of quantum gravity: the causal set [2]. I am using causal set theory as an inspiration for the following.

Looking at the diagram above, every dot is an event that could potentially happen and there are many potential paths from event A to event B (e.g. pale blue arrows). An event only actually happens if there is a realised causal path leading to it (black arrows). Events are separated in time on the horizontal axis, so the green box selects all potential events at a particular time. Starting from event A (assumed to have actually happened), it has a broadening range of consequences in its future, limited by the speed of light, depicted as the ‘light cone’ with blue shading (everything in the shaded area is disallowed by Special Relativity: the edge opens up as the vertical dimension of the white space increases at the speed of light). Event B, also assumed to have happened, has an expanding light cone of possible pasts (yellow shading). What actually happened could have been any continuous path leading from A to B in the white diamond. As depicted only one event happens at the selected time, but in general, at any given time there could be a multitude of them, selected from among all possible (lots of things happen simultaneously). If they were unrelated, then there is no possibility of agency. However, if we find that some events at time t make other events, either in the future or past of the causal diamond (white area) more (or less) likely than would be expected at random, then there is a correlation among these events. For that to be true, there must be mutual information among them and that requires there to be an informational structure that transcends, i.e. exists at a higher spatiotemporal scale than, the events themselves. This is what to look for in the hunt for an agent.   

A coalition of causes, meaning that they act in some organised way together, is difficult to explain using causal set theory (see e.g. Yurchenko, 2023 for his criticism of such an idea), though it does allow for it (including exotic phenomena such as quantum entanglement). If we think of multiple paths causing multiple simultaneous events that are correlated, the corresponding mutual information can be decomposed as redundant (more than one event carries the same information in its probability), unique (not really mutual at all) and synergistic (information that appears only when the probabilities of all the events are combined) - See Farnsworth 2025a for further explanation and illustrations. Unique and redundant information don’t add anything, so the only room here for the agent is in synergistic information. That makes sense because, by definition, this component of mutual information exists only at an organisational level above that of the event.  An agent is therefore information distributed as synergistic among the force carrying particles that produce physical causation. Recalling that all physical causation results from the constraint of physical forces (emanating from fundamental particles) by physical information, the agent is the additional physical information that is shared among the fundamental particles and the only way this can exist is through a particular configuration of particles persisting by mutual reinforcement (see Farnsworth 2025a for further explanation of that too). More precisely, since this (shared) mutual information includes redundant and unique information which can be accounted for without invoking a new concept such as ‘agent’, the only remaining component - termed synergistic information - is the agent (Farnsworth 2025a again). This is a rather weird phenomenon requiring a good explanation.

In general, there is no good reason to associate a set of causes in a time slice with one another, except that they may originate some time earlier in a single event. Mutual information does occur naturally, but tends to decay away under thermal noise etc.. However, theoretical biology has identified a peculiar causal structure that so far we have only identified in living organisms - that is closure to efficient causation (also called organisational closure). This structure creates and maintains mutual information among components of a living organism. Indeed, only in living organisms is every part (e.g. biomolecule) both responsible for the existence of the others and depends for its existence on those others. This association is the best justification for the existence of agents in the universe and the implication is that every living organism (including single cells such as bacteria, but not viruses (Farnsworth 2021)) is an agent. Poria Azadi has provided a formal analysis leading to what I think is the same conclusion (Azadi 2025). This definition of agent is independent of intention [3], so we can now add that as a qualifier to obtain the ‘philosophical definition’ of agency.

Why Intention is the fulcrum of free will

Intention is where the philosophical question of ‘could do otherwise’ is to be found: an intentional act is one chosen from a set of more than one option. Of course it has to be the agent ‘making the choice’ and if a choice is made in the physical world, then it too is a physical act: the result of a force-mediated arrangement of matter/energy in spacetime. A choice is a selection of one spatiotemporal pattern over others that is caused by embodied information constraining the action of physical forces. For that to be associated with an agent, then the information that physically contributes to the cause of the choice must be part of the agent. For the choice to be intentional in the sense of being explicable in teleonomic terms as in some way perceived by the agent, or likely to be, beneficial to the agent, it must at least be non random, i.e. it must be informed. So at least some of the synergistic information we are calling the agent must be the deciding constraint on which outcome (the choice) occurs, before we can call the determination agent causation. For that to be ‘intentional’ it further has to be an outcome that, on a reasonable assessment of all the alternatives, gives a relative benefit to the agent. The only benefit that makes sense at this level of scrutiny is one that makes the persistence of the agent more likely.

Every living organism incorporates a multitude of causal mechanisms that fulfil those requirements for intention. These are the homeostatic set-points of the organism that regulate its behaviour in every way (see Farnsworth 2025b). Although homeostasis is not a mental act, it does qualify as intentional action and can be located within an agent, using the definitions I am favouring (for their independence of assumptions about brains and minds). A homeostatic set-point is a piece of information that determines a choice among options and is embodied within a system that has closure to efficient causation and therefore qualifies as an agent.

For an organism with a brain, we could say the role of the brain is to keep the body alive and flourishing, the role of the body is to keep the brain alive and flourishing. Both tasks depend upon a large battery of homeostatic set points, each a little bit of personal information, making the organism unique and for which it can therefore be held accountable. In a somewhat comically reductionist sense, Elvis was a collection of set points that collectively contributed to all of his behaviour, along with other information that he embodied. The point is they were his because he was an agent because he was closed to efficient causation. Elvis was emphatically not the material by which these informational structures were embodied: as the continuity between the youthful and the old Elvis demonstrates (it’s the same Elvis, though all the atoms are different ones). 

You cannot have free will without constraints

Current physical theory tells us that the universe is fundamentally random, but the original small amount of anisotropy in its form has been reinforced by the action of fundamental forces to produce order in the form of atoms, molecules, galaxies, stars, solid objects such as planets and the geochemical processes that unfold on and within many of these. All that order is information embodied in the distribution of matter and energy. Every case of information is a case of constraint, since it specifies one particular spatiotemporal distribution from among all the possibilities. Among the geochemical processes, on at least one planet, a remarkable phenomenon emerged: closure to efficient causation, whereby every part of a complex system owed its existence to every other part and also was instrumental in the existence of every other part. That phenomenon makes it possible to associate particular bits of information with a subsystem of the universe - like a little spiral loop in the tree of causal relations, an agent is a self-sustaining package of synergistic information that includes the constraints making its persistence more likely. Human beings are a (rather fabulous) elaboration of that, to the point that they can support additional layers of informational constraint through anticipation based on mental modelling of themselves and their environment (including social). All this is a very elaborate arrangement of constraints on physical forces that in a causal sense belongs to the agent that embodies them. The resulting complex of information and its mutual interactions is beyond normal comprehension, but we have developed a sense of it, which we call our free will. What we feel it to be is a metaphor, but make no mistake, underlying that, free will is physical and real.

Notes.

[1] Tracing back should lead to the big bang, but that is only because the paths have already been chosen by the time that tracing back can implemented. This is one of the deep differences between forwards and backwards in time (a temporal asymmetry) - hindsight is deterministic because it has already happened, but foresight can be no more than probabilistic because causal outcomes at each junction are underdetermined. This was true from the beginning - since the big bang (predicted by General Relativity) actually ‘breaks’ spacetime, efforts to explain it concentrate on quantum gravity and one very promising avenue for that is causal set theory and there, all possible outcomes from every cause are equally probable at the elemental scale (see e.g. Yazdi, 2024). More prosaically and macroscopically, the weather (often cited in this context) is typical, rather than unusual in its underdetermined dynamics which can only be known in hindsight.

[2] Not surprisingly, its meaning is hard to grasp as it was developed in an effort to explain fundamental processes at the extremes where Einstein’s General Relativity runs into trouble - specifically singularities (e.g. black holes) where spacetime looses its meaning, the mother of all of them being the (still hypothetical) big bang that started the universe as we know it. Causal sets allow gravity to be quantised (the other fundamental forces are already well treated that way), eliminating the infinities by quantising spacetime. Thanks to Sergei Yurchenko for introducing me to this idea.

[3] This definition of agent is also consistent with, but more deeply derived than, the so called “minimal agency” of Barandiaran et al. (2009). They considered an agent as a entity unified by producing and maintaining itself and capable of independent action that accords with a normative goal. The only way an entity can be ‘unified’ is by its parts sharing mutual information and the only way it can be causally distinguished from its environment is by having closure to efficient causation. The requirement for a normative goal is what is meant by intention and Barandiaran et al. (2009) admit homeostatic control within a bacterium as qualifying for normative action.

References

Azadi, P. Computational irreducibility as the foundation of agency: A formal model connecting undecidability to autonomous behavior in complex systems. BioSystems 256, 105563. doi:10.1016/j.biosystems.2025.105563.

Barandiaran, X.E., E. Di Paolo, and M. Rohde, 2009. Defining Agency: Individuality, Normativity, Asymmetry, and Spatio-Temporality in Action. Adaptive Behavior 17(5): 367–386.

Del Santo, F. and Gisin, N. 2019. Physics without determinism: Alternative interpretations of classical physics. Phys. Rev. A, 100. 6, 062107. doi:10.1103/Phys-RevA.100.062107.

Farnsworth, K.D., 2021. An organisational systems-biology view of viruses explains why they are not alive. BioSystems 200, 104324. doi:0.1016/j.biosystems.2020.104324.

Farnsworth, K.D., 2022. How an information perspective helps overcome the challenge of biology to physics. BioSystems 217 doi:10.1016/j.biosystems.2022.104683.

Farnsworth, K.D., 2025a. How physical information underlies causation and the emergence of systems at all biological levels. Acta Biotheoretica 73, 6. doi:10.1007/s10441-025-09495-3.

Farnsworth, K.D., 2025b. Homeostatic set-points are physical and foundational to organism autonomy. BioSystems 258, 105634. doi: 10.1016/j.biosystems.2025.105634.

Pattee, H.H., 1969. How does a molecule become a message? Dev. Biol. Supp. 3, 1–16.
doi:10.1007/978-94-007-5161-3_3.

Pattee, H.H., 2001. The physics of symbols: bridging the epistemic cut. BioSystems 60, 5–21. doi:10.1016/S0303-2647(01)00104-6.

Potter, H.D.,  Ellis, G.F.E. and Mitchell, K., 2026.  Reframing the free will debate: the universe is not deterministic. Synthese  207.71. doi:10.1007/s11229-026-05455-7.

Sapolsky, R. 2017. Behave: The Biology of Humans at Our Best and Worst. Penguin Press. ISBN: 978-1-59420-507-1

Schlosser, M. 2019. Agency - in - The Stanford Encyclopaedia of Philosophy. https://plato.stanford.edu/entries/agency.

Yazdi, Y. K. 2024. Everything you always wanted to know about how causal set theory can help with open questions in cosmology, but were afraid to ask. Mod. Phys. Lett. A. 39.01, 2330003. doi:10.1142/S0217732323300033.

Yurchenko, S.B. 2023. Is information the other face of causation in biological systems? Biosystems 229, 104925. doi:10.1016/j.biosystems.2023.104925.