To cut to the chase, I'm now pretty happy that a perception-action analysis is appropriate at this particular cellular level. I set a high bar for this (mostly by reading Turvey & Carello papers, which should illustrate that height pretty clearly :) but it seems clear the cells are behaving with respect to information, and not simply being buffeted by forces. Applying some key criteria, and resting on the hard work of Turvey & Carello showing that intelligence isn't about brains but about behaviour, I will claim here that Bentley's endothelial cells are agents that exhibit intelligent behaviour, and there is a clear need for a behavioural scale contribution to any explanation of that behavior.
Do we need an ecological-scale analysis to explain cell behaviour?
This question is about whether there is a need for an ecological scale analysis. As we wrote in the Ecological Representations paper (pg 236),It is relatively uncomplicated to tell a causal story about simple, mechanically linked events. If we are sitting in a tree and the branch we are sitting on breaks, the force of gravity causes us to be displaced. ... It is more challenging when events are not so obviously mechanically linked. For instance, if we move because a branch falls nearby, or because it is windy and we are worried that a branch might fall, what is it that causes our behavior to change? In this example, there was no mechanical linkage between the accelerating tree branch and our body; in the latter case, there was not even an accelerating tree branch at all, just the worry of one. Yet, we moved. Obviously, we do not move by accident or by magic. There are reasons and explanations, but these reside in a psychological level of explanation, which is why we need a psychological theory to fill the gap.The relevant entity that crosses the gap is ecological information, of course. In Turvey's formulation, the organism moves, not because it was affected by a sufficiently large force, but because it interacted with a 'relatively low-energy media' which contained structure that shaped the behavior. I am not literally buffeted by light energy into doing something; I am in no danger of being knocked over or moved by ambient light. But it is still a key part in the causal chain of events that leads me to behave one way, rather than another.
The endothelial cells Bentley studies change their form (growing and retracting filopodia and lamellapodia), they change location (they migrate towards the area where new cells are required), and they change state (from stalk to tip cells and back again). Exactly none of this happens because of physical forces. All of it happens (primarily) in response to variations in the distribution of vascular endothelial growth factor (VEGF), which leads to various changes in cell processes, which in turn alter access to the VEGF distribution, and so on in a loop. That VEGF distribution is caused by changes in cell dynamics (related to, for example, hypoxia) in cells that are distant from the endothelial cells; no mechanical contact. From the first-person perspective of the endothelial cell, the relevant dynamical event it needs to organize it's behavior with respect to is 'over there', in exactly the same way as it is generally true for human-scale organisms. Equally importantly, as I reviewed last time (see the post and Bentley & Chakravarula, 2017), these changes in cellular processes are not solely the result of the action of a internally clocked, central pattern generator triggered by the simple detection of VEGF. As I wrote,
...it's clear that angiogenesis is the kind of self-organising, emergent dynamic operating under multiple constraints that we observe at the scale of organism behavior. While one of those constraints is internal (the CPG, analogous to the brain being in charge), the differential adhesion constraint is clearly an embodiment constraint, and there is the clear potential for the VEGF constraint to be an information constraint that requires active perception.So, there is a gap in any explanation of the cell's behavior that cannot be filled by physical forces or an internal process. There is a need to invoke behavioural-scale entities, such as cell embodiment and perceptual information. There is room for an ecological analysis.
Hey, but cells don't have nervous systems!
It would never really occur to the average non-ecological psychologist to claim that a cognitive process like perception or action is occurring here, because there's nothing to implement the required representations of those distant events. Specifically, there's no nervous system, or any analog of one. These cells do live in organisms with nervous systems (humans), but this particular behavior is not under neural control. So why invoke anything like a psychological scale process here?Obviously, the ecological approach is not brain-centric. Nervous systems are cool, and it could be the case that they are necessary before invoking something like perception, but the ecological approach would never assume that this is true until it had ruled it in explicitly. Luckily for me, Turvey and Carello (in particular) have spent time doing their usual excellent work considering this issue, and (no surprises) find no reason to think intelligent behavior needs neurons.Plant Intelligence
Carello et al (2012) analyse the necessity of nervous systems for intelligent behavior in the specific context of plants. It's become quite trendy in biology to talk about plants as intelligent, but of course most of the work is spent looking for plant analogues of brains. Carello et al instead focus on plant behaviour, and ask questions about whether it can be explained by a mechanical story (like moving because the branch broke) or whether it requires behavioural-scale entities (like moving because you are worried the branch might break). She discusses numerous examples where it's clear plants are not simply 'buffeted by forces'.Cellular Intelligence
Turvey's new favorite thing to talk about is a Difflugia, a single-cell protist that builds a shell out of rocks. Difflugia selects appropriately sized rocks, and selects rocks with different properties for different stages of the shell construction. Like plants, it's behavior cannot be explained by a simple mechanical, physical story or one invoking any central process.Intelligence
Intelligence, from their ecological point of view, can be defined as 'end-directed behavior marked by the making of meaningful distinctions made possible by perception-action cycles'. I've already laid out above that the endothelial cells seem to do this. They then add some more specific requirements for the behavior to be considered to have agency; prospectivity, retrospectivity, and flexibility (Turvey & Carello, 2012). ProspectivityThis asks whether the behavior in question is future-oriented (note, as per the outfielder problem, this does not necessarily entail prediction!). They show plants do this, and endothelial cells clearly do this too. Specifically, their behavior is organised with respect to making possible a future state where they can grow new cells in the required location. Their behavior makes no sense without understanding this fact.Retrospectivity
This asks whether the behavior is shaped by things that have happened in the past (effectively, does the system have a memory?). Plants show this, as do the cells. For example, how they act now reflects where they have been, and what they did while there. Their sensitivity to VEGF changes over time in response to lateral inhibition from neighbouring cells also trying to alter their sensitivity to VEGF, and the answer to the question 'which cell becomes a tip cell?' depends on when you ask. So there is memory in the system, although as with plants, this doesn't require a nervous system storing anything; the memory is embodied in the temporal dynamics of the cell behavior. This should remind you of Smith & Thelen's analysis of the A-not-B error, which is going to inspire some specific experiments and modelling work. More on this in the next post. Flexibility
If there aren't at least two behavioural options for a system, no-one would want to call that an intelligent agent. When the branch under you breaks, you have no choice except to begin falling and that's all that's required to explain why you moved that way. Of course, because you are an agent, you can allow the fall, try to catch yourself, etc, and so any explanation of those behaviours requires a behavioural-scale explanation. A simple example from the cells is that they are working towards being in one of two states (tip cell or stalk cell) and this 'decision' happens in a single context. We've already established that the decision isn't the compulsory result of an internal or physical process, and so we have minimal flexibility to explain.
