Interventionist/Manipulationist models of causation (e.g. Pearl, 2000; Woodward, 2003) are rooted in the intuition that if some event A causes some event B, then one should be able to manipulate A in some way and see corresponding changes in B after changing A. If A is modeled as causing B, then there should be an intervention on A (in at least some state of the model) that results in B changing its value. These kinds of models of causation basically describe what scientists already do to determine causal relationships among variables. Scientists have for centuries tried to control for all independent variables (Woodward calls this “screening off” the other variables besides A that likely partially cause B, by holding their values constant) save one, A, which they vary, in order to see the consequences or changes expressed by some outcome or dependent variable B. If B changes with an intervention on A, it is concluded that A causes B.
A counterfactual formulation of interventionism would be: "If A had not occurred, with all screened off variables that may cause B held constant, then B would not have occurred." A core point of criterial causation is that we need to enhance the interventionist account by saying : "If A had not occurred, with all screened off variables that may cause B held constant, **and with the parameters by which B evaluates its inputs also held constant**, then B would not have occurred."
To sum up, standard interventionist models of causation carry out some intervention on A to determine what effects, if any, there might be on B (and other variables). If instead of manipulating A, or A’s output to B, however, we instead manipulate the criteria, parameters or conditions that B places its input (including on input from A), which must be satisfied before B changes or acts, then changes in B do not follow passively from changes in A as they would if A and B were, say, billiard balls. Inputs from A can be identical, but in one case B changes in response to A, and in another case it does not, depending on B's criteria for responding. This reparameterization of B is what neurons do when they change each other's synaptic weights, such that a neuron now responds optimally to different inputs than prior to the act of reparameterization or criterial resetting. Criterial causation emphasizes that what can vary is either outputs from A to other nodes (the traditional and I would say incomplete view of causation), or how inputs are decoded by receiving nodes B, B’, B’’ and so on. On this view, standard interventionist (hearkening all the way back to ‘Newtonian’ models of causation that emphasize energy transfer and conservation; e.g. P. Dowe’s views (1992)) are a special case where B places no particular conditions on input from A that have to be met before B changes state. But the brain, if anything, emphasizes causation via reparameterization of B, by, for example, rapidly changing synaptic weights on post-synaptic neurons. Let me emphasize that I do not think that Woodward or Pearl are wrong. But they also make no mention, as far as I can tell, that causation might partly depend on reparameterizations of B.
Changing the code or parameters or criteria that B uses to decode, interpret or respond to input is a manipulation that might make no apparent changes to A or any other variable in the system for long and uncertain spans of time, until just the right pattern of inputs comes along. For example, the Mossad might program a cellphone to explode only when a particular phone number, known 'only' to their target, is dialed. Manipulating ‘A’ here appears to have no effect on any dependent variable ‘B’ and might not, in principle, for as long as you like (think of a booby trap in a king's tomb set by the pharaoh’s builders that only kills archaeologists millennia later). It might take years to get this phone into the hands of their targeted Hamas leader. But when the bombmaker dials his ‘secret’ number to call his uncle in Paris, his head is blown off. This kind of reparameterization of B need not have immediate noticeable or measurable effects within the system, so seems to violate the assumption on the traditional view that causation is transferred at some fast speed (say the speed of light). But reparameterization of B is a causal intervention nonetheless, even though this subclass of causation has been relatively ignored so far.
I believe, however, that it was the ‘discovery’ of this class of causation by evolution that really led to the explosion of physical systems that we now call biological systems. Once causation by reparameterization came not only to involve conditions placed on physical parameters (e.g. molecular shape of a neurotransmitter before an ion channel would open in a cell membrane), but also conditions placed on informational parameters (fire iff the criteria for a face are met in the input) that were realized in physical parameters (fire iff the criteria on the simultaneity of spike inputs are met), we witnessed a further revolution in natural causation. This was the revolutionary emergence of mind and informational causation in the universe, as far as we know, uniquely on Earth, and perhaps for the first time in the history of the universe.
Recent Comments