Metacorrelation: a reimagining of cause-and-effect as cultural narrative
Lately I found that cause-and-effect seems a) very ingrained in both intuitive and scientific models of the universe and b) rather a strong property to impose on the universe.
I’m more of the purist scientific school of ‘nothing can be certainly known’, and ‘the universe could upset even your strongest theories at any moment’ (ie, all theory is provisional, non-universal and subject to revision in light of new observations). Those attitudes don’t sit at all well with something as prescriptive as cause-and-effect, which seems to knit the elements of the universe into some sort of a deterministic clockwork mechanism.
Correlation on the other hand, is a nice tool: it doesn’t posit anything about the nature of what is being observed, it merely highlights patterns in observations. Correlation doesn’t make any assumptions about reality, it merely requires that observations can be made. In contrast, cause-and-effect assumes that reality obeys some orderly narrative, and it’s that assumed property (conformance to orderly narrative) that inevitably gives rise to the correlations we observe.
Cause-and-effect, being a stronger property, is on the face of it a more useful property. A good observed correlation between A and B only allows me to know that A and B were correlated in the past. With a leap of faith, I might imagine that they might be correlated in the future. But I don’t know whether controlling A will affect B, or vice versa, or neither, or both. It doesn’t tell me anything about anything that isn’t A or B. In contrast, a good cause-and-effect theory gives me a nice intuitive story (if my knowledge creation framework is ergonomic) with which to understand richly how A and B are related, including how they are related to other phenomena, how controlling various variables affects other variables etc.
That is, cause-and-effect theories are stories created from many observed correlations. When many observed correlations are simplified into a cause-and-effect theory, it is usual that the simplified theory is consistent with more than just the original observed correlations. This is the predictive property of cause-and-effect theories, and is one of the primary advantages of cause-and-effect theories over plain correlation: prediction gives us hints as to where to find previously unobserved correlations (as opposed to only be able to hope for recurrences of previously observed correlations). Without cause-and-effect theories we’d have nothing but hunches and intuition with which to look for other interesting correlations.
However, a big drawback of cause-and-effect theories is that they draw us into thinking that reality OBEYS the theory, when in fact the theory is really just our way of simplifying a lot of observations into a rule of thumb. If we fall into the trap of thinking that reality OBEYS the theory, we easily lose sight of the possibility that the universe could sidestep our theory and behave (seemingly or actually) in contradiction to it.
So, we can get all the benefit of cause-and-effect with none of the blinkered shortsightedness of it by simply recognising that a cause-and-effect theory is merely one way (out of unlimited ways, depending on how you observe your observations) of compressing multiple observed correlations into a convenient story: essentially a correlation of correlations: a metacorrelation.
So, a model for theorising, avoiding imposing any properties on an underlying reality:
- observe correlations
- describing all those correlations longhand is unwieldy, so apply descriptive compression to observed correlations to obtain metacorrelation. Generate multiple compressions, including mutually exclusive ones.
- see what other correlations drop out of the previously synthesized metacorrelations, that weren’t in the set of input correlations: these other correlations are predictions
- test the predictions to obtain new correlations
- rinse and repeat to obtain more knowledge of your observations, to inform your actions (which otherwise would just be uninformed flailing in a meaningless sea of undifferentiated observation, ah, like very many actions that I see!)
Another key aspect of the above model is that it has plenty of room for mutually exclusive theories. I think intolerance to noticed contradictions or equally descriptive theories is a big hindrance for most knowledge institutions (eg science): it reduces the opportunity for new ideas and cross-fertilization between apparently incompatible ideas. Also, intolerance to ideas in general means ideas have to exist more serially, rather than in parallel, which limits the population of ideas.
And clearly one of my goals here is to remove the need for an underlying reality. I find a world composed entirely of observation a much neater and compelling model of the world. Occam’s razor says get rid of extraneous elements in your theory, perhaps an underlying reality is an extraneous element. I think it is.
Ramble. Over.
