The study of animal and human behavior is fraught with quantitative potholes. Months ago I explored these with a fellow researcher and was put back into place simply by my lack of knowledge. However, after months of study where I understand a smidge more than I knew thing (not much more because I learned how much more I don’t know!), I’m confident in my identified modeling potholes.
In the majority of behavioral experiments and studies researchers plot and analyze stimuli, responses, response rates and some physiological/biological events. The stimuli can be classified in various ways and same with the responses and biological events. Common terms used are consequences, reinforcers, punishers, fixed action patterns, action potential, conditioned responses, s-r, s-s, discriminant stimulus… to name a few.
And now the rub.
What exactly are these? Peeling back the layers, what’s underneath these? what are their discrete quantifiable parts?
Let me explain… when we measure a response, we consider the whole of that “response.” Consider the classic examples of pigeon pecking. We measure the “pecking behavior” rates. It can be a quick peck, a succession of pecks, hard peck, soft peck and so on. In most studies what researches are looking for is the relationship (not necessarily the quantitative/magnitude relationship) between the pecking behavior and various stimuli or reinforcers/punishers. Typically a simple graph of time vs. rate of response is plotted to show strengthening or reduction in the rate of response.
All of this is very good for verifying that there is a relationship between variables. However, it has not been helpful at all for defining a reliable, falsifiable, quantitate law like F=MA or the wave equation or anything. It these approach never can deliver that! Why? because what you are studying is quantitatively subjective. You have no way to measure pecking behavior versus drinking behavior versus fighting and so on. You can never arrive at a generalized rule that says if I use a VI 6 schedule of reinforcement I always generate a response rate curve of 1/t^2 or something similar ( rate of response drops off as inverse square of the time).
This makes it impossible to apply one researchers finding to another situation other than in relational terms like “we know variable interval schedules tend to increase the time to behavior extinction”. I can’t say by how much, how long.
Up until recently this was ok because for the last 100 years we didn’t even know, understand and appreciate the fundamentals of how we learn and adapt behaviorally.
And make no mistake this same problem exists in producing quantitative models of natural selection. What exactly are you measuring with variation in a species? How do you assign some magnitude of variation? How do you compare to variations and their impact? This green frog is X different from this blue frog…. X is????
I’m not sure it’s even possible logically much less computationally to get at models of natural selection and human behavior. These are more “organizing principle” type theories/laws. This is where my brain goes haywire.
Genes mutate. We can see the consequences play out. We can measure environmental variables, population densities, genetic expression and pretty much every aspect involved in playing out natural selection. Yes, we can’t really look at a species and individuals in a species and apply an equation to predict it’s proliferation or extinction.
And so it is with behavior. I can measure responses, catalogue stimuli, map neural networks, and plot the consequences. I cannot look at an organism and predict with numerical precision how it will behavior – when, where, how much.
I can do this relatively. I can make probability statements but not even those end up like the nice wave equation in quantum mechanics. The probability statements are not reliable from organism to organism.
Part of this trouble comes from the huge variation between organisms (biological, environmental). We can control much more strongly the stimuli (then again, the stimuli is relative to the observer, oy.)
Some ask why care about this at all? So we can’t predict natural selection or human behavior, what difference does that make. Besides that’s one of those intractable problems, Russ. You are better off just observing how it plays out than coming up with computational models that will never end. (computational irreducibility in some sense). I agree on a macro level. We’ll never be able to simulate / predict human behavior faster/more efficiently than letting it play out. Any thing we do to get that efficiency will cost us in detail or clarity (which in behavior and natural selection can produce critical modeling errors down the road!).
We can’t really produce absolute models in other disciplines either. We can pick out relatively simple situations (meaning – we can isolate very small subsets of the overall picture with lots of assumptions and controls). That said, even isolated sub systems can provide profound and useful insights and helpful predictions.
That’s what I’m looking for in behavior. An isolated subsystem I can model. Something I can figured out and repeatedly, consistently and quantitatively measure over and over. To do even that, there must be a unit of behavior and unit for stimulus or something to that effect. Without a unit we can take from one situation to the next, we’re left with a description of relationships.