“The Map is not the Territory”
For starters, I just want to be up front about the fact that the subjects covered in this post are very vast subjects. There are dozens of very long, technical books written about them, and this post is just a basic introduction. Actually, it’s more like an introduction to an introduction – I’m currently 20,000 words into a book talking about all of this stuff in much more depth, though still not the depth of any one book drilling down really deep into one specific topic.
Also, don’t skip the abstract stuff. There are several concrete examples, but they’ll make more sense if you actually take the time to understand the conceptual bits.
With that out of the way, let’s dive in.
Your body is insanely complex.
Humans, with all of our scientific knowhow and the aid of vast computational power from supercomputers, have just reached the point of being able to model a single cell of the world’s simplest organism. We’re still a long way from having a comprehensive model for a single human cell, let alone modeling, from the bottom up, how individual cells interact, or how entire organs signal back and forth with each other, or how the human brain works in its entirety, or how it interacts with, influences, and is influenced by the other tissues of the body, and how we interact with other complex organisms (each other) and our environment.
We, as a species, know a lot, and we’re quickly learning more every day. But we still have a long way to go to understand all of the workings of a single one of our cells.
Just let that sink in for a moment.
A nihilist, when faced with this realization, would throw his hands in the air and lament, “compared to how much there is to know, we know effectively nothing. There’s no way to understand all of this stuff, so why even try?”
Luckily, I’m not a nihilist, and I think that response is nonsense. Not knowing EVERYTHING doesn’t mean we don’t know anything. Far from it. We know enough to treat many diseases, put a man on the moon, and split the atom. Heck, hundreds of years ago Isaac Newton could describe, with stunning accuracy, how the planets move the way they do with nothing but a telescope and some calculus. We, as humans, are really good at doing a lot with astoundingly little (relatively) information.
But, because we don’t know everything, we have to construct models.
Models are our way of wrapping our minds around complex systems that we don’t know everything about, distilling them down to their most important features, and being able to have a basic idea of how they work and being able to predict how they’ll respond to various challenges (stimuli or stressors).
A good model has three main features:
1) It captures enough of the system’s complexity to be useful in describing how it works and how it will respond.
2) It accounts for few enough factors to actually be user-friendly
3) It actually works
It captures enough of the system’s complexity
I want to use the study Effect of squat depth and barbell load on relative muscular effort in squatting by Bryanton et. Al. (2012) as a lens through which to see this issue.
The study is very straightforward and very well-done.
They researchers measured each subject’s maximal plantar flexion, knee extension, and hip extension strength.
Then the researchers got people to squat with weights from 50% to 90% of their 1rm squat, working up in 10% increments.
They set up a camera directly to the side of the people when they were squatting, and analyzed the net joint moment (NJM) for each joint at each point in the movement. NJM is the minimum amount of torque necessary to keep the joint turning at the observed rate.
Then, with the NMJs determined, they calculated relative muscular effort (RME) – a measure of how much torque is required at each joint, relative to the maximal amount of torque the subjects were capable of producing at that joint.
Here were the results:
Just to break this all down a bit: Squat depth and barbell load affected how much torque was required at the hips. Barbell load had the biggest effect on plantar flexion RME. Depth, but not barbell load, increased knee extensor RME.
Just at face value, without understanding the limitations of this type of research, you’d make a couple surprising conclusions.
- Ankle plantar flexor RME gets a lot closer to 100% than knee extensor RME. This would effectively mean – at face value – that squats work your calves harder than they work your quads.
- Since barbell load didn’t affect knee extensor RME, squats at 50% of your max would train your quads just as hard as squats at 90% of your max.
However, as the authors explain in the discussion section, that’s clearly not what’s going on. Because, while RME provides a useful biomechanical model for analyzing a movement, there are variables it doesn’t account for.
Most saliently here – it’s based on Net Joint Moments which are the minimum amounts of torque required to keep movement going at that joint.
Here’s the problem: we have pesky antagonist muscles that make sure we’re never only having to produce the minimum required torque from our prime movers at any given joint.
For example, with the quads, although RME may be virtually identical between 50% and 90%, since plantar flexor RME and hip extensor RME were increasing as the load increased, that means the gastrocnemii and hamstrings were contracting harder and harder as the load increased.
The gastrocs are plantar flexors, but they’re also knee flexors. The hamstrings are hip extensors, but they’re also knee flexors. So at 90%, although the nice tidy description of the physics of the situation says “the minimum amount of necessary torque at this joint hasn’t changed from 50%,” in reality the quads DID continue having to work harder, because they were fighting against more force from their antagonists.
So, returning to our discussion about the usefulness and drawback of models, we can see that models are only useful insofar as they account for enough complexity to make them a decent enough approximation of what’s actually happening – and that you have to be aware of the limitations of the model you’re using so you don’t come to a silly conclusions like “squats at 50% are just as hard for your quads as squats at 90%.”
In this study, the model took into account three important factors:
1) The pure physics of the situation (net joint moments)
2) How much torque the subjects were maximally capable of producing at each joint
3) The contributions of antagonistic muscles (though not quantified).
Just relying on RME, you come to less accurate conclusions because your model is accounting for fewer factors (only the physics and maximal torque, without taking into account antagonists). You decide that squats are a calf exercise more than a quad exercise.
Just relying on physics, you lose another variable – maximal torque at each joint. You know how much torque is required, but have no idea whether the resulting numbers are big or small because you have nothing (maximal torque at each joint) to compare them to.
As each model accounts for fewer and fewer factors, it manages to account for less and less complexity, and it becomes less and less useful.
In this instance the simplest model – an analysis based purely on physics – is useful for perhaps pointing you in the right general direction of what’s going on, and nothing more.
It accounts for few enough factors to actually be user-friendly
A perfect example here is calories in and calories out.
We all “know” calories in minus calories out equals caloric surplus or deficit equals weight loss.
While this may be “true” from the perspective of pure physics, things are a little fuzzier in the human body – it’s essentially impossible to pin down an exact value for either “calories in” or “calories out” under reasonable conditions.
Different macronutrients (carbs, fats, and proteins) take different amounts of energy to digest and process in your body. They can also influence various hormones like leptin and thyroid hormones that change your metabolic rate.
A caloric excess or deficit is met with regulatory responses from your body to naturally adjust how active you are or how many calories your metabolism will burn at rest. They’ll also affect hunger, which mediates how much food you’ll want to consume without forced self-restraint or gluttony.
Not everything you eat is even absorbed by your body to be utilized as fuel – you naturally excrete a small percentage of what you eat, which can change a bit with dietary composition. Furthermore, some foods will be used as fuel by your intestinal bacteria to a greater or lesser extent, meaning more or less of it is actually “left over” to be used by YOU.
Of course, then you toss in the monkey wrench that nutritional labels only have to be within 20% of the actual energetic values of the food – and that regulation isn’t always followed to a “t” by food manufacturers or restaurants. So even if you COULD know what your body was going to do with the food you ate, you still wouldn’t ever know for sure exactly how many calories you were putting in your body unless you made two identical meals, ate one, and tossed the other in a bomb calorimeter.
Also, even if you could know the exact number of calories that were going into your body or being expended by your body, other hormones like cortisol alter the relative amounts of each type of fuel your body is using –fatty acids, proteins, or carbohydrate. So being able to predict changes in weight with perfect accuracy still wouldn’t mean you could predict changes in body composition with dead-on precision.
Then, pressing further, you can’t know exactly how many calories your body is expending in day to day activities unless you live in a metabolic chamber in a lab. Different people display various degrees of efficiency in movements, so two people who are the same size who run the same mile will burn slightly different amounts of energy in doing so.
So am I proposing we throw the baby out with the bathwater and scrap CICO? Of course not! What would we replace it with, or how would you improve it?
Could the model account for more complexity? Sure.
However, let’s go back to the fact that for a model to be useful, it has to be user-friendly.
Attempting to account for ALL that complexity would make the model much less user-friendly. You could fine-tune the calories in and the calories out sides of the equation if you burned your feces in a calorimeter, accounted for fluctuations in lean and fat mass, measured the concentrations of various hormones a few times per day, took your temperature at regular intervals, and measured your daily activity by wearing an accelerometer all the time…but who’s going to do that.
There would be issues with gathering data (who wants to burn their poop and draw blood a few times per day?), and there would be issues analyzing data (the equation would be quite a bit more difficult to use than calories in – calories out).
And, as a segue into the next topic, although CICO is not a perfect model, it works well enough.
It actually works
This is what it all comes down to. Does the model work?
The first two factors – accounting for enough complexity and being user-friendly, are necessary factors, but they aren’t sufficient.
Any model, no matter how elegant or thorough it may appear, is ultimately of little value if it doesn’t actually describe the system well and lend itself to making predictions about that system that are fairly accurate (or more accurate than a competing model).
You can’t assume that a model is automatically a good model if it meets the first two criteria. Heck, you can’t even assume it’s an accurate model because it meets the first criterion, user-friendliness be damned (accounting for so much complexity that it’s no longer user-friendly).
Bringing this full circle, refer back to the initial part of this post about complexity and how little we know.
If we simply don’t know enough about a system, a model built on everything we know is still not going to be a good model. Even if we know enough to construct a good model, if we need a massive computer to account for enough factors to run the model, it’s still not going to be very useful to a coach or an athlete in-the-moment in the gym.
Imagine you have a machine that you feed a number into and, through a massively complicated algorithm you can’t understand fully, it spits another number out the other side, and the number it spits out isn’t always the same if you feed the same number into it repeatedly, though the output usually falls within a reasonably small range of values.
For example, if you input “5,” the machine may spit out 33, 37, 32, and 35, but not 2 or 13243.
You’re playing a game with a friend where you have to get the machine to spit out the biggest number possible.
Through trial and error, you find a range of inputs that tend to results it high outputs. You don’t know WHY it works, but you know that it works.
Your friend, on the other hand, knows more about math and computer science than you, and he does his best to figure out the algorithm. He constructs the best model he can to describe how the machine will respond, based on what he knows, though he can’t yet account for the full complexity of the machine’s operation.
When you play the game, you consistently get the machine to produce higher values than your friend does. His model looks better on paper (accounting for as much complexity as he possibly can vs. simple trial and error), but yours is better at reliably getting the machine to produce higher numbers.
This is the reality test. The ultimate usefulness of a model is not in its construction, but in its results.
If trial and error produces better results than a model accounting for everything we know, there may have been a problem with the actual construction of the model, or it may just be that we don’t know enough to construct an adequately good model.
You see this in exercise technique and program design a lot.
There’s nothing wrong with trying to build a model for ideal exercise technique, or proper program design. But does it produce results? Does it produce better results than competing models?
If it’s been tried and it’s failed, it’s not a good model, the elegance or complexity of it be damned.
If it’s been tried and it works better than the other models out there, it’s a better model even if it seems rudimentary or simplistic on paper.
If it simply hasn’t been tried, you have to treat it as an untested hypothesis – you can’t assert that it would work better than the other things out there, because what should work isn’t always what does work (refer to the multitude of “can’t-miss” drugs that fail badly when put through human trials).
A bit about humility
So, going back to CICO, even though there are a lot of factors it doesn’t account for, is it a good model? YES! Because it simply works. It produces results that are within 5-10% of what would be predicted by the model in the vast vast majority of cases. For a model as simple as CICO, trying to describe the behavior of an enormously complex system, perfection is an unrealistic standard – 5-10% is truly exceptional.
However, we can’t forget what we’re dealing with.
We are dealing with models.
Models are not the system. Models approximate the system and its behavior.
Models help us wrap our minds around and work with a set of factors, circumstances, and interactions that can’t be (at this time, potentially ever) fully known. Building and using effective models helps inform practice and helps us make useful predictions, but they are not Fact. They are not Truth.
They are maps, of varying degrees of quality. Your body and the world it interacts with are the territory. A perfect map of the USA doesn’t tell you what it’s like to be American.
As such, don’t fall into the lazy intellectual trap of treating your model as the facts about the situation. It simply helps you deal with facts that aren’t fully known.
Your body changes day to day, and it won’t respond exactly the same way to an identical stimulus if it meets it twice. Your body is different from someone else’s, and theirs won’t respond exactly the same way yours does.
It’s usually weak people who try to argue that one exercise technique or one program is the best. Chasing optimal is a fool’s errand.
There are very few raw lifters who I’d instruct to squat as wide as I do, but I have hips that let me drop into an almost-full split with no stretching, but that go bone-on-bone with very little straight-ahead flexion.
There are very few lifters who I’d recommend to squat heavy once every other week in pursuit of a 1000 pound squat like Eric Lilliebridge.
If you construct models and treat them as Truth, you’d think I squat wrong (still the highest raw drug-free squat at 242 all-time), and you’d think Eric Lilliebridge programs wrong (totaled 2000 when he was a teenager, and highest raw total of all-time at 275).
I’ve never had a strong person (someone who understands what it takes to actually get results) tell me I should squat differently, and I doubt Eric has ever had a strong person tell him he should program differently.
Your model (exercise technique, program, diet plan, etc.) may be a useful approximation of the facts for a lot of people, but it’s not the best for everyone, and it’s not the truth, the whole truth, and nothing but the truth.
This is not to say that everything comes down to trial and error. It’s not the nihilist “we can’t know, so why bother,” position. Gathering more facts and trying to build progressively better models as we learn more and more is a worthwhile pursuit. If we ever get to the point that we CAN model the human body from the bottom-up, it will doubtlessly save us a lot of time and resources in just about every branch of biological science. If we COULD build a model from the bottom up (taking into account individual differences) for proper exercise technique, it would save people a lot of trial and error and frustration.
But for the time being, we’re not there. Learn, experiment, build models, test hypotheses, and troubleshoot, but be humble about your conclusions.
More than anything, never lose sight of the single most important question: Does it work?
Please share. I’m not asking for selfish reasons. If people actually understood this, we could dispense with 90% of the pointless arguments in the fitness industry and elevate the conversation.
• • •