The Use and Misuse of the OODA Loop
|Venkatesh Rao||Apr 13, 2018|
Boyd's famous OODA loop (observe-orient-decide-act) is probably the most misunderstood and misused diagram of practical significance in decision making. It is to decision-making what the five elements (earth/fire/water/wind/ether) model from alchemy is to modern chemistry. Or what Freud's id-ego-superego model is to neuroscience. It is in fact the same sort of model as Freud's: a prescient analytical model that attempts to describe cognition rather than a synthetic model that attempts to directly prescribe it.
As with Freud's theories, the problem with the OODA loop is that it was prematurely codified and popularized as an abstract conceptual superstructure before the empirical foundations had matured sufficiently. The map was drawn before the territory was properly explored. The empirical foundations of the two are actually the same: neuroscience, group dynamics (families being the key special case for Freud, military action for Boyd), organizational mapping, social network analysis, and so forth.
Though the terms and concepts of the OODA loop are intuitive and coherent at least at a first pass, there is fundamentally no way to use them entirely safely without making gross errors. Just as there is no way to do modern chemistry using alchemical concepts without making gross errors. Or trying to do neurosurgery based on an id-ego-superego map of the brain without making gross errors. The concepts of the OODA loop, like alchemical and Freudian concepts, are still useful so long as you recognize their limits, and approach them with a sufficiently poetic sensibility. And so long as you recognize the main potential modes of misuse, and firmly cordon them off.
In other words, you need to develop literacy in the OODA language. Starting with the way the diagram is usually drawn, which is frankly terrible.
Some species in the OODA loop zoo
Drawing the Diagram
Chances are, most of you have only seen the OODA loop as a diagram used to describe the concept on a single slide, not as a working construct in support of active decision-making. You may recall it looking like an ordinary feedback loop diagram such as you might encounter in many other kinds of material, but with lots of extra interconnections, annotations, and cryptic bits.
Here's the thing: the OODA loop is best thought of not a single prescriptive diagram, but as a class of description diagrams that can capture the state of a decision-making system/mind of a decision maker, using a visual alphabet of four primitive elements (observe, orient, decide, act).
In other words, it is not so much a diagram as a diagramming language. Or at least, that's what I think it ought to be.
The usual way of drawing the thing, with arrows all over the place, and annotations for every possible pathway of information flow, and every possible influencing factor from the context, is wrong in multiple ways.
It is wrong in the way a kitchen-sink cookie (with all possible ingredients) is aesthetically "wrong". It also tastes awful the way a true kitchen-sink cookie (the term comes from the idea of throwing in all available ingredients except the kitchen sink) would taste.
It is also wrong the way a composite fMRI image with every possible part of the brain lit-up is wrong. It makes for good chart porn, and perhaps has some minor utility in showing all the possible configurations in one visual, but isn't actually usable for anything serious.
And finally, it is also wrong in the way pointing at a diagram of a complicated, folded protein molecule and saying "that's chemistry" is wrong. The most complex example of a class of phenomena is not the same as the class it belongs to.
On the other hand, some people go the other extreme and reduce the OODA loop to some sort of minimalist, iconic mandala, or a 4-stage mnemonic circle. I don't know what you're supposed to do with those versions either.
I mean, do you really need a mnemonic device to remember that you observe, orient, decide and act to produce behavior? That's like needing a mnemonic to remember that you eat by Inserting Food, Jaw Up, Jaw Down, Swallow. It's too basic. If you forget that level of basic thing, you're probably dying. Or maybe you're supposed to stare at these minimalist versions until you reach an "Aha!" enlightenment about what to do, tapping into your inner Napoleon to find your coup d'oeil.
This kind of reductionism happens when you treat the diagram as a symbol of a philosophy (like the yin-yang symbol for the I Ching) rather than a working construct.
So how should you actually draw an instance of the drawing? What are the rules of grammar here?
The useful way to draw the OODA loop has only two rules:
Make orientation distinct: For reasons I'll get to, orientation is categorically different from the other 3, so make it visually distinct
Model a decision-making condition, not the idea of the loop: This means only drawing active flow arrows where information of some significance is flowing.
My usual way of drawing the diagram, when I'm actually using it to think, is to make a triangle out of observe-decide-act, with orient in the center. The sketch above shows six examples which I'll walk through in a bit. I've seen a few people draw it my way, but other equivalent ways are possible. For example, put "orient" on the outside of a linear observe-decide-act sequence with the arrow flowing back, and orient on top. This is Chet Richards' way, as well as very close to the way a control engineer would draw the diagram.
I like my way because it is easy, quick, and reflects what I consider the most important phenomenology driving it.
The most basic way to fail at using the OODA loop is to treat it as a read-only construct. As a diagramming language, it is the foundation of a kind of literacy within a particular idiom of decision-making. You have to decide when it is the appropriate language, and then use it in a literate way. It is not appropriate in all situations, and there are literate and illiterate ways of using it.
The OODA diagraming language is to Boyd's theories what Transactional Analysis is to Freud's. The core of TA is a interpersonal interaction diagramming technique using stacks of 3 circles labeled P, A, C (Parent, Adult, Child, Eric Berne's improved labels for superego, ego, id). You represent each person in an interaction using such a stack, and draw arrows showing how communication works/does not work in each "game" situation. As you might expect, the result (which you will find in Berne's famous book, Games People Play) is a combinatorial universe of game types. The idea of TA is to get literate in that game universe using the diagramming technique, and learning to recognize the most common ones.
In fact, you could argue that TA is actually OODA for interpersonal interactions, with the diagramming being focused on the actors rather than the action. It's a "be somebody" version rather than a "do something" version. To "recognize" a game condition and break out of it in TA is similar to recognizing an OODA condition and reorienting out of it. Every stable game is a harmonized set of player orientations.
I'll get to how to read the diagram correctly in a bit, but let's talk about three ways of misreading it.
The bureaucratic misreading: The bureaucratic way to read the diagram is to fussily fetishize all its arrows and annotations, in an effort to force-fit all situations to it. It is a textualist approach that is something like constitutional originalism (a political ideology, prominent in the US, that is based on seeking to derive all judicial and legislative thinking from a close reading of the constitution). As with all bureaucratic misreadings, this is a case of fetishizing the map to the point that you become blind to the territory.
I see this kind of misreading most often when consultants, middle managers (especially ones in staff roles), and other support types project their own lack of direct agency onto the situation. They use the diagram as a color-by-numbers template for "documenting" the situation as a consolation prize, rather than solving a decision problem, since they cannot influence it. Those who can, do. Those who can't, diagram the situation. You also see such overwrought usages among hedgehog-type people who lack other comparable mental models, and go "deep" into OODA, looking for more depths there than there are available to be plumbed.
The mystical misreading: This is a close cousin to the bureaucratic misreading, but instead of focusing on the diagram itself, it seeks to mine Boyd's briefings, lore from the Boydian community, and cryptic, aphoristic pronouncements passed down for generations (all 2 of them) to turn the OODA loop diagram into something like a divination tool for "What Would Boyd Do?"
So you take a cryptic Boydism like "get inside the opponent's OODA loop" or "fold the adversary's action back onto itself" and use the diagram as vagueware on the way to an intuitive leap to a creative idea about how to apply it. This is similar to the "What the Founding Fathers Really Intended" school of political philosophy, where you read the constitution in the light of Jefferson's letters.
I see this kind of misreading most often in military acolytes of Boyd, who have perhaps something of an undue reverence for him. Again hedgehogginess tends to compound the problem in the mystical misreading. If OODA is the only theory of decision-making you know, it will trip you up. Ignorance of other traditions and disciplines of decision-making that offer alternative mental models (such as neuroscience, control theory, AI, game theory, economics, among others) with different domains of appropriate use, can keep you from getting into OODA mysticism.
As an agility book-keeping system: This kind of misreading is most common in Silicon Valley, where you view a simplified version of the diagram, shorn of all its rich interconnections, parallelism (O, O, D, A all happening in parallel all the time, with varying intensities) and nonlinearities as a sort of circular stage-clock to replace the phase-gate roadmaps of waterfall planning. For example, you might treat the four stages as mutually exclusive stages in a circular flow, where your mental "cursor" is in one of the 4 boxes at any given time, and you might (for example) decide you're spending too much time in the "Observe" box and should budget more time for the "Orient" box.
What time is it? It is Orient o'clock. Let's meet for dinner at Decide o'clock. I can't, I need to be spending more time in Act o'clock.
This is not even wrong because the whole point of the OODA loop is to have a way to grok and work with the complexities possible in decision-making systems; the various mental-homeostasis "gears" you could be in, and the variety of regimes you could be operating in. If you just want to prune away all the potential complexity, why bother with a diagram?
I like to think of the OODA loop as a way to turn a regular feedback loop into a "hairy" feedback loop, where you tastefully toss a couple of monkey wrenches into a simple, somewhat degenerate (ie short of full complexity) default decision-making flow.
One significant example of going wrong in this book-keeping way is common in the lean startup world, where the OODA loop is occasionally treated as isomorphic to something like a Scrum product development cycle, with the temporal boundaries of a single circuit of the loop matching the sprint boundaries. In the worst case, there is a "pivot" (or reorientation) in every circuit, a recipe for incoherent thrashing. Or treating every cycle as a trial-and-error experiment which must necessarily maximize learning by driving straight into the most uncertain part of the environment.
If you're doing something like this, you probably have bigger problems than an illiterate OODA loop, like not being product-driven enough, or lacking a sense of true north in the market.
OODA illiteracy leads to many kinds of errors, but I want to say more about one in particular, "faster tempo." Sometimes all of Boydian decision theory is reduced to this.
This is the "transform lead into gold" driving myth of OODA. Or the "regression" in Freudian therapy.
Supposedly, if you simply operate at a faster tempo, miraculously all challenges will be overcome. Your product will develop on schedule, your customer will be delighted, and your enemies will die horrible deaths.
That logic is exactly as plausible as the logic that "if you could turn all your crap into gold, you'd be rich." Or "if you could remember all your childhood traumas, you'll be mentally healthy."
What makes the error a persistent one that gets perpetuated is that increasing your operating tempo is often the right thing to do, so naive Bayesianism would have you accelerating all the time. Versions include:
The player operating with the faster tempo wins
Iterate as fast as possible
You get inside the adversary's decision cycle by operating at a faster tempo
Here's a better rule-of-thumb on tempo.
You should aim to operate at a faster tempo in some situations all of the time, all situations some of the time, but not all the situations, all the time.
This error most often arises from the book-keeping misreading. It results from optimizing for the metric of "iteration rate" or more subtly, "learning rate," in a context-blind way. And as we know from Goodhart's Law, once a measure becomes a metric, it ceases to be a good measure.
Any dumb feedback-driven process can be made to go faster, faster, faster all the time. What takes intelligence is creating a feedback-driven process that can vary its tempo to suit the context.
Operating tempo is just one among many variables you can tune in creating an effective OODA configuration, and depending on the situation, the right tempo might actually be very slow. This is reflected in the interestingly paradoxical special-forces dictum: slow is smooth, smooth is fast. If you're operating at the right tempo, rather than the fastest possible, you will make faster progress overall. I like to think of it as finding semantic resonance with the environment. Product-market-fit generalized to behavior-environment fit.
The Greenland shark is one of the slowest fish around. It is also the longest-lived fish around. Its bodily op-tempo is a perfect match for its environment.
Fortunately such mistakes are rare in good teams in my experience (whether or not they are driven by misreading of an OODA loop, or some other source of conceptual errors). Common sense kicks in, and most people seem to intuitively operate by the understanding that there are times you go fast, and times you go slow. That you reorient when things start to go weird on you, and pivot on occasion when the environment changes enough to merit it.
You don't rush headlong into uncertainty with a battle cry of "Tempo, tempo tempo!" and a fixed two-week pivot schedule synchronized with feature releases. You're not a windup toy that runs headlong into a wall, then bounces off and rushes off towards the other wall, repeating till it runs out of spring power, celebrating each bounce as a "pivot."
The Territory of the Map
What are you diagramming when you draw an OODA loop?
Not the situation itself. That requires other kinds of maps suited to the context, such as literal geographic ones or financial accounts. What you're mapping is the state of the decision-making system. If it is a person, it is a state of mind. If it is an organization, it is a management culture and structure.
An OODA diagram is not a structural localization. If you go looking for parts of the brain doing Observation, Orientation, Decision, and Action, and trying to see if they're currently wired up the way you think, you're missing the point. The OODA diagram is a functional diagram. It captures the effective condition of the decision-making system, not necessarily its actual structure.
For example, if you're engaged in a real-time tactile-sensory task, like correcting how you aim a baseball, the "Orientation" role might be played by some Purkinje cells in the cerebellum, while the eyes, cortex, and arm muscle neurons do the observing, deciding, and acting. On the other hand, if you are doing some offline brainstorming about a creative 10-year plan to grow your company, the orientation might be happening in the default mode network.
You can observe with your fingers, and "act" with your eyes.
You can even observe with your mind, by reprocessing sensory memories, and orient with your fingers, by drawing diagrams.
You can similarly find varied situation-specific mappings when the decision-making entity is a group of people sitting around a table chatting (with or without alcohol), or an entire organization "maneuvering" in a market.
Sometimes marketing is the eyes and the CEO is the decider, sometimes it is the other way around.
But across the complexity of ways an OODA loop can be embodied by a specific system, you do notice one thing: the Orient part is sort off of to one side, generating interruptions, overrides, and exceptional course corrections, while the other three flow into each other serially, generating a smoother base behavior. Orient is conceptually at a higher level of abstraction than the other three, which is why it is something of a category error to visualize it in the same way. Orient is also often a locus of recursion, maintaining internal models of the whole system, including itself. Maps within maps.
There are rare conditions under which the orientation level collapses into the observe-decide-act level, creating a sort of pristine flow condition, but it is always local in time and space. You can't stay in that state indefinitely, or for an unbounded scope of action.
At every level, you can go to any granularity you like. "Decision maker" is a recursive entity concept. At one level you might be an unconflicted decision maker described by a single well-tuned OODA loop generating coherent, harmonious, unfolding action. At a lower level, you might be a conflicted battlefield where multiple internal demons are trying to hack other's OODA loops to try and steer you.
In fact, that's a good way to think of the territory the OODA loop is meant to diagram: it's a recursive stack of demons, all the way from small neural circuits selfishly trying to get their own way in the economy of the brain, to the largest human organizations. Or zooming out in time, selfish genes adversarially fighting selfish memes.
It is dueling demons all the way down, and demons all the way up, in space and time. This is why modeling the presence of an adversary is so important in OODA loop thinking. Even if there is no explicit adversary or declared conflict, there's always an adversary somewhere. But before we get to that, let's say a few things about actually reading the diagram.
Reading the Diagram
Literacy is writing and reading. To read an OODA diagram, you have to ask yourself, what decision-making condition is it portraying? I've illustrated 6 very common ones above, which I'll describe here. Reading from left to right from the top left, we have:
Normal: The orientation block is monitoring a functioning observe-decide-act loop.
Simulation: Action is temporarily unplugged and possibilities are being simulated (ie a domain model within the orientation is being used to speculatively estimate the possible outcomes of actual action)
Tuned closed loop: Orient is disconnected. This is not necessarily a pathology. It might be a groove or a rut. Taking orientation offline increases risk, but lowers cost. You might expect to use routine kinds of feedback, but not exceptional kinds of overrides or interrupts.
Tuned open loop: Both orient and observe are disconnected. Again, not necessarily a pathology. A traditional washing machine is a tuned open loop. If you don't expect things to go wrong routinely OR exceptionally, you can operate very cheaply with just a decide-act feedforward circuit. Feedback is never free, so you should eliminate it when you can.
Situation tracking: Observe and orient are in a loop, and act and decide are isolated. This does not mean you're not acting or deciding. You might simply be stirring the soup, making no decisions, and thinking about some other unrelated situation that you're not actively working. Or of course you could be in analysis paralysis.
Adversary inside your OODA loop: Here is one way an adversary could be inside your OODA loop: a man-in-the-middle type conceptual attack where your actions are being driven by a distorted picture of reality (like gaslighting). If this situation progresses, you will slowly exhaust your resources, and collapse into a cheaper open-loop condition as shown at the lower right. This is (one pathway to) "action folding in upon itself"and being pwned by an adversary driving you to exhaust yourself.
The point of my list of sketches and examples (apologies for the poor readability of the drawing, I had to hand draw since I don't have my iPad/Pencil handy) is not to provide you with a definitive encyclopedia of OODA diagrams, but to get you thinking of OODA as a diagramming technique rather than a single visualization of a theory, or a system state description.
Half the time, just getting the qualitative diagram right will solve your problem, and you won't need to go to more complex ways of modeling decision systems involving math and software.
Let's say a little bit more about adversaries, since the strength of OODA-thinking is its incorporation of adversaries.
There is Always an Adversary
Though it is not immediately apparent from the canonical diagram, the greatest strength of the OODA language is that it assumes an adversary. This does not mean it is a "wartime" construct useful only when there is an explicit adversary. It is useful for modeling any kind of non-aligned intentions in a situation.
Starting with the inside of your own head.
You are rarely a single coherent decision-maker. Usually, multiple parts of you contending for control over external behavior. And they don't always play fair with each other.
A naive way to think of the OODA model of your mind is to treat it as a well-governed hierarchy. GluttonMe wants to eat a bag of chips, while StoicMe wants to go the gym. Does each part of me tamely offer up a candidate "orientation" to a higher level OODA loop for adjudication and rational choice? Is there a CEOme that chooses whether GluttonMe or StoicMe will be in the driver's seat as prevailing orientation for Mission NextFiveMinutes, with the rest falling in line to observe/decide/act appropriately?
Hell no, it's a battlefield in there. Different bits of you are constantly trying to fool each other to get their way. There's always many MiniMes contending for control, but no Dr. Evil.
The OODA diagramming language is good for this because it gives you a tractable set of primitives with which to capture what's happening. In my example above, maybe StoicMe can simply hide the bag of chips, thereby pwning the OODA loop of GluttonMe, and exhausting its drive to eat chips by removing the sensory reinforcement of desire. Once I weaken that drive enough, hey, I can go to the gym.
Speaking of chips, just writing that paragraph made me remember I have a bag in my laptop case. Now I want to take a break and eat it. GluttonMe just got inside the OODA loop of PamphleteerMe. See how this works?
The point is that each OODA diagram is like a molecule, and the Observe, Orient, Decide, and Act bits are like atoms. And many molecules can react in interesting ways, but only if you allow for interesting breakdowns, recombinations, and reactions.
This stuff gets harder as you go up the recursive demons stack. Here is a bunch of diagramming problems for you to think about:
Diagram an attentive waiter serving a family of diners at a restaurant
Diagram the decision-making style of a co-dependent CEO and COO at a toxic company. Who is inside whose OODA loop?
Diagram the situation in the movie Gaslight
Diagram the emerging consensus of a workgroup of 4-5 people at a software development sprint meeting
Diagram the deliberations of a parliament in a 2-party system
Diagram a first-past-the-post election
Diagram a two-way symmetric declared war
Diagram an asymmetric three-way undeclared war
Diagram a long-term demographic shift over 2 generations
For each, the way to use the OODA diagramming approach is to make aesthetic judgment calls about the right level of representation for decision-makers, and how their OODA loops are intertwingled and getting inside each other.
I'll be honest: It takes effort, it takes practice, and is rarely as useful as you think it is going to be.
It's a lot of fun to do though, even if all it does is get you primed or warmed up for using some other decision-making approach.
When to Use OODA DIagramming
In your decision-making toolkit, OODA is the mirror. You use it to see yourself.
I occasionally use OODA style diagramming (very free-form) to think about my own challenges, but only rarely with others. A good decision-maker can refine their thinking with OODA, but OODA won't turn a bad decision-maker into a good one. Just like a mirror will help a beautiful person with dressing and make-up, but not turn an ugly person into a beautiful one.
If two people are roughly equally literate in OODA, it can serve as a compact language for talking about things, but if they are at different levels, or both at a low level, it is best to dispense with OODA language and simply talk in English.
OODA diagramming is most useful when the characters are relatively clear, and the plot is complex, but character-driven, and the situation is more ambiguous than uncertain. Surprises should be coming more often from misread conditions than from entirely unexpected new events.
In other words, OODA diagramming is good for figuring out how you're stuck, and how to try and shift your behavior patterns so you get unstuck, especially when the adversarial element in the situation is strong.
But it is not the only way, not always the best or most reliable way, and most importantly, not a formulaic way. It requires taste, literacy, an artistic familiarity with the idiom it expresses, and practice to apply well.
If you use the OODA loop like an alchemist looking to turn lead into gold, you're going to be disappointed.
Feel free to forward this newsletter on email and share it via the social media buttons below. You can check out the archives here. First-timers can subscribe to the newsletter here. More about me at venkateshrao.com
Copyright © 2018 Ribbonfarm Consulting, LLC, All rights reserved.