I’ve been reading Boyd’s work again, this time focusing on the primary sources. I’ve not really begun to internalize it all yet, so expect more posts on this in the near future. This is just a set of initial, tentative thoughts about the OODA loop.
The first thing to note is that the OODA loop as finally envisaged by Boyd is significantly more nuanced than the general description. The basic OODA loop theory is that you want to cycle through the stages of the loop faster than the adversary. If you manage to orient and act faster than they can orient, then they are effectively orienting to situations that have already changed, so their actions will be wrong, and then this failure will compound and your advantage will increase.
This approach emphasises relative speed as a key criterion in victory. The aim is to move faster, more unexpectably, and more nimbly than the enemy can expect,and in this way crush their ability to respond.
This all seems reasonable and likely works well, depending on the domain. In many domains speed is a critical advantage. An important nuance is that it is only relative speed that matters. You only have to orient and act faster than the other guy can orient to gain a compounding advantage. Due to this, one of the key capabilities of any successful system is that of “fast transients”. I.e. the ability to command fast “accelerations” where acceleration is meant in some abstract sense.
This is effectively what is meant in organizational terms as being “nimble”. That is, the ability to make rapid decisions which can be executed quickly to achieve rapid changes in behaviour which are unpredictable is key to success in this model. Boyd has a good discussion with this in jet fighters. The key advantage of the jet fighter in a dogfight is not necessarily the top speed (at least this was presumably true in the 1970s), but rather the ability to rapidly accelerate and decellerate with a short turning circle compared to the other guy.
Is this true in the general sense? I would argue no. It only applies in certain strategic situations, which are primarily focused around strategic contests with small numbers of participants, where actions are faster or of the same order of magnitude duration as orientation, where a good model of the strategic game is largely known to all participants, where the downstream effects of actions are predictable to at least the first order, and where actions are rapidly reversible and don’t have high costs, but also have immediate strong effects.
Let’s look at these conditions in turn. It is worth noting that these conditions apply in many military situations, potentially up to the level of theatre strategy (i.e. in Blitzkriegs) and extremely well in Boyd’s initial domain – air-to-air dogfights. They also apply to some business situations, especially at lower levels, most sports and games, and diplomacy at some level. First though I want to go through why I think these conditions limit the applicability of the theory to certain situations.
1.) Actions are faster or of the same order of magnitude as orientation:
- this one is simple. The whole idea of the OODA loop is to orient and act faster than the other guy can orient. This is impossible if actions take a significantly longer time to achieve than an orientation step. For instance, consider competition between a bunch of heavy-industry competitors. Suppose one competitor decides to build a new factory. This can take years to get up and running and is essentially impossible to disguise from competitors. The competitors are also able to run analyses on the new factory and orient their own strategy to it (over a time-course of weeks or months) compared to the years for the factory to come online. In this scenario rapid transients are impossible and so OODA loop theory is not especially useful here. (Interestingly, this would change if, for instance, the new factory had some rapid prototyping tools to iterate on new products over a timescale of weeks, for instance. This means that rapid transients are again possible and OODA loop theory can apply. Also important to note is the relative speed. Weeks for a new product is not fast by some objective measure, but is fast compared to how rapidly competitors can orient, and OODA loop theory applies).
2.) Strategic game has a small number of participants.
- This is necessary because the OODA loop theory requires a single or a couple of distinguished adversaries. If there are a large number of players, for instance in financial markets, then each player is not really orienting on what you specifically are doing, but to what some aggregate of activity over all players are doing. This does not mean that rapid orientation and action are not useful in this scenario, they very definitely are. Being able to act on new information faster than other market participants is a huge advantage, but it removes the compounding effect of the OODA loop. Your adversaries orientation ability is not progressively destroyed by your faster cycling within the loop. The benefits are constant rather than cascading.
3.) A good model of strategic game is known to all participants; downstream effects of actions are predictable.
the OODA loop is only effective if a solid dynamics model of the world is known beforehand, and so are the effects of actions. It is impossible to really formulate any kind of strategy without this kind of knowledge and understanding of your domain. Doing stupid things faster is not necessarily any better than doing them slowly. A fighter “pilot” who doesn’t know how to fly and has no idea what the buttons in the cockpit do is not going to outperform one that does, even if the one with no idea can react 10x faster to the situation. The result will just be a lot of random flailing without much effect rather than building a consistent advantage. I suspect a lot of the “action” that occurs in the worlds of business and politics is of this latter type, due to a lack of good understanding of how the domain actually works.
In the case of lacking a good world model understanding of a domain, the most important thing to do is try to obtain such a model which is more important than any specific strategy. Generally strategy and effective action can only procede when such a model of the domain is available. Anything before that happens is not really a strategy in a full sense.
4.) Actions must have large effects. * - Rapid transients are required both to be rapid and to have large effect sizes. There is not much point in rapid action if your actions don’t do much. This just results in strategic flailing. The idea of the OODA loop is NOT to flail around doing lots of things randomly. It is to have a well considered strategy, and execute it and adapt on it rapidly. The military equivalent of this would be marching troops quickly back and forth around randomly without any specific objective. This might well confuse the adversary, but not really do much damage or be a good strategy.
Supposing all the conditions of the OODA loop are met. What then? The idea must be to optimize various aspects of the loop for speed and effective action. In general there are three key places in the loop to optimize:
- The rapidity with which informative observations are sampled and conveyed to the orienting phase.
- The rapidity with which orientation can take place and decisions made
- The rapidity with which actions can be executed.
In the first case, getting observations and data to where they are needed fastest is crucial. While only a minor part of contests with direct sensory contact with the environment (i.e. in a jet fighter), this is a very important factor in any kind of decision-making within large organizations. One of the major advantages of small organizations is that observations can be conveyed to the top very quickly, while in larger organizations they are typically filtered through multiple layers. The CEO is often the last person to know about important changes in environment. The ways to ameliorate this are pretty well known. There must be rapid channels of information transfer right to the top. Must be a combination of formal and informal channels. Upper management cannot become too disconnected from day to day concerns or else they will also lose the metic knowledge of changing conditions on lower levels.
Rapid channels are easy to implement. What is hard to implement is then filtering the data that comes from the rapid channels. Because most of it will likely be useless and noise. Large organizations effectively implement filtering ad-hoc through hierarchy, but this is likely not the optimal way to do it and leads to certain biases. Better filtering could involve deciding ahead of time and through experience what information is directly useful and what is not. For instance, supposing the organization is implementing feedback policies, information directly related to those policies should be conveyed as rapidly as possible. The issue will be preserving the kind of inofficial “gut feelings” or metic knowledge of workers on the edges upwards, which are often woolly and hard to quantify and don’t sound official but are often the most valuable indications of how things are going. (Feedback policies are an interesting question. Very few strategies seem to have any sort of integral or derivative component in them and are entirely based on proportional control. PI or PID strategies is likely an interesting area to consider).
The other alternative is to develop automatic tools or thinking tools to allow people at the top to digest larger amounts of information and sift it themselves. This is both a technological, educational, and cognitive challenge but can likely be done with reasonable visualization and modelling tools.
A third alternative in organizations is obviously delegation, which is what Boyd strongly argues for through aufstragstaktik -i.e. decentralizing decision making down to the lowest levels in closest contact with the situation. This delegation speeds up observation speed, may or may not speed up orientation (people at lower levels likely have more gut instinct or fingerspitzengefuhl while those at the top typically have a greater holistic picture and potentially more experience in general although not in the specifics) and also speeds up actions by eliminating communication lag. The issue then is with alignment of multiple independent actors which is another organizational problem to solve.
In individuals, increasing orientation speed primarily comes from learning a direct and intuitive model of the world. The fastest orientations and actions are intuitive. This likely comes from training and long experience. Technology can also help with this in creating visualizations, interactive models, or other seeing and thinking tools which enable rapid iteration and planning of strategies. Intuition and initiative, underlain by a long period of training and experience is what drives success here. At an organizational level, these factors are critical for the key decision makers, but then there is also the communication time to consider too. The decisions and orientations formed at the top must be transmitted throughout the organization for it to have oriented as a whole. This requires effective and rapid top-down as well as bottom-up channels of communication. These channels must not be overwhelmingly noisy.
The third step is increasing action speed. In some domains action speeds are inherently fixed by the dynamics of the world. In others they are amenable to technogical solutions. Boyd has a great example where the soviet MIG fighter is more manouverable in simulations than the American fighter, but not in practice simply because the American fighter had power-steering while the Soviet fighter did not. This means that the Soviet pilots were unable to reach the theoretical maximum manouverability of their aircraft while the American pilots could get significantly closer to that limit. In terms of organizational dynamics, rapid action is best ensured by trained and experienced personnel and effective communications.
These are some initial thoughts on Boyd’s work anyhow. May come back to it later.