Reading the New Research Directions update from MIRI, I was struck by the description of deconfusion:
By deconfusion, I mean something like “making it so that you can think about a given topic without continuously accidentally spouting nonsense.”
I find this concept deeply impressive. I have also lately been considering the problem of strategy or lack thereof, so the idea popped up almost immediately: strategy is deconfusion in the action domain.
I'm going to draw on three sources for this post: the first is the aforementioned New Research Directions post from MIRI; the second is a paper from Parameters 46 Winter issue by Jeffrey W. Meiser, "Are Our Strategic Models Flawed? Ends+Ways+Means = (Bad) Strategy"; the third is an article from November 2012 in The Atlantic by Thomas E. Ricks, "General Failure." I recommend them all individually, but I won't assume you have read them.
Returning to the concept of deconfusion, it is easy to change the quoted section only a little to capture what I mean:
By strategy, I mean something like "making it so that you can act towards a given end-state without continuously accidentally wasting effort."
I think this is an important connection to draw. There is a lot of information available on strategy: each military philosopher of note supports an entire corpus of commentary, and likewise for every conqueror; every war has lots of official and academic analysis done on its results; there is an absurd profusion of filtering the military and historical information through the lens of self-help or business-speak. It has all very consistently failed to guide the development and execution of good strategies, and I think deconfusion does a good job of pointing to why.
Failure to Notice Confusion and the Lykke Model
Meiser's paper is about the current norms in the US military and the Army in particular. The focus of the paper is the Lykke model and the ways in which it encourages bad strategy. It consists of the following:
“Strategy equals ends (objectives toward which one strives) plus ways (courses of action) plus means (instruments by which some end can be achieved).”
The problem in practice, Meiser argues, is that strategy development is dominated by means-based planning. This is because the theory is from 1989, developed in reaction to the failures of the Vietnam War; the thinking went that if resource constraints were better taken into account, we could prevent "unrealistic strategies." Ends are treated as given and what people mostly do with planning is look at the means table, match them against the ends table, and then call it a day. The problem with the model is that it promotes a kind of plug-and-chug approach.
Meiser uses General Stanley McChrystal's plan for Afghanistan after he took command of that theater as an example:
What emerges from journalistic accounts of the 2009 Obama administration strategy-making process is the observation that the entire discussion by civilian officials and military officers was about the number of troops, not strategy.
. . .
After repeated presidential requests for at least three distinct options, all Obama ever got was slight variations of the original ones. All options were based on the amount of resources being thrown at the problem.
The debate was actually moot. No one knew how to really use those resources, and the military did not notice their confusion about the problem. Of course, this is not the only shortfall.
Assuming Confusion Away
Returning to the MIRI post, this is given as an example of confused thinking about AI risk:
People who are serious thinkers about the topic today, including my colleagues Eliezer and Anna, said things that today sound confused. (When I say “things that sound confused,” I have in mind things like “isn’t intelligence an incoherent concept,” “but the economy’s already superintelligent,” “if a superhuman AI is smart enough that it could kill us, it’ll also be smart enough to see that that isn’t what the good thing to do is, so we’ll be fine,” “we’re Turing-complete, so it’s impossible to have something dangerously smarter than us, because Turing-complete computations can emulate anything,” and “anyhow, we could just unplug it.”)
From the Atlantic article, quoting a slide from a classified briefing:
“What to Expect After Regime Change”:
Most tribesmen, including Sunni loyalists, will realize that their lives will be better once Saddam is gone for good. Reporting indicates a growing sense of fatalism, and accepting their fate, among Sunnis. There may be a small group of die hard supporters that [are] willing to rally in the regime’s heartland near Tikrit—but they won’t last long without support.
Comparing these two is only marginally appropriate; the first quote is from (at the time) amateurs who were deeply engaged with the problem they were thinking about, whereas the latter is from nominal experts who were negligently hand-waving the problem away. I say that both suggest confusion about how to consider the problems at hand. I go further and say that the latter is a more pernicious sort of confusion, because they assumed there never was any from the beginning.
Also from Ricks, regarding General Tommy Franks at the beginning of the Iraq War:
In many ways, Franks is the representative general of the post-9/11 era. He concerned himself principally with tactical matters, refusing to think seriously about what would happen after his forces attacked. “I knew the President and Don Rumsfeld would back me up,” he wrote in his memoir, “so I felt free to pass the message along to the bureaucracy beneath them: You pay attention to the day after and I’ll pay attention to the day of.” Franks fundamentally misunderstood generalship, which at its topmost levels must link military action to political results.
This did not improve as the war went on:
Not long after the Anaconda battle, Franks spoke at the Naval War College, in Newport, Rhode Island. A student heard his talk and then posed the most basic but most important sort of question: What is the nature of the war you are fighting in Afghanistan? “That’s a great question for historians,” Franks said. He then went on to discuss how U.S. troops cleared cave complexes.
Nor was it a problem unique to Franks. Regarding his successor, Lieutenant General Ricardo Sanchez:
Sanchez inherited no real war strategy from Franks or the Bush administration, and he did nothing to remedy that deficit. This lack of any coherent strategy manifested itself in the radically different approaches taken by different Army divisions in the war. Observers moving from one part of Iraq to another were often struck by the extent to which each division was fighting its own war, with its own assessment of the threat, its own solutions, and its own rules of engagement.
Eventually on the third try someone noticed that they were confused about what the Army was doing, in the person of General George Casey:
He knew the Army needed to start operating differently in Iraq. He developed a formal campaign plan, something Sanchez had never done. More significant, he asked two counterinsurgency experts, Colonel Bill Hix and retired Lieutenant Colonel Kalev Sepp, to review the actions of individual units and make suggestions. Sepp, a Special Forces veteran of El Salvador with a doctorate from Harvard, reviewed the commander of every battalion, regiment, and brigade in Iraq and concluded that 20 percent of them understood how to properly conduct counterinsurgency operations, 60 percent were struggling to do so, and 20 percent were not interested in changing and were fighting conventionally, “oblivious to the inefficacy and counterproductivity of their operations.” In other words, a vast majority of U.S. units were not operating effectively.
But he, too, took the ends as both given and uncomplicated. While Baghdad was in the throes of civil war:
Casey’s lack of awareness began to undercut his support at the top of the Bush administration. On August 17, 2006, during a video briefing to top national-security officials, he said he wanted to stick with his plan to turn Baghdad over to Iraqi security forces by the end of the year. Vice President Dick Cheney, watching from Wyoming, was troubled by that comment. “I respected General Casey, but I couldn’t see a basis for his optimism,” he wrote later.
From the beginning of the conflicts in both Afghanistan and Iraq, the possibility of being confused about the goals was not seriously considered, because the military assumed the problem away.
Theory of Success and Re-enter Deconfusion
Current strategy norms don't admit the idea of confusion, which is a problem. In Meiser's paper he offers a different definition of strategy which he hopes will promote "creative and critical thinking," which I have taken the liberty of interpreting as addressing the confusion problem. Pleasingly he moves into terms and concepts we are familiar with (emphasis mine):
The two definitions that come closest to articulating a distinctive meaning for strategy are offered by Barry Posen and Eliot Cohen. Posen defines grand strategy as “a state’s theory about how it can best ‘cause’ security for itself.” Cohen defines strategy as a “theory of victory.” The key insight by Posen and Cohen is the inclusion of the term theory. If we define theories as “statements predicting which actions will lead to what results—and why,” we can move toward a better definition of strategy that is general, but not too inclusive, and captures the essence of the concept.
If we use the Posen-Cohen approach with a more general definition of purpose, we arrive at a sufficient working definition: strategy is a theory of success. This creates the expectation that anything called a strategy will be a causal explanation of how a given action or set of actions will cause success. Most strategies will include multiple intervening variables and conditions. Defining strategy as a theory of success encourages creative thinking while keeping the strategist rooted in the process of causal analysis; it brings assumptions to light and forces strategists to clarify exactly how they plan to cause the desired end state to occur.
This puts strategy on the same type of conceptual ground that motivated deconfusion in the first place; it is the primary reason I see deconfusion being so valuable. A secondary reason I see deconfusion as being valuable is shifting from the current paradigm. Currently formal strategic methods don't account for confusion, and lazy or negligent approaches to those methods can make it impossible to resolve. Viewing deconfusion as fundamental means that my assumption is that I am confused.
The military example of being confused about what to be confused about makes me think there is a confusion equivalent of the knowns. From the famous quip:
It feels like the transition from unknown to known probably looks like this:
Confused confusions (an unknown)
Deconfused confusions (these seem like what we have been calling 'disentangled' in Agent Foundations)
Deconfused deconfusions (a known)
There is an idea I have used by implication in the OP, but might benefit from being identified specifically. This idea is that the level of abstraction where a concept is applied matters.
To illustrate what I mean, consider the end of the confused statements quote from the MIRI post:
Today, these conversations are different. In between, folks worked to make themselves and others less fundamentally confused about these topics—so that today, a 14-year-old who wants to skip to the end of all that incoherence can just pick up a copy of Nick Bostrom’s Superintelligence.
I think it would be reasonable for someone reading my post to look at that section of the MIRI post and then ask: so what is the Superintelligence of strategy? My answer is that there isn't one yet; this is what Sun Tzu and Clausewitz tried and failed to accomplish. I don't believe we have a good enough understanding of the component disciplines of strategy to write one, either (consider our mastery of computer science and information theory relative to our mastery of political science, economics and psychology). We are too confused.
I think the key insight of Meiser's approach is that he applies scientific reasoning as a generative rule for a strategy instance, rather than trying to describe a science of strategy in general and leaving the instance as an exercise for the reader. In other words, he took the scientific perspective and aimed it one layer of abstraction down. This allows us to account for confusion.
The level of abstraction is a big reason deconfusion is so awesome: it works no matter where you aim it, even aiming-at-aiming.
anything called a strategy will be a causal explanation of how a given action or set of actions will cause success.
This is very helpful. It should be in bold.