Strategy is the Deconfusion of Action

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.