Aria’s Past, Present, and Future

On divining the esoteric truth of Neo-Venezia through holes in world-building.
anime, criticism, computer-science, SF, cats
2011-07-132013-08-03 in progress certainty: highly unlikely importance: 3

Here is a con­spir­acy the­ory for you to munch on: just as is ac­tu­ally about how Kyon is god and not Haruhi (see ), so the true story be­hind has to do with the cats.

Even the su­per­fi­cial1 overview of the Wikipedia ar­ti­cle men­tioned that they are nearly as in­tel­li­gent as hu­mans (ex­plain­ing some of their grossly out­sized head­s). They seem to do noth­ing. But we know they live for cen­turies as ev­i­denced by how the cat-pres­i­dent of Aria Com­pany does­n’t vis­i­bly age even as the founder of Aria goes from a young girl to a wiz­ened granny, while hu­mans in Aria seem to live not much longer than we do (the old peo­ple on Neo Venezia don’t seem to date back to the ter­raform­ing of Mars/Aqua, which was only a cen­tury or 3 ago). Fur­ther, there is the mys­te­ri­ous char­ac­ter of Caith Sith (!) who is enor­mous, eas­ily as large as 3 or 4 hu­mans, and who com­mands all the cats; he too dates back cen­turies—he an­nu­ally vis­its Neo Venezia cos­tumed as the Venet­ian ad­ven­turer , and those vis­its have been go­ing on as long as any­one knows. Why does each gon­do­lier agency have a cat as pres­i­dent? Per­haps this re­flects the hid­den truth of their con­trol; per­haps it gives them li­cense to be con­stantly wan­der­ing the city spy­ing on peo­ple and vis­i­tors. Per­haps they are the small­est of the cats, suit­able un­ob­tru­sive and in­vis­i­ble—­Cait Sith’s in­tel­li­gence agents!

Or con­sider the Aria the Nat­ural episodes where Cait Sith is re­vealed to have power over evil ghosts, and to be the con­duc­tor of a Galaxy Rail­way2. Con­fus­ing­ly, the evil ghost he dis­pels from Akari is pre­sented as an an­cient leg­end from Earth about an ex­e­cuted crim­i­nal (need­less to say, no one has ever been ex­e­cuted on Aqua, al­though we hear of deaths in ac­ci­dents)—although at the end of the episode we learn the leg­end and ghost post-date Aqua and Neo-Venezi­a’s found­ing! How can all this be rec­on­ciled?

Let’s pull back and look at Aqua. Every sin­gle craft and in­dus­try on Aqua seems aimed at ei­ther ba­sic needs (the many wind­mills sup­ply ba­sic elec­tric­ity need­s), or aimed at tourists (the en­tire un­dine sys­tem, all the lit­tle coffee shops, the is­land of glass­blow­ers, etc.). The sys­tem is oth­er­wise ab­sur­d—why are highly ed­u­cated ex­pen­sive hu­mans spend­ing their lives row­ing peo­ple around on er­rands over dis­tances of a few hun­dred me­ters? Even ve­hi­cles slowed down to the point that their wakes did­n’t dam­age Neo-Venezia would be eco­nom­i­cally su­pe­ri­or, to say noth­ing of all the la­bo­ri­ous hand­i­crafts and ob­so­lete pro­fes­sions prac­ticed. Se­ri­ous­ly, they have a bustling postal office? We are shown/­told this by a nar­ra­tor who is con­stantly send­ing in­ter­plan­e­tary emails (in the ani­me). Also keep in mind, Aria is set in the dis­tant fu­ture, many cen­turies or maybe even mil­len­nia from now, after un­known cranks of 3 or other as-yet un­known laws, at a time where Earth­-Mars travel is rou­tine and hu­man­ity has the power to ter­raform en­tire worlds and gen­er­ate grav­i­ty/an­ti-grav­ity and who knows what other mir­a­cles? And yet, here’s an en­tire city of 100% nor­mal hu­mans with ex­tremely old-fash­ioned so­cial arrange­ments (here’s how old fash­ioned they are: I can’t think of a sin­gle ho­mo­sex­ual pair­ing) go­ing about their busi­ness, with the only inkling of the far fu­ture be­ing the oc­ca­sional preter­nat­u­rally in­tel­li­gent cat, su­per­nat­ural oc­cur­rence, and float­ing ship/is­land? Bo­gus; fail.

Or… is it? Every­one in Neo-Venezia is dis­gust­ingly hap­py, and even when they’re sad they do it in a fun quaint way. They all seem far more sat­is­fied in their ca­reers than we are, and cer­tainly their food and art seem bet­ter. Their ar­chi­tec­ture may be an­cient, but gosh darn it all, does­n’t Aria Com­pa­ny’s house look clean and liv­able?

Sim­plic­ity of life, even the barest, is not a mis­ery, but the very foun­da­tion of re­fine­ment: a and white­washed walls, and the green trees, and flow­ery meads, and liv­ing wa­ters out­side; or a grimy palace amid the smoke with a reg­i­ment of house­maids al­ways work­ing to smear the dirt to­gether so that it may be un­no­ticed; which, think you, is the most re­fined, the most fit for a gen­tle­man of those 2 dwellings?4

That is more than I can say of Amer­i­can hous­ing.

What Neo-Venezia is tru­ly, I think, is a pic­ture of la dolce vita—the sweet life, the pleas­ant one. The in­hab­i­tants of Neo-Venezia don’t strive for ma­jor ac­com­plish­ments: they com­pete to be­come Pri­mas or for pro­mo­tions as sala­man­der­s/g­nomes only so far as the com­pe­ti­tion lends their lives a lit­tle spice.

The in­hab­i­tants are, in fact, closely akin to ’s ‘eloi’. They barely work (one can­not imag­ine the word used in Neo-Venezi­a), are never shown in the count­less tech­ni­cal roles that an in­ter­plan­e­tary civ­i­liza­tion ought to re­quire, and are as­tound­ingly ig­no­rant. One ro­man­tic episode cov­ers how one char­ac­ter is so ig­no­rant of ba­sic physics that grav­ity has to be ex­plained to her. Even in our age it’s hard to be ig­no­rant of grav­ity un­less one is as­tound­ingly il­l-e­d­u­cated or stu­pid, and the char­ac­ter would seem to be nei­ther. Her ig­no­rance is likely quite gen­er­al. How do the grav­i­ty-gen­er­at­ing ‘stones’ work? Even the char­ac­ter whose ca­reer is main­tain­ing them does­n’t seem to re­ally know. They’re just there. Any an­ti-grav­ity sys­tem sta­bi­liz­ing moons would re­quire in­tim­i­dat­ing math­e­mat­ics and pow­er­ful com­put­ers and net­works of sen­sors, yet that char­ac­ter stud­ies dusty old books. Ob­vi­ously he’s noth­ing but a tech­ni­cian punch­ing but­tons, and maybe not even that.

Of course, there are no in Aria to put our beloved char­ac­ters in dan­ger. They just float along hap­pily through life. No Mor­locks that we ever no­tice, any­way.

In­ter­est­ing enough a view, but what does it have to do with Cait Sith? Well, if Neo-Venezia is a post-post-post…-mod­ern par­adise, then is­n’t there some­thing miss­ing—a sense of mys­tery, of grandeur? Mys­tery and a lack of con­trol is one of the fun­da­men­tal things sep­a­rat­ing an au­then­tic ex­pe­ri­ence apart from a con­trolled cor­po­rate one; Neal Stephen­son re­flects in his es­say “In the Be­gin­ning was the Com­mand Line” that every spot in Dis­ney is cal­cu­lat­ed, and cal­cu­lated to such an ex­tent that if you see an­cient In­dian ru­ins in An­i­mal King­dom, then by gol­ly, those ru­ins look “more like what I have just de­scribed than any ac­tual build­ing you might find in In­dia.” And Dis­ney’s “seam­less il­lu­sion” re­quires an ab­sence of au­thors and ties to a spe­cific his­tor­i­cal ori­gin or process: “…the au­thors’ names are rarely if ever men­tioned, and you can’t buy the orig­i­nal books at the Dis­ney store. If you could, they would all seem old and queer, like very bad knock­offs of the pur­er, more au­then­tic Dis­ney ver­sions.” If that mys­te­ri­ous gar­den in a for­mer monastery known only to a se­lect few were ac­tu­ally Mys­te­ri­ous Lo­ca­tion #44, tended every week by Neo-Venezia Lt­d’s gar­den­er­s—it would in­stantly lose most of its charm.

Caith Sith and his cat fol­low­ers sup­ply that mys­tery; they are the au­thors. Where does that train go? How does that cafe ap­pear only to Akari and to every­one else as a dere­lict? What do the cats dis­cuss at their con­claves? Why does Cait Sith an­nu­ally per­form as Casanova dur­ing Car­ni­vale? This mys­ter­ies con­cern not just Akari, but every­one—the ru­mors spread, and peo­ple on the pe­riph­ery are affect­ed. Per­haps Caith Sith et al’s only role is to cre­ate this mys­tery, per­haps they are charged with this mis­sion by the ‘heart of Aqua’ that Ali­cia speaks so vaguely of. The ghost may be man­u­fac­tured, and Sith sim­ply saves any­one who is care­less enough to be taken by her (how could the ru­mor spread if she only ap­proaches soli­tary undi­nes, and kills each one? What liv­ing wit­ness­es?) The per­va­sive il­lu­sion of seam­less­ness can en­rap­ture and lead to the fa­mous iyashikei effec­t—a sense of peace and heal­ing. The sim­plic­ity and con­sis­tency lull us, and draw our minds like how an­i­mals draw us (it is no co­in­ci­dence we see so many do­mes­tic or tame an­i­mals in Aria and in iyashikei works in gen­eral5); Schopen­hauer com­ments:

What a pe­cu­liar plea­sure it affords us to see any free an­i­mal look­ing after its own wel­fare un­hin­dered, find­ing its food, or tak­ing care of its young, or as­so­ci­at­ing with oth­ers of its kind, and so on! This is ex­actly what ought to be and can be. Be it only a bird, I can look at it for some time with a feel­ing of plea­sure; nay, a wa­ter-rat or a frog, and with still greater plea­sure a hedge­hog, a weasel, a roe, or a deer. The con­tem­pla­tion of an­i­mals de­lights us so much, prin­ci­pally be­cause we see in them our own ex­is­tence very much sim­pli­fied.6

But the same con­sis­tency and sim­plic­ity and sur­face ap­peal can, with a twist, plunge us into a sort of ex­is­ten­tial hor­ror sce­nar­io, where we de­cide the en­tire world is a lie and the il­lu­sion is not for our ben­e­fit, where our hap­pi­ness (Solon: “call no man happy un­til he be dead”) turns out to a ve­neer that is very thin in­deed—is the truth some­thing we were ar­guably bet­ter off with­out, as in , or is it some­thing whose ab­sence will de­stroy us ut­ter­ly, as in ? Stephen­son again:

…Dis­ney World works the same way. If you are an in­tel­lec­tual type, a reader or writer of books, the nicest thing you can say about this is that the ex­e­cu­tion is su­perb. But it’s easy to find the whole en­vi­ron­ment a lit­tle creepy, be­cause some­thing is miss­ing: the trans­la­tion of all its con­tent into clear ex­plicit writ­ten words, the at­tri­bu­tion of the ideas to spe­cific peo­ple. You can’t ar­gue with it. It seems as if a hell of a lot might be be­ing glossed over, as if Dis­ney World might be putting one over on us, and pos­si­bly get­ting away with all kinds of buried as­sump­tions and mud­dled think­ing.

The true civ­i­liza­tion re­mains un­known. Per­haps it is run by AIs—per­haps this is what looks like on the other side. There are worse utopi­as, after all, than ones that give us a bo­hemian Eu­ro­pean lifestyle. Like (“See­ing A Post-Sin­gu­lar­ity World Through Pre-Sin­gu­lar­ity Eyes”) or , one might call Aria a “weird­topia”.

Hu­mans kept as happy pets? (The cats then would be a supreme irony on the part of the AIs.) It’s more likely than you think.

To sum­ma­rize: Aria de­picts a vast fu­tur­is­tic con­spir­acy in which cute girls are ma­nip­u­lated by even cuter-look­ing amoral alien en­ti­ties to cover up an op­pres­sive and soul-killing re­al­i­ty. So it’s prob­a­bly set in the uni­verse of :

A vi­sual com­par­i­son of Aria Com­pa­ny’s cat and Madoka’s Kyubey.

  1. A su­per­fi­cial­ity en­cour­aged by deep trends in Wikipedi­a’s ed­i­tor com­mu­ni­ty; see . The prospects are poor for this be­ing reme­died by the , given to work on Wikipedia specifi­cal­ly.↩︎

  2. Likely a ref­er­ence to the movie or its de­scen­dant, ; al­lu­sions to Japan­ese me­dia seem gen­er­ally rare in Aria, height­en­ing the sense of clo­sure and be­ing an­other world.↩︎

  3. Moore’s law has a ways yet to run. It would only take an­other decade or two for some truly im­pres­sive com­put­ing pow­er. And even if Moore’s law stopped, we could ex­pect con­sid­er­able effec­tive turns of Moore’s law, as the ab­sence of Moore’s law now means that it is worth­while to in­vest in al­go­rith­mic im­prove­ments and op­ti­miza­tion: this is cur­rently ex­pressed as “ : com­pil­ers will dou­ble the speed of ex­ist­ing pro­grams every 18 years.” (See also Arnold et al 2000.) This is be­cause in part, there is con­sid­er­able low-hang­ing fruit which re­main un­plucked be­cause an­other few dou­blings will equal the gains and there are—cur­rent­ly—­bet­ter things to do, like adding more fea­tures.

    are often huge. For ex­am­ple gained a con­stant fac­tor of >10 just by switch­ing to a of its core data-struc­ture, the ex­tremely well-un­der­stood . (In the same vein, the ob­streper­ous Ul­rich Drep­per has writ­ten , demon­strat­ing how var­i­ous uses of RAM and caches can lead to con­stant fac­tors greater than 10.) A Russ­ian pro­gram­mer in ~2010 has demon­strated a 20% gain on the highly op­ti­mized Java im­ple­men­ta­tion of (pos­si­bly the most stud­ied al­go­rithm in an old & ex­tremely stud­ied sub­-field, due to the fact that sort­ing used to be the most com­mon com­puter task, see ’s Sort­ing and Search), called Du­al-Pivot Quick­sort. Com­piler op­ti­miza­tion, often as­sumed to be played out, has much room for im­prove­ment by in­cor­po­rat­ing or ; al­ready im­ple­mented fea­tures like are rarely used. Op­er­at­ing sys­tems have barely ex­plored op­ti­miza­tions demon­strated by ob­scure and aban­doned ar­eas of re­search like the or . (All of the above is usu­ally done in pure soft­ware, but hard­ware and soft­ware are a con­tin­u­um, and large con­stan­t-fac­tor speedups can be gained from cus­tom hard­ware like FPGAs or al­ter­na­tive ar­chi­tec­ture like s or spe­cial­ized ASICs; re­lax­ing re­quire­ments can also yield large gains, eg. Joe Bates claims that float­ing point cal­cu­la­tions could re­quire 100 times fewer tran­sis­tors if small (1%) er­ror bars were ac­cept­ed, and Ben­jamin Vigoda offered sim­i­lar speedups for sta­tis­ti­cal cal­cu­la­tions us­ing slightly er­ro­neous ana­logue cir­cuit­s.) The area of AI is al­most en­tirely un­ex­plored—and .

    Al­go­rithms re­search some­times re­sults in use­less galac­tic al­go­rithms, but some­times re­sults in mind-bog­gling gains. Im­prove­ments in or are some­times said to have re­sulted in speedups greater than Moore’s law since the 1970s. has seen sim­i­lar al­go­rith­mic speedups, which have been quan­ti­fied (from “Re­port to the Pres­i­dent and Con­gress: De­sign­ing a Dig­i­tal Fu­ture: Fed­er­ally Funded R&D in Net­work­ing and IT”, em­pha­sis added):

    The al­go­rithms that we use to­day for speech recog­ni­tion, for nat­ural lan­guage trans­la­tion, for chess play­ing, for lo­gis­tics plan­ning, have evolved re­mark­ably in the past decade. It’s diffi­cult to quan­tify the im­prove­ment, though, be­cause it is as much in the realm of qual­ity as of ex­e­cu­tion time.

    In the field of nu­mer­i­cal al­go­rithms, how­ev­er, the im­prove­ment can be quan­ti­fied. Here is just one ex­am­ple, pro­vided by Pro­fes­sor Mar­tin Grötschel of Kon­rad-Zuse-Zen­trum für In­for­ma­tion­stech­nik Berlin. Grötschel, an ex­pert in op­ti­miza­tion, ob­serves that a bench­mark pro­duc­tion plan­ning model solved us­ing lin­ear pro­gram­ming would have taken 82 years to solve in 1988, us­ing the com­put­ers and the lin­ear pro­gram­ming al­go­rithms of the day. Fifteen years lat­er—in 2003—this same model could be solved in roughly 1 min­ute, an im­prove­ment by a fac­tor of roughly 43 mil­lion. Of this, a fac­tor of roughly 1,000 was due to in­creased proces­sor speed, whereas a fac­tor of roughly 43,000 was due to im­prove­ments in al­go­rithms! Grötschel also cites an al­go­rith­mic im­prove­ment of roughly 30,000 for mixed in­te­ger pro­gram­ming be­tween 1991 and 2008.

    Grötschel re­port­edly is draw­ing on Robert Bixby’s “Solv­ing re­al-world lin­ear pro­grams: a decade and more of progress”, who notes that in­te­ger pro­gram­ming al­go­rithms im­proved even more than lin­ear pro­gram­ming al­go­rithms (pg2):

    In­te­ger pro­gram­ming makes di­rect use of all the ad­vances we will dis­cuss in LP al­go­rithms. In ad­di­tion, there have been other ma­jor ad­vances that are do­main spe­cific to in­te­ger pro­gram­ming, such as the use of cut­ting planes and in­te­ger-pro­gram­ming-spe­cific pre­solve tech­niques. These two classes of meth­ods alone often trans­form mod­els from be­ing un­solv­able to straight­for­ward. There is lit­tle doubt that the over­all im­prove­ment in pre­sen­t-day in­te­ger-pro­gram­ming codes ex­ceeds that for lin­ear pro­gram­ming.

    And the im­prove­ments to the lin­ear pro­gram­ming al­go­rithms work even on ‘chal­lenge prob­lems’ (pg5):

    The de­gen4 model is a larger ver­sion of the netlib mod­els de­gen2 and de­gen3, and is much more diffi­cult. It is an early in­stance of an air­line fleet-as­sign­ment mod­el. In late 1989, de­gen4 was pre­sented as a chal­lenge prob­lem to op­ti­miz­ers and com­puter ven­dors…What was miss­ing was steep­est edge for the dual. That fi­nal piece of the puz­zle was pro­vided by For­rest and Gold­farb (1992), who in­tro­duced a par­tic­u­larly effec­tive ap­proach to steep­est-edge pric­ing for the dual. This mod­i­fi­ca­tion not only works well on de­gen4, but in gen­er­al. It is one of the key rea­sons why the dual sim­plex al­go­rithm has emerged as a pow­er­ful al­l-pur­pose al­go­rithm for lin­ear pro­gram­ming.

    Cus­tomized al­go­rithms were de­vised for these chal­lenge prob­lem­s—and then su­per­seded by gen­er­al-pur­pose im­prove­ments (pg7):

    Be­cause of the diffi­culty of these mod­els, they have re­ceived con­sid­er­able at­ten­tion in the LP lit­er­a­ture, and sev­eral spe­cial-pur­pose al­go­rithms have been de­vel­oped. To my knowl­edge, the most re­cent and best of these al­go­rithms is de­scribed in Cas­tro (2000). The largest model solved by Cas­tro was pds90, with a so­lu­tion time of 21,781 sec­onds on a 200 MHz Ul­tra­Sparc. As we shall see, Cas­tro’s al­go­rithms are now dom­i­nated by cur­rent gen­er­al-pur­pose im­ple­men­ta­tions of the dual sim­plex al­go­rithm.

    (Sec­tion 5.3 notes that iron­i­cally enough, the im­prove­ments are so large that they com­pli­cate bench­mark­ing & com­par­i­son.) For some other ar­eas in­clud­ing in­te­ger fac­tor­ing, chess & Go play­ing, and ma­chine learn­ing, see . And that is to say noth­ing of fu­ture re­search. Here’s an ad­mit­tedly un­likely sce­nar­io: were re­search to dis­cover with rea­son­able con­stant fac­tors, this alone would be worth decades of Moore’s law.↩︎

  4. , speech, Lon­don (1880-03-10)↩︎

  5. Al­though it is un­usual that Aria is set in a city at all, given cities’ as­so­ci­a­tion with stress & mad­ness. It is prob­a­bly worth not­ing that Venice is one of the few real world-fa­mous cities where an iyashikei ap­proach can work; an­other work set in, say, Tokyo or Ky­oto, would prob­a­bly be no more than a or work. I like , but one could not call it iyashikei.↩︎

  6. , “Psy­cho­log­i­cal Ob­ser­va­tions”, Es­says↩︎