2.3.8 algorithm ididekile

Ukuziphatha kumahlelo amakhulu wedatha akuyona yemvelo; iqhutshwa imigomo yobunjiniyela yezinhlelo.

Nakuba imithombo eminingi yedatha enkulu ingasebenzi ngenxa yokuthi abantu abazi ukuthi idatha yabo ibhaliwe (isigaba 2.3.3), abacwaningi akufanele bacabange ukuziphatha kulezi zinhlelo ze-intanethi ukuba "kwenzeke ngokwemvelo." Eqinisweni, izinhlelo ze-digital ezirekhoda ukuziphatha isungulwe kakhulu ukudala ukuziphatha okuqondile njengokuchofoza kuzikhangiso noma ukuthumela okuqukethwe. Izindlela imigomo yabakhi bezinhlelo ezikwazi ukuletha amaphethini ku-data kuthiwa i- algorithmic confounding . Ukuphazamiseka kwe-algorithm akungaziwa kwabesosayensi bezenhlalakahle, kodwa kuyinkinga enkulu phakathi kwabesosayensi abaqaphele. Futhi, ngokungafani nezinye zezinkinga nge-digital traces, ukuphazamiseka kwe-algorithmic kubonakala kungabonakali.

Isibonelo esilula kakhulu sokuphazamiseka kwe-algorithmic siwukuthi ku-Facebook kukhona inani eliphezulu labasebenzisi abangabangani abangaba ngu-20, njengoba kutholakala nguJohan Ugander kanye nozakwethu (2011) . Ososayensi bahlaziya le datha ngaphandle kokuqonda ukuthi i-Facebook isebenza kanjani ngokungangabazeki idala izindaba eziningi mayelana nendlela engu-20 uhlobo oluthile lwenombolo yomphakathi yemilingo. Ngenhlanhla, u-Ugander nabalingani bakhe babenokuqonda okukhulu kwenqubo eyenza idatha, futhi bazi ukuthi i-Facebook yakhuthaza abantu abanokuxhumana okuncane ku-Facebook ukwenza abangani abangaphezulu baze bafinyelele abangani abangu-20. Nakuba u-Uganda nezakwabo bangasho lokhu emaphepheni abo, kungenzeka ukuthi le nqubomgomo idalwe ngu-Facebook ukuze kukhuthazwe abasebenzisi abasha ukuba basebenze ngokwengeziwe. Ngaphandle kokwazi ngokukhona kwalolu mgomo, kulula ukudweba isiphetho esingalungile kusuka kwedatha. Ngamanye amazwi, inani elimangalisayo labantu abanabangane abangaba ngu-20 lisitshela kabanzi mayelana ne-Facebook kunokuziphatha kwabantu.

Kulesi sibonelo esandulele, ukuphazamiseka kwe-algorithmic kuveza umphumela we-quirky wokuthi umcwaningi ocophelelayo angase ahlole futhi aphenye okuqhubekayo. Kodwa-ke, kukhona inguquko eyinkimbinkimbi ye-algorithmic confounding eyenzeka lapho abaklami bezinhlelo ze-intanethi bezazi izinkombandlela zomphakathi bese babhalela le mibono ekusebenzeni kwezinhlelo zabo. Ososayensi bezenhlalakahle bathi lokhu kusebenza : uma inkolelo ishintsha umhlaba ngendlela eletha izwe ukuthi lihambisane ne-theory. Endabeni yokwenza ama-algorithmic adidekile, isimo esiyinkimbinkimbi sedatha kunzima ukubona.

Isibonelo esisodwa sephethini esakhiwe ngobuciko ukuguqula ekuxhumaneni komphakathi kwe-intanethi. Ngomnyaka we-1970 nango-1980, abaphenyi bathola ngokuphindaphindiwe ukuthi uma ungabangani bobabili u-Alice noBob, u-Alice noBob banamathuba okuba bangabangane nomunye ngaphandle kokuthi babe ngabantu ababili abakhethiwe. Leli phethini elifanayo (Ugander et al. 2011) sezenhlalo ku-Facebook (Ugander et al. 2011) . Ngakho-ke, umuntu angaphetha ngokuthi amaphethini obungane emaphethini aphindaphindiweyo e-Facebook angabangani abangekho ku-intanethi, okungenani ngokwemvelo. Kodwa-ke, ubukhulu bokuguquka kwemvelo ku-Facebook social graph kuqhutshwa kancane yi-algorithmic confounding. Lokhu kungukuthi, ososayensi be-data ku-Facebook babesazi ngokucwaninga ngemibono nangokwamazwi mayelana nokuguquka komzimba bese bewabheka ukuthi i-Facebook isebenza kanjani. I-Facebook ine-"People You May Know" isici esikisela abangane abasha, futhi enye indlela i-Facebook enquma ukuthi ubani ongakuphakamisa ukuguquka. Okungukuthi, i-Facebook ingasikisela ukuthi ube abangane nabangani bakho. Lesi sici sinomthelela wokwandisa ukuguquka komqondo ku-Facebook social graph; Ngamanye amazwi, inkolelo yokuguquguquka kwemvelo yenza izwe lihambisane nokubikezela kwenkolelo (Zignani et al. 2014; Healy 2015) . Ngakho-ke, lapho imithombo emikhulu yedatha ibonakala ikhiqiza izibikezelo zenkolelo yezenhlalakahle, kumelwe siqiniseke ukuthi le ncazelo ngokwayo ayizange ibhekiswe endleleni isistimu esebenza ngayo.

Esikhundleni sokucabanga ngemithombo emikhulu yedatha njengokubheka abantu esimweni esingokwemvelo, isifaniso esiphezulu esingaphezu kokubona abantu e-casino. I-Casino inemvelo enhle kakhulu eyenzelwe ukudala ukuziphatha okuthile, futhi umcwaningi akasoze alindela ukuziphatha ekhasino ukuze ahlinzeke ngefasitela elingapheli ekuziphatheni komuntu. Yiqiniso, ungafunda okuthile ngokuziphatha kwabantu ngokutadisha abantu kuma-casino, kepha uma ungahambanga iqiniso lokuthi idatha idalwe ekhasino, ungase uthathe iziphetho ezimbi.

Ngeshwa, ukubhekana nokuphazamiseka kwe-algorithmic kuyinkimbinkimbi ikakhulukazi ngoba izici eziningi ze-intanethi ziyi-proprietary, zibhalwe kahle futhi ziguquguquka njalo. Isibonelo, njengoba ngizochaza ngokuhamba kwesikhathi kulesahluko, ukuphazamiseka kwe-algorithmic kungenye yezincazelo zokuhlukana kancane kancane kwezinguquko ze-Google Flu (isigaba 2.4.2), kodwa lokhu kubanga kunzima ukuhlola ngoba ukusebenza kwangaphakathi kwe-algorithm yokusesha ye-Google i-proprietary. Uhlobo olushukumisayo lwe-algorithmic confounding luhlobo olulodwa lwe-system drift. Ukuphazamiseka kwe-algorithm kusho ukuthi kufanele siqaphe nganoma yisiphi isimangalo esiphathelene nokuziphatha kwabantu okuvela ohlelweni olulodwa lwe-digital, kungakhathaliseki ukuthi lukhulu kangakanani.