Text
E-book Democratic Algorithms : Ethnography of a Public Recommender System
Thisisabookaboutsocialorder.Morespecifically,itisaboutthecompli-cated relationship between machine learning algorithms and the formation of democratic order. And there are good reasons for such a book. Algorithms appeartobecomeadefiningmomentinthedigitizedsocietiesofthe21stcentury – and it seems that no domain of life is spared from the rational and seeminglypowerfulinfluenceofthecomputationallogicofalgorithms.Atleastthisishowthe“alluringandcompellingdrama”(Neyland2016,51)isbeingtold.Anditseemstrue.Filterbubblesthreatendemocraticdiscourseandopinion formation, algorithmic risk scores are being applied in law enforce-ment and judicial reasoning, and policy is increasingly based on algorithmic evidence.Algorithmsspecifically,anddigitaltechnologiesingeneral,havebecome deeply embedded in our social life, in contemporary societies, and in theinstitutionsofourdemocracies(BergandHofmann2021).Thisfactevenprovoked the question, whether democracy will survive the rise of AI (Helbing etal.2017).The focus of this book is related to these questions, but also inverts them: instead of asking how algorithms are changing our contemporary democracy, Iwillcriticallyexaminetheeffortsofademocraticinstitutiontomakealgorithmsmoredemocraticallyaligned.Startinginearly2016,Ijoinedtheinitiative of a public broadcaster in Germany that sought to develop a new website with a video-on-demand system. To accomplish this, a software devel-opment team was gathered to design and implement the new site, and which would “air” the same shows from the linear program – but in a non-linear way. Further,becausethiswebsitewasintendedtoimplementthelatestavailablefeatures, a recommender system was envisioned as a central element of the setup. This, however, created some challenges on the normative side. Public broadcasting entities in Germany have a legal obligation to adhere to the German constitution, which states that such broadcasters must distribute a broad range of information about political, social, and cultural events in Germany.Recommendersystemsdoexactlytheoppositeofthat;theyselectinformationpiecesbasedonsimilarity,notdiversity–theytendtocreatefilterbubbles. The relation between democratic order and algorithmic systems should havenowbecomeapparent.However,thespecificchallenge–notonlyforthisspecificbroadcasterbutforalldemocraticinstitutions–is:canwebuilddemocratic algorithms? Can we translate our central shared values, beliefs, and norms into machine learning systems and, if so, how? Part of the answer is that we have to go beyond a technical understanding of algorithms, but without losing the technical systems out of sight in the process. Instead, I argue that algorithmic systems are complex socio-technical systems in which the algorithm, as an entity, is enacted. This enactment is a symmetrical process, in which the technical aspects must aligned with their surrounding practicesandexpectations–andviceversa.Further,algorithmicsystemsareenactedinmultiplewayswithdifferenttranslationsandoutcomes.InthespecificcasethatIdiscusshere,therecommendersystemwasenactedinthreedifferentwaystoformdifferentcollectives:adisciplinaryenactmentofthe algorithm as technical object, an institutional enactment of the algorithm asorganizationalentity,andapoliticalenactmentofthealgorithmasanormative and legal challenge. Each of these enactments tried to shape the algorithminitsownway,enablingorhinderingspecificimplementations.Thechallenge for the actual implementation of the algorithm was to align these threedifferentenactmentsintoabroadersocio-technicalsystem.Theresultcomplicates the notion of the algorithm, because the boundaries of its enact-ment become a political question as well.
Tidak tersedia versi lain