Apurahat ja residenssipaikat

Tiede

The Statistical Construction of Place

Hakemuksen tiivistelmä

Statistical modelling shadows our every move today. Indeed, natural language processing, a subfield within artificial intelligence (AI), can now be used to represent how, when, and where we attach meaning. Modelling places, the very moments and sites in which we cohere as social beings, is increasingly at the core of research by companies such as Airbnb and Google. For my PhD, I will study and apply statistical and machine learning models to spatially and temporally encoded, linguistic data. In the first part of my thesis, I will produce a critical genealogy of how meaning-making in general, and, its temporal and spatial context in particular, have been theorised and modelled in AI and statistics research in industry and academia. In the second part, I will flip the perspective, and compare the representations of place that are produced when implementing in practice the models studied in the first part. How have cities changed over time, when viewed through historical records of Airbnb reviews and descriptions? Where does one place begin, and another end, when their boundaries are defined using the meanings attached to sites in massive archives of digitised literature? How do the answers to these questions change, as the parameters and assumptions in our models change? Through my work, I will make visible the inherent tension between the simplifications that are necessary for any mathematical model, however complex, and the singular poetics of our lived experience.

This doctoral dissertation deals with the question of "information" in the social sciences at various levels: (1) Enclosure, (2) capital, (3) metaphor, and (4) method. Today, machine learning approaches relying on information metrics and using textual data have become commonplace in all of the social sciences and humanities. Behind this turn, we find a peculiar paradox: Information, a concept which originally was strictly about engineering, is today the arbiter of "meaning", semantics, and even place. In chapters one, two, and three, I explore this paradox through a set of genealogical studies on the various moments of translation that had to be established for information to become the (1) enclosure within which meaning can be disclosed. The relationship between information and meaning has gone through a fundamental shift, where, initially, it was meaning that appeared as the agreement which established the conditions for information. As the roles have reversed, the large language model (LLM) has become an avenue for accumulating and producing a new type of (2) "cosine capital" which determines which truths can be disclosed, and which cannot. In this sense, information today also acts as a (3) habitually naturalized metaphor. Challenging prevailing theories on opacity and information asymmetry in critical housing and algorithm studies, we develop in chapters four and five a new concept of information asymmetry in the housing system and demonstrate its utility in a comprehensive study of landlord networks in the Montreal rental market. The last two chapters move from information as metaphor to information as quantitative (4) method, closing the circle from the beginning of the dissertation through two overlapping studies of Airbnb descriptions and reviews in New York City. Here, we take prevalent operationalizations of information theory as granted, focusing instead on developing a new framework for quantitative analysis in critical toponymy studies.

Takaisin apurahalistaukseen