AI as a Universal Solvent: autonomous decision systems and their 'disjointed instrumentalism'

Enviar para

Título:
AI as a Universal Solvent: autonomous decision systems and their 'disjointed instrumentalism'
Autor: Taylor, Simon
Assuntos: algorithms ; artificial intelligence ; data ethics ; drones ; history of technology ; infrastructure studies ; machine learning ; robotics ; science and technology studies ; Sociology and social studies of science and technology
Descrição: Artificial Intelligence capabilities accelerated from 2010. This machine learning era is called a “golden decade” for autonomous decision systems and “the biggest experiment of classification tasks in human history” (Crawford 2019). Yet the genealogical foundations and technical elements underpinning uses in large-scale AI remain under-examined and inadequately theorised. This thesis takes its title from historian Paul Ceruzzi’s claim that as computation is mobilised across domains of expertise and social arenas it can be conceived a “Universal Solvent” (2012). This positions AI as a form of non-domain-specific problem solving. AI impacts knowledge in internet use, scientific epistemologies, financial analytics, robotics, administrative systems, and medical diagnosis, to name a few. Furthermore, this is constituted by generalised and cross-domain techniques in the trade of datasets, statistics, models, algorithms, and devices that are hard to get a grip on and to explain. The implication is this leads to “black-boxes” in AI systems. These have a critical relevance to the public understandings of how technical decisions are made, their accountability, and to the impacts on society, people and disparate fields of knowledge. Using a methodology grounded in a subset of science and technology studies (STS) called infrastructure studies, a multiple case-study approach is undertaken. As a line of enquiry to examine “the trading zones”of AI development the research investigates four use-cases that initially appear unrelated – namely facial recognition, operational sensing, statistical classification, and drone datasets. The thesis shows how each use-case requires prototyping, trading of scientific development, and AI infrastructure. The research reveals how machine learning trades in networks – scientific, technical, historical and social – and is driven by a) historical legacies, b) toolsets across domains and in digital systems, and c) the routinisation and standardisation of autonomous processes. This development in AI computation shows how specific instruments are made amenable to local uses, but can apply to anywhere. This can distribute risks and errors in systems, across domains, transform site-specific uses into generalised applications, host incompatible or excessive data reduction, and remove aspects of human oversight. The objective of this thesis then is to reveal, explain, and understand a “disjointed” trade of decision-making built in AI tools and systems. By investigating these as foundations for machine learning there is a potential to contribute to regulatory concerns within AI infrastructure, for data provenance, transparency, and accountable actions. Source: TROVE
Data de criação/publicação: 2023
Idioma: Inglês

AI as a Universal Solvent: autonomous decision systems and their 'disjointed instrumentalism'

Taylor, Simon

2023

Buscando em bases de dados remotas. Favor aguardar.