Название: Algorithms: Technology, Culture, PoliticsАвтор: Tobias Matzner
Издательство: Routledge
Год: 2024
Страниц: 193
Язык: английский
Формат: pdf (true)
Размер: 10.2 MB
Algorithms: Technology, Culture, Politics develops a relational, situated approach to algorithms. It takes a middle ground between theories that give the algorithm a singular and stable meaning in using it as a central analytic category for contemporary society and theories that dissolve the term into the details of empirical studies.
The book discusses algorithms in relation to hardware and material conditions, code, data, and subjects such as users, programmers, but also “data doubles”. The individual chapters bridge critical discussions on bias, exclusion, or responsibility with the necessary detail on the contemporary state of information technology. The examples include state-of-the-art applications of machine learning, such as self-driving cars, and large language models such as GPT.
In consequence, as soon as programming becomes more than the most basic listing of machine code, it depends on algorithms in a two-fold manner: First, writing code is not just programming algorithms but most often calling other algorithms that are readily available in programming libraries. Second, code needs to be processed by specific algorithms, such as compilers, before it can execute the algorithm it “implements.”
Many textbooks from computer science describe this manifold transformation of code by other algorithms as a set of translations where an abstract mathematical algorithm is implemented in code which is in turn translated into machine instructions. This implies a platonic idea of an algorithm that stays the same throughout its different materializations as mathematical notation, program code, and machine operations. It resonates strongly with the cybernetic conviction that information processing is decisive, not the carrier of information.
Machine Learning (ML) is the main reason why data have become a key element, sometimes even more emphasized than algorithms, in critical discussions of digital technology. Data are particularly important in Machine Learning because they are no longer just the input of an algorithm once it is programmed. Data are also used to create – or as it is often framed “train” – the algorithm in the first place. Very generally speaking, machine learning can be described using two algorithms instead of one. Rather than programming an algorithm for a task, programmers specify how the task is to be solved but leave specific elements open. I call this the task algorithm. Most of the time, programmers specify that a particular probabilistic model is to be used – such as the infamous artificial neural networks (ANNs) – but leave the parameters of that model unspecified. These parameters are determined by a second algorithm – the training algorithm – that uses training data to find good estimates of these parameters. Thus, the training algorithm completes the elements in the task algorithm that had been left unspecified by the programmers. There are cases where the training is done just once and then the task algorithm is deployed and runs like any other algorithm. Often, though, training is done constantly. That is, both task and training algorithm are deployed and the training algorithm constantly adapts the task algorithm to new incoming data.
Скачать Algorithms: Technology, Culture, Politics