Glossary

Ether’s Bloom: A Programme on Artificial Intelligence

This glossary was co-authored in collaboration with the LARGE LANGUAGE MODELS ChatGPT and Bard and translated into German with DeepL. Large Language Models are being increasingly used to generate texts across different fields. This glossary is an attempt to develop an understanding of how these models work and how they can be put to work to arrive at factual, interesting or curious results. As these technologies are becoming more common, it is crucial to learn how to instruct them to work effectively and in doing so to also identify their limits. Based on a list of terms that are circulating in the context of ARTIFICIAL INTELLIGENCE (AI), the request  – which is called prompt – was to give short definitions that should take ethical considerations into account: “Can you give me a short definition of [...] in an ethical and accessible way?”. The responses were edited by the project team along with the guidance of Maya Indira Ganesh and blended into one entry per term. To make these perspectives visible, the italic font highlights the editorial revisions. This glossary is not exhaustive or universally applicable and will grow over the course of Ether’s Bloom: A Programme on Artificial Intelligence.

ALGORITHM

An algorithm is a set of instructions or rules that a computer programme or system follows to solve a problem or accomplish a task. It is used in all sorts of different applications, from the simple (like sorting a list of numbers) to the complex (like advertising products to customers). Algorithms can have a significant impact on individuals and society, influencing individual and collective decisions. One such example is when the algorithmic dissemination of false information manipulates the outcome of elections. However, the algorithm is made up of many other parts, like DATA and statistical techniques, and is designed by people: programmers are in constant interaction with algorithms, shaping them to do different kinds of things depending on the context.

ARTIFICIAL INTELLIGENCE

Artificial intelligence (AI) is a rapidly evolving field, and there is no single definition. However, most definitions share the understanding that AI is the ability of machines to perform tasks that are typically associated with human cognitive abilities, such as learning, reasoning and problem-solving. AI involves multiple technical infrastructures: big data, MACHINE LEARNING MODELS, data science techniques, COMPUTER VISION, natural language, AUTOMATION, among others. AI manifests as many kinds of technologies developed by varied organisations through global supply chains. It adopts specific, situated theories of the human and intelligence and builds on distinct fantasies of the future.

AUTOMATION

Automation is a process of using mechanical, cyber-physical, or digital technologies to scale a task so it might take place faster and with fewer errors. Automation usually requires a clear set of instructions and the power to implement them. There are many different kinds of automation all around us and the industrial age saw a great leap in automation, when workers and machines had to work together to increase productivity. 

AVATAR

An avatar is a virtual representation of a person or entity in digital environments. It can take different forms such as an animated character, a voice or a three-dimensional model. Avatars are commonly used in online games, social media platforms, virtual reality experiences and other digital spaces. Because they are disembodied, avatars can work as powerful tools for communication and interaction, offering a proxy for identity and expression in the digital realm. 

BIASED DATA

Biased data refers to information that is distorted and unfairly represents certain groups or characteristics in a dataset. This can happen when the data collection process or the sources used are not diverse or inclusive enough, leading to inaccurate conclusions. There are different ways that DATA is biased and these tend to overlap: selection bias, where the data is collected from a sample of the population that is not representative, and this might emerge from the invisibility of certain communities, or limited resources; measurement bias where the data is collected in a way that introduces bias; and processing bias where the data is analysed in a way that introduces bias. All of these forms of bias can reproduce discrimination against certain groups of people. Through education, mitigation and identification, biased data can be located and addressed to avoid harmful results. However, it is important to recognise that in research and statistical terms, all data is biased in some way or the other. A dataset is just a sample. ‘Bias’ may not always be a value judgment; it is also just the nature of statistical sampling techniques. 

COMPUTER VISION

Computer vision is a set of technologies that enables computers to interpret visual information from the world, emulating visual perception in humans. It involves teaching computers to perceive, and then recognise objects, people, text and other visual elements in images and videos, and in the physical world. Computer vision is applied in self-driving cars, facial recognition, medical imaging and augmented reality, among others. By analysing imagery, computer vision can identify and detect different aspects of life that can simplify daily tasks. Privacy issues and biases in ALGORITHMS amplify the harmful outcomes of the application of such technologies.

DATA

Data refers to digital and non-digital information collected, stored and processed by any kind of information technology, including computers. Data can be collected in many different ways, through surveys, experiments, observations or tracking of digital activities. They can be stored and organised in formats such as spreadsheets, databases, or text files. Data about data is called metadata and is in itself powerful because it can tell us about the contexts of data collection. Data is fundamental to AI.

DATA CLEANING

DATA that “feeds” MACHINE LEARNING is sourced from many custom and generic sources (such as social media); but it is rarely just ready “off the shelf”. Data needs to be prepared. Data cleaning is the process of identifying and correcting errors or inconsistencies in datasets to ensure that it is accurate, reliable and ready for use. Some types of data prep  require human work like labeling and annotating images for machine vision. Data cleaning can be done manually or with the help of software. The specific steps involved will vary depending on the type of data and the problems that need to be addressed encompassing tasks like removing duplicates, filling in missing values, and addressing outliers. This process helps prevent misinformation, bias, or unfairness that can arise from flawed or false data. It, all aimed at producing high-quality data that can be used for various purposes.

DEEP LEARNING

Deep Learning is a subfield of MACHINE LEARNING that involves training complex neural NETWORKS, which are computer systems inspired by the structure of the human brain, to process and understand large amounts of data. It allows machines to recognise patterns, to make predictions and to solve tasks of image and speech recognition. Deep learning is a powerful tool, but as it is often trained on BIASED DATA it reproduces existing forms of discrimination

INTELLIGENCE

Intelligence refers to the ability of an individual or system to understand, learn, reason and solve problems. It encompasses the capacity to gather and process information, apply knowledge, adapt to new situations and make informed decisions. However, intelligence is neither fixed nor universal and can be developed and improved through time, learning, empathy and experience and encompasses social and emotional intelligence. A broad and inclusive understanding of intelligence extends beyond human-centred notions and recognises that different species – and indeed, different kinds of humans – possess diverse forms of intelligence.

INTERSPECIES

The term interspecies refers to interactions or relationships between different species. It recognizes that the world is not just composed of one type of living being, but a diverse array of species living in complex relationships on planet earth. Interspecies interactions can take many forms, such as cooperation, mutual benefit, competition or even parasitism. Understanding and respecting interspecies relationships involves acknowledging the needs and rights of all species beyond biological concepts as well as their roles within ecosystems and life cycles.

KINSHIP

Kinship refers to connections and relationships that go beyond biological connections and the human or living realm and extend to the cultural, emotional and social aspects of family, community and cohabitation. Kinship plays a vital role in shaping our sense of belonging, support systems and responsibilities towards one another.

LARGE LANGUAGE MODEL

Large Language Models (LLM) use DEEP LEARNING techniques and massively large DATA sets to analyse, summarise, generate and predict new human-like text. The data in LLMs come from social media posts, online articles, online courses, news: essentially anything that is freely available online. Typically, LLMs are not based on carefully curated datasets. While the underlying technology, transformer models, have existed for at least five years, usable applications like ChatGPT have become available only recently.

MACHINE LEARNING

Machine Learning enables computers to complete tasks without being explicitly programmed to do so. This is done by exposing ALGORITHMS to large amounts of existing DATA to determine patterns and correlations therein. Computational models represent these patterns and can make predictions when exposed to new data it has never been exposed to before. Machine Learning is present in different applications such as spam filtering, medical diagnosis, and in COMPUTER VISION used in driverless cars.

NETWORK

A network is a system of interconnected devices that can communicate with each other, so that they can share information and resources. Only if the network centres accessibility, it ensures that information and resources are available and usable for many individuals. It allows people to access services, engage in online communities and participate in the digital realm, making it an essential part of our interconnected world – of which many are continuously being excluded.

PLANT INTELLIGENCE

Plant intelligence refers to the ability of plants to perceive and respond to their environment. While the the term “plant intelligence” is still being debated by scientists, as it can be misleading or anthropocentric, there is growing evidence that plants are more “intelligent” than biologists have considered. While not the same as human or animal intelligence, it includes processes like sensing light and gravity, communicating with other plants and adapting to changing conditions. Thus the concept challenges traditional notions of intelligence and highlights the complexity of the natural world. 

PROMPT

A prompt is a short piece of text, a sentence or a question, that is used to initiate a conversation or to provide instructions. Within LARGE MODELS for text or images, a prompt is the input provided by a user to generate a desired response. Prompts serve as a way to guide the conversation, generate an image or information from an AI model.

SPIRITUAL INTELLIGENCE

Spiritual intelligence refers to the capacity to explore and understand one’s inner self and beyond, connect with deeper meanings and values in life, and find purpose outside of material or superficial pursuits. It is a complex and evolving concept that aims at an increase in self-awareness with a focus on psychological depth and understanding.