AI for Research

Glossary

Artificial Intelligence (AI) — 1) a field of computer science that aims at developing machines able to simulate human intelligence and problem-solving capabilities [IBM]; 2) a machine learning system “that generates high-quality outputs such content, forecasts, recommendations or decisions for a given set of human-defined objectives” [ISO/IEC 22989:2022]. 

Artificial General Intelligence (AGI), or Strong AI, or Artificial Super Intelligence (ASI) — a theoretical form of AI where a machine would have an intelligence equal to humans; it would be self-aware and would have the ability to solve problems, learn, and plan for the future [IBM]. 

Artificial Narrow Intelligence (ANI), or Weak AI — AI system trained to perform (a) specific task(s), e.g. OpenAI’s ChatGPT, Apple's Siri, Amazon's Alexa, self-driving vehicles etc. [IBM]. 

Generative AI (GenAI) — deep-learning models that can generate high-quality text, images, and other content based on the data they were trained on, e.g. GPT, Claude, Midjourney etc. [IBM]. 

AI ethics — a multidisciplinary field that studies how to ensure that AI is developed and used responsibly to optimise its beneficial impact while reducing risks and adverse outcomes. This means adopting a safe, secure, unbiased, and environmentally friendly approach to AI [Coursera; IBM]. 

AI guardrails — safeguards put in place to ensure that AI-enabled technology operates safely within ethical and legal boundaries [Techopedia]. 

AI hallucinations — incorrect or misleading results that AI systems generate, presented as fact. These errors can be caused by a variety of factors, including insufficient training data, incorrect assumptions made by the model, or biases in the data used to train the model [Google]. 

Algorithm — a sequence of instructions given to a machine to perform a task or solve a problem. 

Big data — extremely large and diverse collections of structured, unstructured, and semi-structured data that continues to grow exponentially over time. These datasets are so huge and complex in volume, velocity, and variety, that traditional data management systems cannot store, process, and analyze them [Google]. 

Chatbot — a software or web application designed to imitate human conversation through text or voice interactions. 

Computer vision — a subfield of computer science that focuses on enabling computers to identify objects in images and videos [Microsoft]. 

Data mining — a process aimed at discovering patterns, correlations, anomalies and other valuable information in large data sets [Kaspersky]. 

Data science — an interdisciplinary field that combines principles and practices from mathematics, statistics, and computer science to analyse large amounts of data and extract meaningful insights for business [Amazon]. 

Deep learning — a subset of machine learning that uses multilayered neural networks, called Deep Neural Networks (DNNs) to learn from data. With multiple hidden layers, deep learning algorithms are potentially able to recognise more subtle and complex patterns [IBM]. 

Hyperparameters — in machine learning, parameters that are set manually before the model or algorithm is trained. The choice of hyperparameters is a trial-and-error process aimed at creating a model that is both accurate and efficient [Techopedia]. 

Image recognition — a computer vision task aimed at identifying objects (people, places, text etc.) and their features in an image or video.  

Large language model (LLM) — a deep learning model pre-trained on vast amounts of data that can perform a variety of NLP tasks, such as generating and classifying text, answering questions in a conversational manner, and translating text from one language to another. The label “large” refers to the number of parameters the model can change autonomously as it learns, e.g. Cohere’s Command R+ has 104B parameters, Open AI's GPT-3 — 175B parameters, and Anthropic’s Claude 3 Opus— 2T parameters [Techopedia]. 

Machine learning — a branch computer science that focuses on building systems that learn, i.e. gradually improve performance, in a supervised, semi-supervised or unsupervised way based on the data they consume [IBM]. 

Natural language processing (NLP) — a subfield of computer science that combines the power of computational linguistics with machine learning algorithms to enable computers to understand and generate human language, or data that resembles human language [IBM; DeepLearning.AI]. 

Neural network (NN), or Artificial Neural Network (ANN) — a type of machine learning algorithm inspired by the human brain and composed of interconnected nodes called neurons, organised in layers [Google; IBM]. 

Parameters — in machine learning, variables whose values are learned from the data during training. The values of the parameters determine how the model or algorithm works [Techopedia]. 

Prompt — an input that a user feeds to a generative AI system to get a desired output. 

Prompt engineering — a process of creating and refining instructions used as input to a generative AI system to get high-quality outputs [Amazon]. 

Reinforcement learning — a branch of machine learning that focuses on decision making by autonomous agents, which make decisions and act in response to its environment independent of direct instruction by a human user [IBM]. 

Semi-structured data — the “bridge” between structured and unstructured data that does not have a predefined data model and is more complex than structured data, yet easier to store than unstructured data, e.g. in JSON, CSV and XML formats [IBM]. 

Semi-supervised learning — a branch of machine learning that combines supervised and unsupervised learning by using both labeled and unlabeled data to train models [IBM]. 

Structured data, or Quantitative data — data that has a standardised format for efficient access by software and humans alike. It is typically stored as tables in a relational database and queried with the Structured Query Language (SQL) [Amazon; IBM]. 

Supervised learning — a branch of machine learning that uses labeled datasets to train algorithms to predict outcomes and recognise patterns [Google; IBM]. 

Token — in NLP, the atomic unit that models are training on and making predictions on, usually a word or a subword unit [Google]. 

Training data — the data used to train an algorithm or a machine learning model. 

Unstructured data, or Qualitative Data — data that has no predefined data model and can be stored in non-relational (NoSQL) databases and data ltoken akes, e.g. texts, images, videos, emails etc. The vast majority of existing data, ca. 80-90%, is unstructured [Amazon; IBM]. 

Unsupervised learning — a branch of machine learning where models  learns from data without human supervision [Google; IBM] 

Vision language models (VLMs) — a multimodal deep learning architecture that combines computer vision (CV) and natural language processing (NLP) capabilities. By simultaneously learning from image and text data, it can tackle such tasks as visual question answering or image captioning [HuggingFace].