AI - Short form of Artificial Intelligence.
Algorithm - An algorithm is a set of instructions given to a piece of computer software. Algorithms use data to make a decision and perform an action. At a basic level this might use simple logic tests; if the correct password is entered, then log the user in. More complex algorithms use many more data, rules and calculations to make more complex decisions.
Artificial General Intelligence (AGI) - Artificial General Intelligence (AGI), also known as General AI, should not be confused with Generative AI. AGI refers to an advanced form of artificial intelligence that possesses the ability to understand, learn, and apply knowledge across a wide range of tasks and domains in a manner comparable to or exceeding that of human intelligence. Unlike narrow AI, which is designed for specific tasks, AGI aims to exhibit general cognitive capabilities, including reasoning, problem-solving, perception, learning, and adaptability, across diverse contexts and challenges. Achieving AGI represents a significant milestone in AI research and could have profound implications for various fields, including technology, society, and the economy. AGI does not yet exist. See also Narrow AI.
Artificial Intelligence (AI) - The construction of computer programs that engage in tasks that are currently more satisfactorily performed by human beings because they require high-level mental processes such as: perceptual learning, memory organization and critical reasoning.” (Minsky et al., 2006)
Bias - Bias in GenAI is where output produced can be unfair, prejudiced or perpetuates stereotypes. Bias is often a result of Training Data that contains limited or prejudiced data, for example if an image generator was trained using a photo library where most of the photographs of doctors were white men, it might later assume that when you ask for an image of a doctor you expect to see an image of a white man.
Bot - A computer program that is designed to perform an automated, repetitive task. It is programmed to look at the data given and then perform certain tasks accordingly.
Chatbot - A Chatbot is a computer program that is designed to simulate a conversation between a person and a computer, usually with the human asking questions and the computer program attempting to answer. Chatbots can range in how advanced they work; simpler programs pick out keywords or phrases from a question and answer with links whereas more complex AI-driven chatbots can handle more complicated conversations and answer more naturally.
ControlNet. - ControlNet is defined as a group of neural networks refined using Stable Diffusion, which empowers precise artistic and structural control in generating images. This means you can use an existing image to influence the output of an AI generated image along with the prompt. For example, it can be used to specify a particular pose, the colour palette or the artistic style of your output image.
Data Poisoning - When AI generated data pollutes the training set of subsequent AI models, leading to a degradation in outputs. Also see Model Collapse and Recursion.
Deepfake - Pictures, video and voice often altered to deliberately generate misinformation and disinformation see Martin Lewis Deepfake on YouTube. As AI makes the technology more accessible it is not exclusively used by scammers but in the television and movie industry, see Deepfake Luke Skywalker on YouTube.
Freemium - Freemium product or software is one that allows users limited use of the service for free with the option for users to pay for extra features or unlimited use. The free version often limits how many times you can use it, doesn’t provide support or blocks key functions until you purchase a subscription.
General AI - See Artificial General Intelligence (AGI).
Generative AI - Gen AI is a type of artificial intelligence model that generates new data or content that is similar to, but not exactly the same as, the data it was trained on. GenAI learns the underlying patterns and structures of the training data and then uses that knowledge to create new instances that resemble the original data. GenAI models can be used for image generation, text generation, music generation and video generation.
Hallucination - GenAI can produce outputs that are surreal, bizarre, nonsensical or unexpected. These are known as hallucinations. Hallucinations can occur in AI-generated content for various reasons, including biases in the training data, the quality of the training data, errors in the model's understanding of context, or simply the probabilistic nature of the model's generation process. Fake references could be considered a form of hallucination in the context of GenAI models.
Large Language Model - A Large Language Model is a type of GenAI that uses huge amounts of text-based Training Data. Large Language Models work by looking at how often words appear together and using this to predict what word should come next, much like a highly complex predictive text.
Machine Learning - A type of AI that allows a system to learn and improve from examples without all its instructions being explicitly programmed. Machine learning systems learn by finding patterns in training datasets. They then create a model (with algorithms) encompassing their findings. This model is then typically applied to new data to make predictions or provide other useful outputs, such as translating text. Training machine learning systems for specific applications can involve different forms of learning, such as supervised, unsupervised, semi-supervised and reinforcement learning.
Model Collapse - Model collapse occurs when AI systems use training data that has been created by existing AI models, rather than real life data. When trained on model-generated content, new models exhibit defects, with degradation in the quality and reliability of outputs, and with becoming more homogenous or increasingly “wrong”. With more text being created using AI tools, the reuse of this text to train AI tools, could lead to data pollution on a large scale. Also see Data Poisoning and Recursion.
Narrow AI - Narrow AI is designed and trained for specific tasks or domains, unlike Artificial General Intelligence (AGI), which would possess human-like cognitive abilities across a wide range of tasks and domains. Generative AI tools, such as ChatGPT, Midjourney and Stable Diffusion and Gemini, are all examples of narrow AI. See also Artificial General Intelligence (AGI.
Natural Language/ Natural Language Processing - Natural language describes the way people talk to each other or describe their needs using words and sentences. With many computer programs you have to convert your commands into a machine-readable format such as computer code, software that allows Natural Language Processing can interpret your commands in human language.
Prompt - A prompt allows you to enter a question or phrase in the language you normally use to describe the task that an AI should perform. A prompt for a text-to-text language model can be a query, a command such as "write a poem about leaves falling", or a longer statement including context, instructions, and conversation history.
Prompt Engineering - Prompt engineering is a technique used to develop or refine the output from GenAI software. Prompt engineering techniques can include providing extra information or context in your prompt or giving the software further commands to refine the output.
Recursion - Recursion can occur when a GenAI model relies too heavily on its own outputs as inputs without introducing fresh or diverse information. This can lead to the degradation of output quality over successive iterations. This degradation can occur due to the accumulation of errors or biases present in the model's initial outputs, which may be magnified or compounded with each iteration. Without the introduction of new, diverse, or high-quality inputs, the model's outputs may become increasingly distorted, repetitive, or nonsensical. This phenomenon is often referred to as "hallucination," where the model generates outputs that diverge from reality or lack coherence due to the limited scope of information it is operating on. Also see Data Poisoning and Model Collapse.
Responsible AI - The practice of designing, developing, and deploying AI with certain values, such as being trustworthy, ethical, transparent, explainable, fair, robust and upholding privacy rights.
Stable Diffusion - Stable diffusion is one of the deep-learning generative AI models for text-to-image generation. It was released in 2022 by Stability AI. The stable diffusion technology is integrated into other Image Generation tools.
Token - A token is the smallest amount of information that the AI software can process. In GenAI programs, tokens are usually individual words or phrases.
Training Data - Training data is the information that a GenAI program uses to perform a task. In Large Language Models training data consists of millions of webpages, documents and text that it has scanned to make predictions. Training data is important as the accuracy, currency and size of a program’s training data can affect how well it produces outputs. Bias in training data (such as only using data in the English language) can also mean that outputs are biased.
Turing Test - The Turing Test, created by computer scientist Alan Turing, was to evaluate a machine’s ability to exhibit intelligence equal to humans, especially in language and behaviour. When facilitating the test, a human evaluator judges conversations between a human and machine. If the evaluator cannot tell the difference, then the machine is said to have passed the Turing Test. The term is often used to express the human-like qualities of AI.
Uncanny Valley - A term coined in 1970 was originally referring to the uncomfortable and creepy feeling we feel towards robots who have become very human-like. The valley as shown on a graph is the point where our infinity towards the robot is lost and turns into discomfort. The term is more recently used to describe the oddities and imperfections in AI generated content, such as the way someone’s mouth moves in a deepfake video.
Please also see: Glossary of AI terms document