Artificial Intelligence Glossary

Algorithm: A set of instructions needed by a computer to complete a task or solve a problem.

Algorithmic bias: This is the bias that AI programs absorb via the data they are trained on. Racial, gender and cultural biases occur in the data machines learn from because biased humans create the data. Algorithmic bias can result in harmful content being produced and it’s one of the challenges facing the implementation of AI.

Artificial intelligence (AI): Machines that can replace functions ordinarily requiring human (or other biological) brain power.

Artificial general intelligence (AGI): Software that has capabilities equal to the human brain across a wide range of functions. This program doesn’t exist yet but it is often thought of as the ultimate quest of AI study.

Automatic speech recognition (ASR): Software that can understand human speech and reproduce it as text or respond to it as a command. Using AI-driven technology called natural language processing, ASR can create real-time captions but as yet isn’t always accurate.

Chatbot: An AI application that is trained to respond to prompts with conversational replies via text, speech or graphics, usually online. Chatbots simulate natural conversation and complete specific tasks.

ChatGPT: GPTs are AI language models developed by the company OpenAI and powered by natural language processing. GPT stands for ‘Generative Pretrained Transformer’ which is important because it tells us that this model has the capacity to generate or produce output and is trained on existing data. ChatGPT is an AI chatbot that responds to queries and produces human-like communication in text format.

DALL-E / DALL-E2: Generative AI models that can create photorealistic images from a natural language text description. DALL-E2 can also edit and retouch photos, as well as replace parts of photos or images with AI-generated images.

Dataset: A defined group of words or numbers that can be used by algorithms to obtain information.

Deepfake: An AI-generated image, video or audio clip that recreates a person’s image or voice and can use it to spread misinformation about that person.

Deep learning: This is the form of machine learning that uses neural networks to find patterns in data.

Generative AI: A program that analyses large amounts of data and finds its own solutions without the need for a specific code input. These types of AI typically generate text, video and other media. ChatGPT is an example of generative AI.

Hallucination: Generative AI models sometimes invent information. These inventions are called ‘hallucinations’ and often sound realistic; it is easy to be fooled by them. Scientists don’t yet fully understand why this happens.

Human-in-the-loop (HITL): This is the process of merging human expertise with machine capabilities for high-quality outcomes. For example, in the language industry human specialists edit machine translations to ensure there are no errors or inaccuracies. Humans can also intervene to evaluate AI models and give feedback for training.

Large language model (LLM): A type of artificial intelligence that is trained on huge amounts of textual data which enables it to track the patterns and links between words and phrases. It then uses this knowledge to create output that closely resembles human language.

Machine learning: A field of artificial intelligence where a computer algorithm looks for patterns in the data it is given and subsequently uses new data to complete a requested task. The program ‘learns’ from the information it receives without the need for other instructions.

Natural language processing: An area of artificial intelligence where algorithms use data to search for and process linguistic connections and sequences between the words, phrases and paragraphs of human language both in written and spoken form.

Neural machine translation (NMT): A machine translation system that leverages neural networks for more speed and accuracy. NMT is best suited to repetitive, high-volume work or where the purpose is to provide gist only. For fully accurate, error-free translations human intervention is needed.

Neural network: A type of algorithm that is able to analyse and understand relationships within the data it’s given. They are modelled on the way neurons function in the human brain. Neural networks are used in machine learning algorithms and can find solutions without a preprogramed or expected outcome.

Open source: Means that the software is freely available to use and edit. This encourages collaboration, glitch fixes and the development of technologies for wider benefit.

OpenAI: This is an artificial intelligence research and development laboratory that has as its mission the development of AI for the benefit of humanity. It is the creator of the GPT and DALL-E models.

Parameter: A key element in a machine learning algorithm. In very simple terms parameters are like a set of functions the AI model uses to fine-tune its capacity to produce the desired output. Generally, the greater the number of parameters the more powerful the model.

Prompt: In the context of AI a prompt is an instruction given to a chatbot or other model to get a specific response or answer. Prompt engineering is the technique of using precise and detailed instructions to encourage a very refined response.

Training data: This is the information that machine learning algorithms are taught with. They are able to analyse large amounts of data (text, numbers, images or sound) and use it to accomplish tasks. ChatGPT, for example, uses textual data that is available on the internet.

Transformer: A specific neural network model that can transform one type of input into another type of output. Google first used the term when it developed a more efficient neural net to translate English to French in 2017. Transformers are a key component of natural language processing applications like LLMs and DALL-E.

Here at t’works our job is to communicate information in a way that makes it accessible to the people it’s aimed at. Our objective is to eliminate any barriers to understanding and the intended audience is always our priority. That’s why we’ve created our glossaries.

In all sectors of the working world, people coin their own terminology to more easily encapsulate the specific processes, techniques, concepts and technology related to their sector. These words, phrases and abbreviations (the latter in particular!) can get in the way of proper understanding if they’re not correctly explained. The language industry is no exception.

Our t’works glossaries are designed to demystify and clarify the words we use when talking about our services and solutions. We hope you find them useful. And of course, if there’s anything else we can do to better illuminate our work, we would be more than happy to chat to you about it. Just get in touch here.