Your quick and easy guide to artificial intelligence
Now you can sound like you really do understand AI
Impress your colleagues with fluent expositions of the fundamentals of AI in fund management, how it can help your organisation make money and how to watch out for those pitfalls
AI is full of intimidating jargon. Impenetrable labels and notations that sound important and meaningful, and yet somehow convey no meaning at all. This guide is designed to prep you to sound like an AI insider. First, some basics, and then some tight explanations of the worst terminology.
Artificial Intelligence
AI is the field of activity in which a computer (or system of computers) develops the ability to accomplish complex tasks that would generally require human intelligence. Such tasks might include decision-making, playing games, translating across languages, visual perception, speech recognition and making recommendations. AI uses data to develop algorithms (or formulas). AI learns by spotting connections in data and observing what tends to improve performance over time.
Machine learning
Machine learning is a subset of AI. It is the application of AI that provides a computer (or system of computers) with the ability to not only learn from an environment but also to apply this learning to make decisions. In practice, machine learning is the application of algorithms to describe and analyse data, spot patterns or connections in that data and then recommend action. Or, indeed, take action. Used very frequently in portfolio optimisation, risk management and asset price forecasting.
Deep learning
Deep learning is a subset of machine learning. Deep learning is the use of advanced algorithms called artificial neural networks. ANNs are designed to replicate the biological neural network of the human brain. They seek to analyse data and draw logical conclusions in a similar fashion to the human brain. It’s called ‘deep’ because ANNs can have many layers which enable them to learn very complex patterns.
Data science
Data science is the use of statistical and scientific methods to gain knowledge and insight from data sources. Data science converts raw data into useful information to help humans make decisions. Data science has different objectives to AI and machine learning, but all three share some tools and techniques. Data science is a fast-developing academic field and draws from a number of disciplines, from economics to physics.
Machine learning types
Supervised
Supervised learning uses labelled data (like time series) to train a model to explain that data and produce predictions of unseen or future data. The sample data is ‘supervised’ because its labels (like dates or names) are known.
There are two types of supervised learning. Regression quantifies the relationship between one or more predictor (explanatory) variables and an outcome which is a continuous response variable. This relationship is then used to predict unknown outcomes. The other type, classification, predicts the categorical class of new instances based on the characteristics of past observations.
Unsupervised
Unsupervised learning makes deductions and draws conclusions from data without labels. There are two types of unsupervised learning. Dimensionality reduction lowers the number of features (or variables) in a dataset without reducing either the information in the data or the model performance. Clustering uncovers hidden structures or relationships in data. Clustering looks to find natural groupings in datasets so that items in each cluster have more in common with each other than with items in other clusters.
Reinforcement learning
Reinforcement learning is a trial-and-error process of learning from experiences. An agent observes the environment, selects courses of action and receives rewards or penalties. The learning agent seeks to maximise its rewards and minimise its penalties using observations of the results of its previous courses of action. Reinforcement learning is usually applied when there is no explicit answer or solution, as there is in supervised learning.
Your AI jargon crib sheet
Agentic AI
An AI system capable of autonomously planning, deciding and acting toward a specific objective, using memory, reasoning, and tools. Often with minimal human oversight.
Algorithm
A list of instructions to follow to complete a specific task. Particularly useful when working with big data or machine learning.
Application programming interface (API)
A set of protocols that determine how two software applications interact with each other. Usually written in programming languages such as C++ and Javascript.
Cognitive computing
The same as AI. A computerized model that mimics human thought processes like understanding natural language, pattern recognition, and learning.
Computer vision
An interdisciplinary field of science and technology enabling computers to gain understanding from images and videos.
Edge AI
Processes data on local devices (like phones or sensors) instead of relying on resident PCs, systems or cloud servers.
Emergent behaviour
The unpredictable or unintended capabilities shown by an AI system, usually apparent when individual components are allowed to interact with each other.
Explainable AI
AI systems designed to make their processes, calculations, recommendations, decisions and behaviour explicit and understandable to humans.
Fine tuning
The adjustments to a pre-trained AI model using additional new, specific data, in order to improve its performance toward a particular objective.
Foundation model
A large, general-purpose AI model (like a large language model, see below) that can be adapted for many different kinds of tasks. Often built upon to tailor output for specific needs.
Generative AI
A technology that uses AI to create new content, including text, video, code and images, using large amounts of data. Like Tilly Norwood.
Guardrails
Frameworks or mechanisms to ensure that AI systems operate within agreed ethical, legal, and technical boundaries.
Hallucination
An unexpected and incorrect response from an AI system. Usually incorrect or false information presented as an outcome, result or as factual information.
Hyperparameters
Parameters (or controls) which manage the learning process. These ultimately determine the values of model parameters produced by a learning algorithm.
Inference
When an AI model or system applies what it has learned to make predictions, recommendations or decisions.
Intelligent agent
An autonomous algorithm, computer or system that can perceive its environment and take actions.
Large language model (LLM)
An AI model trained on large amounts of text to enable it to understand natural language and generate human-like text. Eg ChatGPT, Claude and Gemini.
Limited memory
An AI system receiving knowledge from real-time events and storing it in a database to make better predictions.
Natural language processing (NLP)
A form of AI enabling computers to understand spoken and written human language. NLP powers features like text and speech recognition on devices.
Overfitting
A common problem in machine learning training when the algorithm concentrates too much on specific instances within training data, leaving it unable to make accurate predictions on new data.
Pattern recognition
The use of computer algorithms to analyse, detect, and label patterns and connections in data. This guides how the data is classified into distinct categories.
Predictive analytics
Use of the technology to predict what will happen in a specific time-frame based on the patterns within historical data.
Prescriptive analytics
Use of the technology to analyse possible strategic situations and scenarios, past and present performance, and other resources to help make decisions.
Pre-training
The initial training where an intelligent agent learns patterns from data.
Quantum computing
The use of quantum-mechanical techniques like entanglement and superposition to perform complex calculations. Performs much faster than a classic machine learning program on a computer.
Sentiment analysis
The use of AI to analyse the tone and opinion of text. Also known as opinion mining.
Structured data
Data that is defined, searchable and formatted. Eg data that is organized into rows and columns.
Synthetic data
Data generated artifically to train or test AI models when real data is limited or unavailable.
Token
A basic unit of text that a large language model (see above) uses to understand and generate language. May be a whole word or part(s) of a word.
Training data
The primary information or examples fed to an AI system to enable it to learn, find patterns, and create new content.
Transfer learning
A machine learning system taking already existing, previously learned data and applying it to new tasks.
Turing test
Evaluates a computer’s ability to show intelligence equal to humans, especially in language and behavior. Usually judged by a human. Invented by computer scientist, Alan Turing.
Underfitting
A problem occuring when a model is too simple to capture underlying patterns in data, resulting in poor predictive performance.
Unstructured data
Data not organized in any apparent way. Unstructured data usually needs to be organised with some type of system before any analysis.
Voice recognition
Computers listening and interpreting human dictation (speech) to produce written or spoken outputs. Eg Apple’s Siri and Amazon’s Alexa. Also called speech recognition.