Types of AI Used by Investment Firms
Artificial Intelligence is an umbrella term that refers to machines which simulate human intelligence using systems or agents to perform tasks such as learning, reasoning, problem solving, seeking out effective solutions and decision making.
In this article we will examine three types of AI – Generative, Machine Learning and Agentic.
Generative AI is a subset of systems that can create new, original content rather than simply analysing or acting on existing data. The output can take many forms, including text, images, video or computer code, but on the whole tends not to take the form of numerical data.
Generative models are trained on large datasets representing existing examples, using pattern recognition techniques to learn the underlying structures, styles and relationships within the training data.
Using this acquired knowledge they can then generate outputs similar in style and structure to the training data although being statistically novel and original.
Outputs are often generated probabilistically, meaning there can be an element of randomness or variation, leading to different results even with similar prompts.
As with other AI systems, Generative AI can produce rogue outputs sometimes referred to as hallucinations, caused by a variety of factors including insufficient training data, incorrect model assumptions or biases in data used to train the model. A hallucination is a term coined essentially to represent where an AI system creates incorrect or misleading information but presents this as a fact.
Generative AI has become synonymous with the term AI in recent years, although it is in fact just one form of AI.
Machine Learning includes systems that learn from and make decisions or predictions based on data and algorithms, without being explicitly programmed for each specific task.
A computer program is considered to be learning from an experience in relation to a set of data and a performance metric if its ability to perform those tasks, evaluated against that performance metric, demonstrates the ability to improve as it gains more experience.
Systems can learn, thereby improving their performance on a task over time, or adapting to changing conditions over time. This happens as they are exposed to more information and by further analysing data to find patterns and relationships.
The output is typically quantitative, and often involves a prediction, for example it will rain today, or a decision such as steering the car to the left.
The system is not explicitly told how to find a solution, a crucial differentiator from traditional programming. In traditional programming explicit, step-by-step rules are written for the computer to follow to solve a problem. If the rules need to change, a programmer has to update the code. With Machine Learning, data is fed to an algorithm which "learns" the rules or patterns from the new data. If the data changes, the system can potentially adapt without a programmer rewriting the core logic. It can do this because the algorithm has an embedded forecasting capability, and it is continuously tested throughout that learning process.
Agentic AI refers to systems designed with a degree of autonomy enabling decision making and taking actions to achieve specific goals.
These systems are characterised by two main features: goal-orientation, meaning they pursue one or more specific aims, and autonomy, enabling them to function without constant or direct human input. In essence, Agentic AI marks a shift from AI as a tool that assists humans to interface with AI as an actor that can operate more independently to achieve objectives.
These systems can assess situations, consider options and choose a course of action based on pre-determined goals and an understanding of the environment. The process involves planning, reasoning, problem-solving, and decision-making.
Systems can execute actions such as send an email, make a financial trade or generate code, and can learn from their experiences adapting their behaviour over time to improve performance or handle new situations.
