Artificial Intelligence: A Glossary Of Key Terms

Artificial intelligence is a complex and intricate field that can often be difficult to grasp. People who work in AI often use special terms to explain their research. This makes it hard for others to understand.

In order to help, we’ve compiled a glossary with definitions of some of the most commonly used terms in the AI industry.

AI Agent

An AI agent is a system that utilises artificial intelligence technologies to handle a variety of tasks on your behalf, tasks that go beyond what basic chatbots can do. These tasks may include actions like filing expenses, reserving tickets or restaurant tables, or even managing code.

While the infrastructure to fully support AI agents is still under development, the general concept is that they are autonomous systems capable of using multiple AI models to complete multi-step tasks. However, the exact definition can vary depending on the context.

In the AI realm, chain of thought reasoning involves breaking down a problem into smaller, manageable steps. This approach, typically used with large language models (LLMs), results in more accurate answers, especially in logical or programming contexts.

Although it can take longer to reach a solution, the quality of the outcome is generally improved. Chain-of-thought reasoning is an advancement of traditional LLMs, optimised through reinforcement learning.

Deep Learning

Deep learning is a subset of machine learning that uses algorithms structured as multi-layered artificial neural networks (ANNs). These networks allow AI systems to find complex patterns in data, far beyond the capabilities of simpler models like decision trees or linear regression.

Inspired by the brain’s network of neurons, deep learning systems learn from vast amounts of data, improving their performance over time through repetition and error correction.

While deep learning models can automatically identify key data features without human intervention, they require enormous datasets, often millions of data points, to produce reliable results. Additionally, training deep learning models tends to be more time-consuming and costly compared to simpler machine learning algorithms.

Fine-Tuning

Fine-tuning refers to the process of further training an AI model to enhance its performance for a more specific task or domain. This often involves feeding in new, task-specific data to adjust the model’s output. AI companies frequently use fine-tuning to refine large language models for particular industries or use cases, providing customised solutions based on specialised knowledge.

When a user prompts the model, it predicts the most likely next word based on context, generating human-like responses. While AI assistants may have different names, they all rely on the same underlying LLM technology.

Neural Network

A neural network is the foundational architecture behind deep learning and is responsible for driving the progress of generative AI tools.

The concept of neural networks, inspired by the structure of the human brain, dates back to the 1940s, but it was the advent of powerful graphical processing units (GPUs) that allowed for the development of multi-layered neural networks.

These networks are capable of processing vast amounts of data and delivering impressive results in areas like speech recognition, autonomous vehicles, and drug discovery.

Weights

Weights are numerical values that help determine the significance of various features in the data used to train an AI model. These values adjust during the training process to shape the model’s output.

At the start of training, weights are randomly assigned, but as the model progresses, they are fine-tuned to more accurately reflect the importance of each feature.

For example, in a real estate price prediction model, weights would be assigned to factors like the number of bedrooms, the location of the property, and whether it has a garage. These weights ultimately determine how much each factor influences the predicted price, based on the training data.