.webp)
When something is labelled as ‘AI’ today, it can be difficult to immediately know what that means.
The core of the problem lies in the fact that the frontier of what’s considered ‘AI’ is constantly moving: as technology matures and language evolves, so does the line at which we consider something artificial intelligence. Right now, ChatGPT is the face of AI - but a few decades ago it was Deep Blue, a chess computer.
What also muddies the waters of discussion is fever-pitch hype, backed by record levels of AI investment. There is now plenty of incentive for companies to say they ‘do’ AI.
With that in mind, we wanted to share a transparent perspective on all the different flavours of AI, and how we use them: Pigment’s PoV on everything under the AI umbrella.
Peeling the onion
AI
Artificial intelligence (AI) is any technology that allows computers and machines to mimic human abilities.
Many techniques and technologies that fall under this umbrella have been at work for decades - machine learning algorithms used in credit scoring being a good example.
Machine learning
Machine learning (ML) is a subset of artificial intelligence (AI) that focuses on teaching computers and machines to replicate human learning processes, enabling them to enhance their performance over time. To do this, they use algorithms and statistical models to analyze and understand hidden patterns in datasets.
Within the context of business planning, they can be very useful for forecasting: predicting future values based on historical data and external variables.
Unlike traditional forecasting techniques, which rely on statistical models with fixed assumptions, ML forecasting uses complex algorithms capable of capturing hidden patterns, trends, and nonlinear relationships in data.
It is particularly effective in scenarios with large datasets, multiple influencing factors, and dynamic environments:
- Financial forecasting (FP&A)
Predicting revenues, expenses, cash flows, stock prices, and market trends. - Demand forecasting (supply chain)
Predicting consumer demand for products or services in industries like retail, manufacturing, and e-commerce. - Sales forecasting (SPM)
Predicting future sales based on past performance, which may take seasonality into account.
Neural networks and deep learning
A neural network comes one layer down from machine learning: it describes a machine learning model that mimics the decision-making process of the human brain by simulating how biological neurons work together to recognize patterns, evaluate options, and draw conclusions.A neural network is composed of three layers of neuron-like ‘nodes’: the input layer, the hidden layer, and the output layer.
Let’s imagine you were going to use neural networks to detect spam email. Information is fed into the input layer (an email), where the nodes in the hidden layer apply mathematical techniques to look for patterns in the data that might show the tell-tale signs of spam, and then submit an answer in the output layer.
The first time it happens, the network won’t be very good at producing a correct result. But by feeding it correct answers (confirmed examples of spam and non-spam emails), the model can ‘train’ itself on this data and adjust the way the nodes in the hidden layer work, allowing it to produce more accurate results over time. This is an ML technique known as supervised learning, because it takes place with labelled data.
Another technique, unsupervised learning, involves using a model to find patterns in the data without human input - it’s particularly good for anomaly detection, which can help find evidence of financial fraud, for example.
Deep neural networks, which are what we use to perform deep learning, refer to neural networks where the hidden layer contains three or more levels of nodes. Because it involves more ‘thinking’, deep learning can undertake significantly more complex tasks - for example, image recognition.
Generative AI and LLMs
Generative AI is a type of deep learning that focuses on creating new content, such as text, images, audio, video, or code, by learning patterns and structures from existing data - for example DALL-E 3 is the most well-known model used for image creation.
Large language models (LLMs) are a form of generative AI that focus on producing text - the most ubiquitous is OpenAI’s GPT-4o, but others include Anthrophic’s Claude, Google’s Gemini, and Chinese newcomer DeepSeek.
All these models are trained using a technique called self-supervised learning, where a model learns patterns from data without requiring labeled data. Essentially, the model ‘hides’ a piece of text from a sentence, and then tasks itself with predicting what that piece of text should be based on the context around it.
In practice, this means companies like OpenAI and Google are able to leverage massive volumes of textual data to train their models, which would be too laborious to label manually.
LLMs have been a seminal moment for the field of AI because they completely change the way people interface with technology. They enable machines to understand and interact with human beings in natural language. That’s incredibly powerful.
We use our own adapted version of GPT-4o to allow Pigment to understand user intent. That core capability powers assistant features within the platform that enable customers to instantly search through documentation, and interrogate their data in natural language.
Our models are adapted and trained using planning best practice and Pigment’s own documentation to ensure they have the context required to be helpful to our users.
Agentic AI
Agentic AI is an emerging field that leverages an array of different AI techniques to enable an ‘agent’ to tackle complex tasks autonomously - without human intervention.
To do so, it combines the reasoning power of an LLM with an array of tools it can use to perceive and interact with its environment, as well as ML techniques like reinforcement learning.
Agents represent the next real step forward in the field of AI.
To learn more, come back soon for the next post in this series, Why everyone’s talking about AI agents.
