All these terms are thrown around in conversations, and sometimes you could be forgiven for thinking they mean more or less the same thing.
But they don't.
Some describe how computers learn. Others describe how they work with language, images or information. Still others describe systems that can use these tools to carry out tasks.
So what is the difference?
Artificial Intelligence (AI) is the umbrella term. It covers computer systems that can perform tasks we have traditionally associated with human intelligence. Learning, analysing information, reasoning and solving problems, for example.
Machine Learning (ML) is one branch of artificial intelligence. Instead of programming every rule in advance, we let the computer learn patterns from data. It can, for example, learn to predict which customers are likely to pay their invoices late.
Deep Learning (DL) is a branch of machine learning. It uses neural networks with many layers that can learn complex patterns, for example in images, sound and text.
There is a hierarchy here. Deep learning is machine learning, and machine learning is artificial intelligence. But not all artificial intelligence relies on machine learning, and not all machine learning relies on deep learning.
Then there are terms that describe what we want the computer to be able to do.
Computer Vision (CV) enables computers to process images and video. For example, to detect manufacturing defects or recognise objects in photographs.
Natural Language Processing (NLP) is about working with human language. Among other things, it can be used to categorise emails, analyse text and translate between languages.
Large Language Models (LLMs) are models that have learned patterns in language from vast amounts of data. They can write text, answer questions, explain concepts and help with programming. ChatGPT and Claude are familiar examples of systems built on such models.
But here is an interesting limitation.
A language model can have extensive knowledge from its training without having access to the latest information or your company's documents.
This is where RAG (Retrieval-Augmented Generation) comes in. This approach allows the system to search specific sources for relevant information and use it when answering.
Instead of relying solely on what the model has learned, it can, for example, look through the company's contracts before answering a question about commercial terms.
Multimodal AI can work with more than one type of information, such as text, images and sound. You can show the system a photograph and ask what is in it, rather than having to describe it yourself.
And then there are AI Agents, automated AI assistants that can plan and carry out a sequence of tasks using different tools.
Such a system could, for example, find suppliers, compare quotes and prepare a purchasing proposal.
But just because a system can carry out more steps independently does not mean we should give it unlimited decision-making authority.
So where is the value created?
Imagine a company where completed work does not always make it onto an invoice.
We could train a machine learning model to detect discrepancies, use a language model to read work logs and have an AI Agent prepare reports.
But what if the problem is simply that timesheets are not being compared with issued invoices?
Then relatively simple automation might solve most of the problem without us needing to build a complex AI system.
This is one of the ideas I find most interesting about designing AI solutions.
Technology keeps opening up more possibilities. But more possibilities do not necessarily mean we need more complex solutions.
Perhaps the most important question is not what artificial intelligence can do for us.
But what we are actually trying to solve.
This article is part of my own reflections on artificial intelligence, technology and value creation. One source of inspiration is my study of Designing and Building AI Products and Services at MIT xPRO. Here I try to put what I am learning into the context of my own experience in running businesses and developing solutions.