How apps like ChatGPT actually work (and how we got here)

You might think of artificial intelligence mainly as a new way to draft short emails or outline reports in seconds. If you do, you’re in good company.

To most people outside the scientific community, it probably feels like AI was invented overnight when ChatGPT burst onto the scene in 2022.

But AI isn’t new. What’s changed more recently is that generative AI – tools that create not only text but images, video and music – has become widely accessible thanks to chatbots like OpenAI’s ChatGPT, Google’s Gemini and Microsoft’s Copilot. Because it’s so new to most people, we haven’t developed a shared language to talk about it yet.

So, while there are different kinds of AI, for the purposes of this article my focus will be on generative AI – the type that powers things like ChatGPT.

How we got here

The launch of ChatGPT just a few years ago may have been many people’s first experience of AI, but its origins go back much further. So here’s a whistle-stop history of generative AI, from its roots to its current incarnation:

• 1950s–60s: Foundations of AI
  Alan Turing’s groundbreaking work in logic and computing – including the invention of the Turing Machine, used during World War II to help crack Germany’s Enigma code – laid essential groundwork for modern AI. In 1956, the term ‘artificial intelligence’ was first coined at the Dartmouth Conference, marking the start of efforts to replicate human reasoning.
• 1980s–90s: Emergence of machine learning
  AI research took a big step forward with the rise of machine learning, where computers began to learn from examples rather than needing every action programmed step by step. Instead of being told exactly what to do, machines could now recognise patterns from data and get better at tasks through practice – much like humans learn from experience. This allowed AI to handle more complex tasks, such as understanding speech or identifying images.
• 2000s: The big data boom
  The explosion of internet usage and digital data significantly accelerated AI advancements, powering tools like recommendation algorithms, search engines and voice assistants.
• 2017: The transformer revolution
  In 2017, Google scientists developed the transformer, a powerful new neural network inspired by how the human brain works. Neural networks learn by spotting patterns in data – again, similar to humans learning through experience. The transformer changed the way AI understands language, laying the groundwork for tools like ChatGPT.
• 2020–present: The era of generative AI
  In 2020, OpenAI launched GPT-3, an advanced AI model capable of generating human-like text. Initially, it was mainly accessible to developers and tech companies, but its remarkable ability to produce realistic articles, emails, stories and even code quickly caught public attention.It wasn’t until the public launch of ChatGPT – based on GPT-3 – in late 2022 that generative AI became mainstream. Millions of people around the world could now easily interact with AI through a simple chat interface, transforming tasks from writing emails to brainstorming ideas. Rapid advancements followed, leading to even more powerful and versatile models such as GPT-4, Claude and Google’s Gemini. This saw generative AI being increasingly embedded into everyday life.

How generative AI works

To really get the most out of AI, it helps to understand what’s happening behind the scenes (roughly!). When you type something into ChatGPT, what’s actually going on?

Generative AI models, like ChatGPT, are powered by something called a transformer: a sophisticated neural network architecture developed by Google researchers in 2017. Transformers process massive amounts of text, learning patterns and relationships within language.

But what makes this technology fascinating is that it doesn’t operate linearly, word by word. Instead, it considers your entire prompt at once to predict the most likely next words – rapidly generating responses that seem remarkably coherent and contextually relevant.

Generative AI is essentially a prediction engine, trained on vast amounts of data collected from sources like Wikipedia, Reddit, news articles, books, academic papers and online forums. It predicts responses based on patterns identified in this data, making each reply statistically likely rather than genuinely understood.

That’s why the way you phrase your prompts matters. The more specific context you provide, the more relevant and accurate the AI’s response becomes, as it can better align its predictions with relevant data patterns.

This prediction process occurs inside what’s known as a large language model (LLM). Applications such as ChatGPT and Microsoft’s Copilot wrap these powerful models in simple, user-friendly interfaces – which allows for easy interactions and meaningful conversations.

So, when you’re typing a prompt into ChatGPT, or any other AI chatbot, you’re interacting with a predictive system. Knowing this helps you to make the best use of the tools and to craft clearer prompts, resulting in better, more useful responses.

Click image to enlarge in new tab

Example: How ChatGPT generates a response

Let’s break this process down with a real-world example.

✅ Step 1: You enter a prompt

• User input: ‘Tell me a joke about AI.’

✅ Step 2: The LLM processes the request

• ChatGPT doesn’t think of a joke the way a human does. Instead, it scans its vast dataset, finds patterns in how AI jokes are typically structured and predicts the most likely response.

✅ Step 3: The application generates an output

• ChatGPT response: ‘Why did the AI break up with its partner? It just felt like they weren’t on the same wavelength!’
• This joke isn’t ‘created’ from scratch – it’s an assembly of words that the model has statistically determined to be the most fitting based on patterns in its training data.

The better you understand this, the better your prompts – and results – will be.

Six key uses

Generative AI is reshaping industries and workflows. But how, exactly? Renowned AI expert Daniel Hulme says that it can be broken down into six key uses:

1. 1. Task automation: automating repetitive tasks like data entry, RPA (robotic process automation), chatbots and object recognition.

1. 1. Content generation: creating text, images, video and music – from email, documents and marketing copy to deepfake videos.

1. 1. Human representation: AI-generated personas, deepfake videos and voice cloning.

1. 1. Insight extraction: using machine learning and analytics to generate insights from vast amounts of data.

1. 1. Complex decision making: AI-driven optimisation, decision trees and expert systems that assist with strategic choices.

1. Human augmentation: AI-powered exoskeletons, avatars and cybernetics designed to enhance human capabilities.

What happens next?

Every major technological shift follows a pattern. AI is no different. We can expect its impact to evolve through three broad phases, framed by the Now-Next-Future model:

1. 1. Doing things better (Now): AI is helping us complete existing tasks faster and more efficiently. Think AI-powered email drafting, automated customer service and instant image generation.

1. 1. Thinking of new ways (Next): businesses will start using AI in new, innovative ways. Think AI-enhanced creativity, AI-generated insights and workflows redesigned around AI’s capabilities.

1. Doing new things (Future): entirely new products and industries will emerge, just as the internet led to companies like Google, Amazon and Uber.

Right now, we’re in phase one. AI is helping us work faster. But the real transformation is still to come.

Generative AI isn’t a magic solution. But it can be used to accelerate our thought processes, which is pretty magical. When we use it not to outsource our thinking but to speed up mundane repetitive tasks, it can leave more time for creativity and free thought. It can be a ‘cognitive accelerator’ that helps us think, create and work in new ways.

As individuals, it’s important to understand how AI works so we can understand not only its potential, but its limitations too. If we keep this in mind, we can use AI to be smarter, not dumb us down.

Discover how to use AI to level up your professional writing with this short, intensive masterclass: Superhuman writing with AI.

Image credit: Besjunior / Shutterstock