It’s a Wednesday afternoon in the 90s. You’re a teenager, staring at a blank sheet of paper, struggling with your English literature essay on Romeo and Juliet. Tomorrow’s deadline is looming, and inspiration feels out of reach.

If this scene had unfolded in your 90s bedroom, you would have dismissed it as madness - something out of a sci-fi movie or an overactive imagination. But today, generative AI has turned this once-unthinkable scenario into an everyday reality. Writing essays, creating art, composing songs, tasks we once believed to be uniquely human, are now effortlessly achieved by machines, leaving us to marvel at just how far we have come.

But how does this futuristic ‘magic wand’ work, and what makes generative AI so extraordinary? In AI: What is it really?, I laid the groundwork for understanding how AI mimics human intelligence. In this article, we will take a closer look at generative AI.

What is Generative AI?

Generative AI is a type of artificial intelligence that creates new content such as text, images, music, or even ideas. Unlike traditional AI, which follows rules to provide answers, generative AI leverages what it has learned from large datasets and uses it to produce something following a request from a user.

For example, you can ask it to write a witty best man speech, design a dream vacation itinerary, or create a recipe for a pasta dish inspired by Greek flavours. Tools like ChatGPT, DALL·E, MidJourney, Claude, and Gemini have proven how powerful this technology has become, turning prompts into stories, art, and even creative solutions almost instantly.

How Does Generative AI Work?

Generative AI might feel like a groundbreaking innovation of the 2020s, but the truth is, it has been decades in the making. Researchers like Geoffrey Hinton and John Hopfield laid the foundation of generative AI as early as the 1980s, developing neural networks that mimic how our brains process information. Their pioneering work, which introduced the idea that machines could learn and adapt like humans, earned them the Nobel Prize in 2024 - a well-deserved recognition of their vision.

But while the concepts were in place, the technology was not ready yet. Training generative AI models requires immense amounts of data and computing power, which only became available in recent years. That is why generative AI did not truly emerge until 2022, when advancements in hardware and algorithms made tools like ChatGPT, Gemini and Claude possible.

How It Works: A Story of Learning

Imagine you want to teach someone to write, but instead of giving them a lesson, you hand them the world’s largest library. Every book, article, poem, and essay is at their fingertips. This person starts by reading everything, not to memorise it word for word, but with the aim of understanding common patterns: how sentences flow, how ideas connect, how stories are told.

This is essentially what happens when training a generative AI model. The model is fed enormous datasets which include trillions of words from books, websites, and articles. It learns not by copying but by recognising relationships: which words more often follow others, how grammar works, and even the nuances of style. Over time, it becomes capable of creating its own sentences, mimicking the way humans write, but with its own unique twist.

The Technical Side: Predicting the Next Word

At its core, generative AI works by predicting probabilities. For instance, if you type “Once upon a” the model calculates the most likely next word based on the patterns it has learned. It considers probabilities - “time” has a 75% likelihood of following “Once upon a”, while “dream” has a 10% likelihood, and so on. The model chooses the option with the highest probability but can also sometimes be creative and choose a less predictable option.

This process relies heavily on backpropagation, a technique pioneered by Geoffrey Hinton. Backpropagation helps the model learn from its mistakes during its training. When the model chooses a wrong word or sequence, backpropagation adjusts its internal calculations - such as weights and biases - enabling it to make a better prediction next time. Think of it as a self-improvement loop, which refines the model's ability to generate content with every iteration.

This ability to learn, adapt, and generate has unlocked endless possibilities for how we can use generative AI in our everyday life. From creating art to solving complex problems, it is reshaping industries and redefining creativity. Its impact can already be seen across a wide range of fields, from entertainment to education to business as well as healthcare.

Real-World Applications of Generative AI

Imagine an ad so captivating that people rush to buy the products advertised only to discover that these do not exist. That is exactly what happened with a speculative Adidas campaign created entirely using generative AI. The ad, made without a single product or traditional production crew, drastically cut costs while testing customer interest in a floral-themed collection.

This move demonstrates the incredible potential of generative AI. Brands can create stunning content at a fraction of the usual cost, test demand, and refine ideas, all before investing in production. It is not only about saving money; it is about opening new ways to create, innovate, connect, and solve problems.

The impact of the Adidas campaign is just one example of how generative AI is revolutionising industries. Let's look at some other examples:

You can learn more about the specific use cases and opportunities for using generative AI in accounting in ICAEW’s guidance.

How to Use Generative AI

Using generative AI is like speaking to a genie but one that takes your words very literally. The magic lies not just in what the AI can do but in how you ask it to do it. This is where prompt engineering comes into play.

For example, if you ask, “Tell me about cities,” you will probably get a generic answer. But if you say, “Describe the energy of London at night, focusing on its lights and sounds,” the response transforms into something vivid and unique. The result is only as good as the prompt you provide, and crafting those prompts is an art in itself.

So, what makes a good prompt? While there are numerous guides available, these five principles form the foundation of a good prompt:

By following these steps, you can guide generative AI to produce clear, useful, relevant, and creative outputs.

We cover more on how to master the art of AI prompt engineering in this on-demand webinar.

The Challenges of Generative AI

Generative AI feels like a magic wand, but even magic comes with its challenges. While it sparks creativity and efficiency, it also carries risks that demand careful consideration.

As we have seen, the results of generative AI models depend on the prompts we provide. But what happens when such models are trained on data that is biased? Since generative AI learns from existing data, it can mirror and even amplify societal prejudices, leading to outputs that may be unfair or harmful. Then there is also the rise of deepfakes and misinformation - content so convincing it can fool the public and blur the line between fact and fiction.

Ownership and accountability are also unclear. Who owns an AI-generated masterpiece? What happens when generative AI accidentally borrows too much from copyrighted works? These are not just technical concerns; these are ethical dilemmas we cannot ignore.

In AI Ethics: What are they really?, I explored how we can navigate these ethical challenges, from ensuring fairness to holding AI systems accountable. Generative AI magnifies these questions, making the conversation even more urgent.

Finally, let’s not forget the environmental cost. Training large AI models requires enormous computing power as they consume significant energy. As we celebrate generative AI’s capabilities, we also need to ensure its development does not harm the planet.

Generative AI is a remarkable tool, but its power is a double-edged sword. How we wield it will define whether it becomes a force for good or something far more complicated.

A summary of the risks and limitations to consider can be found in ICAEW’s generative AI guide.

The Future of Generative AI

While generative AI’s power comes with challenges, its potential to transform the world is equally extraordinary. This technology is just getting started, and its future is brimming with possibilities.

Imagine AI systems crafting personalised medicine plans tailored to an individual’s DNA or creating real-time climate models to help combat global warming. Picture immersive experiences where virtual worlds evolve dynamically, responding to human emotions and creativity. Generative AI has the potential to unlock solutions to some of the world’s most pressing challenges, pushing the boundaries of what we thought possible.

But the most exciting aspect of generative AI is not about replacing human ingenuity - it is about enhancing it. In the future, generative AI will work as a collaborator, empowering artists, scientists, educators, and professionals to achieve more than ever before.

Of course, this future depends on how we choose to use it. Balancing innovation with responsibility is key. By establishing ethical guidelines, promoting fairness, and investing in sustainable practices, we can ensure generative AI becomes a force for good, benefiting humanity and the planet.

Back in the 90s, the idea of a wand that could write your essay, compose a song, or create anything you imagined felt like pure magic. Today, that magic is real, but it is in our hands to shape how we use it. Generative AI is not just a tool; it is an opportunity - a chance to redefine creativity, solve global problems, and build a future that once seemed like science fiction.

The question is no longer whether generative AI can create, but what we will create with it and what kind of world we want it to help us build.