What even is AI anyway?

You see it everywhere. AI in your washing machine. AI in your toothbrush. There’s even AI in the cat’s litter tray. Every new gadget seems to be ‘AI-powered’. Every website, app and phone has a shiny new AI assistant. Presidents and prime ministers are talking about ChatGPT on the news. Lamp posts and shop windows are plastered with AI-generated posters and social media feeds are overflowing with AI-slop videos of cats cooking pancakes. Suddenly, everything is AI.

Is it really ‘Artificial Intelligence’? Why is it everywhere now? Should we even care?

I think we should care, at least about some of it. You’ve probably already figured out that AI is just the buzzword of today. Companies slap an ‘AI’ label on it and hope we’re impressed. But behind the hype is a genuinely transformative technology that will probably change your life, whether you want it to, or not.

While AI feels like it has only been around for a few years, its development has taken decades to reach the point we are at now.

Let’s start at the beginning.

The foundations began in the 1950s with Alan Turing, who was the first person credited with asking the deceptively simple question “Can machines think?”. Throughout the 50s and 60s researchers explored the possibilities of computer intelligence, building programs that solved logical problems, played games and built rudimentary neural networks that could learn from examples (put a pin in this, it’s going to be important later!).

In 1966 a researcher at MIT built a program called ELIZA which is considered to be the first chatbot, and was able to convince many people they were actually talking to a human during a test devised by Alan Turing, later known as the Turing Test.

Between the 1970s and 1990s, progress wasn’t smooth. There were periods of optimism followed by disappointment when the technology failed to live up to expectations. Funding dried up and interest declined during periods that became known as the AI winters. Nevertheless, research continued and incremental improvements were made.

In 1997, IBM’s Deep Blue beat the reigning world chess champion Garry Kasparov. Deep Blue was a famous demonstration of computers outperforming humans in a task once considered a test of intelligence. 

There were many developments during the 2000s and 2010s in the field of machine learning. A pivotal moment was in 2017 when a team of scientists and engineers at Google developed a new architecture for machine learning known as the ‘transformer’ (this is the T in ChatGPT). This laid the foundations for the AI tools of today.

From 2018 to 2022 leaps and bounds were made in the development of language models using the new transformer architecture and image recognition and generation models. Up to this point, these tools were largely restricted to the realm of researchers and academics until, in November 2022, OpenAI released ChatGPT.

ChatGPT made interacting with a powerful AI model accessible to anybody with an internet connection. For those who are not familiar, ChatGPT is an example of a Large Language Model (LLM). It is an interactive AI chatbot that you can talk with in the same way you would with a friend, using text, or nowadays, even using your voice. Where this gets interesting though is, imagine that your friend had read every book, seen every website, every photo and every video on the internet and could recall that information in seconds and talk fluently about any subject you can think of. It is pretty impressive. (There is a little hyperbole here, and there are some major caveats we’ll get into later).

Since the release of ChatGPT by OpenAI, every major tech company has released their own AI products. Google has Gemini, Microsoft has Copilot, X (Twitter) has Grok, Meta (Facebook) has Muse, Anthropic has Claude. The biggest and best models are known as frontier models. Fundamentally, all these AI models work in the same way. They have been trained on enormous quantities of books, news articles, social posts, images and videos scooped up from the internet and distilled into the models. 

Exactly how the training works is beyond the scope of this article (and probably beyond the scope of comprehension for most mere mortals) but simply put, for a language model, the training exercise is: predict what comes next. Give it “The cat sat on the…” and ask it to predict the next word. Adjusting the model whenever its prediction is poor. Repeat this not thousands, but billions or trillions of times and it gradually learns extraordinarily complicated patterns in language.

Image recognition works in a similar way. Show a model millions of labelled images – cats, dogs, cars, trees – and train it to recognise the patterns that distinguish them. Generative image models take this idea much further: instead of simply recognising the patterns that make up an image, they learn enough about those patterns to create entirely new images.

Inside AI models is a neural network (I told you this would be important!). The design of neural networks was inspired by neurons in a brain. Repetition and reinforcement improve the strength of certain pathways in the network. This is learning. The analogy isn’t perfect, but for a high-level overview, it serves the purpose.

This is the fundamental difference between a traditional computer program and modern AI models. Computer programs are sets of pre-defined instructions decided by the programmer. AI models are trained by feeding them vast amounts of data and teaching them to recognise patterns and make predictions.

So, what does this have to do with my washing machine?

Well, now we understand a little about how AI works, we can start to understand how some of this technology is creeping into our lives. At its core, modern AI learns patterns from data and uses those patterns to make predictions. So for our washing machine, this can include sensor data from the weight of the load, how dirty the water is, the amount of foam — to predict how much detergent is needed, how long to wash for or how fast to spin. The AI in this case uses the data to improve the quality and efficiency of each wash.

This is the direction we’re heading, for better or for worse. As AI models become smaller, cheaper and more capable, manufacturers are finding more places to put them. Your phone can recognise the faces in your photos, your car can identify a pedestrian in the road and your doorbell can tell the difference between a person and next door’s cat. Some of this genuinely is AI. A lot of it is just clever marketing for technology that isn’t particularly new.

So why is it everywhere now?

There isn’t one single reason. Several things came together: decades of research produced better ways of building AI models, the internet provided unimaginable quantities of data to train them on, and computing power finally became sufficient to process it all.

The culmination of this innovation is generative AI: models that don’t just recognise and classify information, but can generate something new, like ChatGPT. Give them a prompt and they can write an email, create an image, compose music, generate computer code, hold a conversation, or even write an article named “What even is AI anyway?”.

And now that big tech companies have a taste for what generative AI is capable of, they have doubled down, committing trillions in investment for new data centres and infrastructure needed to build and run the increasingly powerful AI models.

This AI sounds great! Doesn’t it?

Well… It certainly is impressive, but it is not without its faults. If you’ve ever interacted with ChatGPT for any length of time you will notice how often it will confidently tell you things that are simply not true. These are commonly known as hallucinations. Remember, models are trained to predict plausible answers, not necessarily factual ones. It doesn’t inherently know the difference between something that sounds right and something that is right.

Then there’s the awkward question of where all that training data came from. AI models have been trained on enormous quantities of material created by other people: books, journalism, artwork, photographs, music and websites. Much of it wasn’t specifically created or licensed for AI training. Whether companies should be allowed to train their models on copyrighted work without the creator’s permission or payment has become the subject of major legal battles.

AI tools are now so cheap and accessible that AI-generated content is flooding the internet. Articles, images, music and videos that once required hours of human work can now be produced in seconds and at virtually no cost. Inevitably, a lot of it isn’t very good. The barrier to creating content has almost disappeared, but unfortunately, so has the barrier to creating rubbish. This is what we now call AI-slop (remember those cats cooking pancakes?).

Most concerning is how it is becoming increasingly difficult to tell what is real. Photographs can be fabricated, voices cloned and convincing videos generated of people saying things they never said. As the technology has improved, “I’ll believe it when I see it” is no longer good enough. Worse still, when anything can be faked, real photographs, recordings and videos can also be dismissed as AI-generated.

Is this the end of the world as we know it?

Well, I hope not. While it is inevitable that some people will lose their jobs, this is not a new phenomenon. Farm labourers were replaced by tractors. Switchboard operators were replaced by automatic telephone exchanges. Bank tellers were replaced by ATMs. Every major technological revolution has made some jobs obsolete, changed others and created entirely new ones.

What makes AI different is the kind of work it can automate. Previous waves of automation were particularly good at replacing repetitive physical or administrative tasks. AI can increasingly perform work we once assumed required a human mind: writing, translating, illustrating, programming, researching and analysing information.

That doesn’t necessarily mean that accountants, programmers or graphic designers disappear. Jobs are made up of many different tasks, and it is more likely that AI will automate parts of those jobs rather than replace them entirely. A programmer who once spent a day writing a piece of code might produce it in an hour with AI. A lawyer might review a hundred-page document in minutes, not hours. A small business might produce its own marketing material instead of paying somebody else to do it.

History gives us some reasons to be optimistic. New technologies have repeatedly destroyed jobs while creating industries and occupations that previously didn’t exist. A hundred years ago there were no software developers, social-media managers or cybersecurity analysts. It is entirely likely that AI will create jobs we haven’t even thought of yet.

That doesn’t mean the transition will be painless. If one person with AI can do the work that previously required five, there is an uncomfortable question about what happens to the other four. Nobody really knows how many jobs AI will replace, how many it will create, or what those new jobs might look like. Anyone who tells you they know exactly what happens next is probably guessing.

So, should we care?

Yes. But probably not because your washing machine has an ‘AI’ button.

We know a lot of what is being sold to us as AI is marketing. Some of it is technology we’ve had in one form or another for years, now sporting a shiny new label.

We’ve learned that behind all that hype is something genuinely significant and how these AI models can learn extraordinarily complicated patterns from enormous amounts of data and use what they’ve learned to produce something new.

The results are simultaneously impressive and imperfect. AI can write and draw, translate and analyse, program and design, but it can also confidently mislead us and be used equally but those who wish to cause harm. It can make people hugely more productive, while making some jobs redundant. It can democratise creative tools, while making it increasingly difficult to tell whether the things we see and hear are real.

And we’re still relatively early in figuring out how to deal with it.

So the next time you see AI-powered plastered across the front of a washing machine, toothbrush or cat litter tray, a little bit of scepticism is probably justified. Not everything with AI written on it is revolutionary.

But AI itself might just be.