What is AI?
Artificial intelligence sounds complicated. The basic idea isn't.
Artificial intelligence, or AI, is technology that allows computers to perform tasks that normally require some form of human intelligence.
That might mean understanding a question, recognising a face in a photograph, translating a sentence, recommending a film, writing an email or spotting something unusual in a medical scan.
You've probably been using AI for years without particularly thinking about it.
When your phone recognises your face, your email filters out spam, Netflix recommends something to watch or Google Maps predicts traffic, AI may be involved.
What's changed recently is that AI has become much more visible, and considerably more capable.
Tools such as ChatGPT can now have conversations, answer questions, write documents, analyse information, create images and even help write computer programs.
So how does any of this actually work?
AI isn't a digital brain
This is probably the most important thing to understand about AI.
AI doesn't think in the way you do.
It doesn't have a little electronic brain inside it contemplating your question and deciding what it thinks the answer should be.
Instead, AI systems use mathematical models that have learnt patterns and relationships from data.
Imagine showing a computer thousands of photographs labelled cat and dog. During training, the system gradually learns mathematical patterns that help distinguish photographs of cats from photographs of dogs.
Eventually, you can give it a photograph it has never seen before and ask: Cat or dog?
It can calculate which answer is more likely.
Modern AI works on the same broad principle, although today's largest systems are vastly more sophisticated.
Ready to go deeper? Once you've grasped the basics, explore our technical breakdown to understand the intricate details of how modern AI systems work.
So what does "artificial intelligence" actually mean?
AI is really an umbrella term.
It covers many different technologies designed to perform tasks that we associate with intelligence.
These include recognising objects in photographs, understanding speech, translating languages, detecting fraud, recommending products, driving vehicles and generating text, images, music and video.
Not all AI is like ChatGPT.
The software detecting a suspicious transaction on your bank card and an AI chatbot answering questions are both examples of artificial intelligence, but they're doing very different things.
Machine learning
One particularly important part of AI is machine learning.
Traditional computer programs are generally built from explicit instructions written by programmers. You might tell a program: If X happens, do Y.
Machine learning takes a different approach.
Instead of programmers having to describe every rule, the computer is given examples and learns patterns from them.
For instance, rather than trying to write thousands of rules describing exactly what makes an email spam, a machine-learning system can be trained using large numbers of examples.
It learns which patterns tend to be associated with spam and can then use those patterns to assess new messages.
What is generative AI?
This is the type of AI responsible for much of the current excitement.
Generative AI creates new content. Depending on the system, that could be:
- text
- images
- computer code
- music
- speech
- video
ChatGPT, for example, is a form of generative AI designed to work primarily with language and other information.
Give an AI image generator a description such as "a small cottage beside a lake on an autumn morning" and it can generate an image that matches that description.
That ability to create rather than simply classify information is one reason AI suddenly feels very different.
How can ChatGPT have a conversation?
Systems such as ChatGPT are based on large language models, usually shortened to LLMs.
An LLM has been trained using enormous quantities of text and other data.
During that training it learns statistical relationships within language.
Very simply, it gets extremely good at answering this question: Given everything that has appeared so far, what should come next?
Suppose I write: The capital of France is…
"Paris" is an extremely likely continuation.
But modern language models have learnt relationships operating at much more sophisticated levels than individual words. Their internal representations can capture patterns involving grammar, concepts, programming structures, styles, relationships between ideas and much more.
When you ask an AI a question, it uses those learnt relationships to construct a response.
And it does this piece by piece.
AI doesn't actually work with words
There's another interesting detail hiding underneath all this.
Language models don't see sentences quite as we do.
Your question is broken into small pieces called tokens. A token might represent a whole word, part of a word, punctuation or another fragment.
Those tokens are represented mathematically.
Inside an AI model are enormous collections of numerical values called parameters. The values learnt during training are often referred to as weights.
A modern AI model can contain billions of them.
Together, those numbers encode an extraordinarily complicated web of learnt relationships.
You don't need to understand the mathematics to use AI, but this explains something important: Underneath the remarkably human-looking conversation is mathematics, not a person thinking. If you're curious about how these parameters are created during training, "How Does AI Actually Work?" explains the training process in detail.
Does AI understand what it's saying?
This is where things become more complicated.
AI can demonstrate remarkably sophisticated behaviour. Modern systems can solve problems, analyse documents, explain concepts, write software and connect information across different subjects.
Whether that constitutes "understanding" in the same sense that humans understand something is a much deeper question. Our exploration of whether AI has feelings dives deeper into this question of understanding versus pattern-matching.
For everyday use, there's a more useful distinction to remember: Sounding as though it understands something doesn't guarantee that its answer is correct.
Language models are designed to generate plausible responses. Plausible and true are not necessarily the same thing.
That's why AI can sometimes confidently give you information that is completely wrong.
These incorrect invented answers are commonly called hallucinations.
Why does AI make things up?
Imagine asking someone to complete a sentence when they're not allowed to say, "I don't know."
They might produce something that sounds convincing but isn't true.
That's not precisely how an LLM works, but it's a useful way of thinking about one of its weaknesses. For a complete discussion of this phenomenon—including practical strategies for handling it—why AI sometimes gets things wrong covers this in depth.
An AI generates its answer based on patterns and information it has learnt, the instructions it has been given and, in some systems, information it can retrieve using additional tools. Want to understand this process in more detail? See "How Does AI Actually Work?" for a deeper dive into training and pattern recognition.
It isn't automatically checking every sentence against a master database of facts.
This is why you should be particularly cautious when using AI for important factual information.
Names, dates, statistics, quotations, legal information and references are all things worth verifying.
Is AI intelligent?
It depends what you mean by intelligent.
AI can already perform some intellectual tasks extraordinarily well and outperform humans in certain narrow areas.
But human intelligence encompasses much more than answering questions or recognising patterns.
People experience the world. We develop relationships, motivations, values and memories. We understand things within the context of living a life.
Today's AI systems are fundamentally different.
Calling them "intelligent" can therefore be useful when describing what they can do, but it can become misleading if we assume that means they think and experience the world as people do.
Is AI the same as a search engine?
No.
This is another important distinction.
A traditional search engine looks through an index of information and tries to find web pages relevant to your search.
A generative AI model generates a response.
Some modern AI services blur that distinction because they can also search the web. The AI might search for information, read the results and then use its language capabilities to produce an answer.
But the underlying concepts remain different.
This matters because you shouldn't assume an AI response came directly from a reliable source simply because it sounds authoritative.
Does AI know everything on the internet?
No.
An AI model isn't simply a compressed copy of the internet, nor does it necessarily have live access to it.
What an AI can know depends on several things, including the data used to train it, when that training happened and whether the particular service has access to additional tools such as web search.
Ask an AI about something that happened five minutes ago and it may know nothing about it unless it can retrieve current information.
Different AI services also have different capabilities.
Is AI always learning from what I type?
Not necessarily.
This is one of the most common misconceptions about AI.
Your conversation doesn't normally rewrite the billions of parameters inside the underlying model as you type.
However, what happens to the information you enter is a separate question.
An AI provider may retain conversations, process them for particular purposes or potentially use certain data to improve its services, depending on the product, account type, settings and provider's policies.
That makes this a useful rule: Before giving an AI sensitive information, understand where that information is going and how the service says it will be used.
Don't assume that because you're having what feels like a private conversation, the technology works like a private conversation with another person.
Should I be worried about AI?
You don't need to choose between believing AI is wonderful and believing it's terrifying.
It's a technology with considerable capabilities, genuine limitations and real risks.
For most people, the sensible approach is to understand enough about it to make informed decisions.
That includes knowing that AI can be wrong, checking important information, being careful with personal or confidential data and thinking before allowing an AI system to take actions on your behalf.
The more capable these systems become, the more important those habits become.
The simplest way to think about AI
If you remember nothing else from this guide, remember these four things:
- AI learns patterns from data.
- It uses those patterns to perform tasks or generate new information.
- What looks remarkably intelligent is ultimately produced by computation, not a human mind inside the machine.
- And an AI sounding confident doesn't mean it's correct.
You don't need to understand billions of parameters, neural-network architecture or advanced mathematics to start using AI.
You just need to understand what the tool is good at, where its limitations are, and how to use it safely.
And that's really the idea behind ABitOfAi.
Clear. Practical. Safe.
