Impact of AI

Artificial intelligence (AI) is software that carries out tasks which normally need human intelligence: recognising objects in a picture or words in speech, processing language, making predictions, and making decisions. AI is software only, so it is not a robot and it is not conscious, even when it replies in a friendly, human-sounding way.

You probably meet AI several times a day without noticing. The table shows the main everyday kinds.

One request can chain several of these together. A pupil on a school trip photographs a poster in another language: the phone's AI lifts the words out of the picture, translates them and can read the result aloud. Each stage is a separate job that would once have needed a person who could read that language and speak it.

These jobs suit AI because they are very hard to write as fixed rules. You cannot list simple instructions that spot any cat in any photo, or make sense of any accent, so AI is used wherever the input is messy real-world material such as pictures, sound and language.

Artificial intelligence is software that carries out tasks which normally need human intelligence, such as recognising a face or spotting an unusual pattern. It can work through far more information than a person could, and the answer it gives is usually a suggestion that a person then acts on.

Recognising and diagnosing

Facial recognition measures a face — the distance between the eyes, the shape of the jaw — and compares it with faces it has already seen. Airport e-gates use it to match you to the photo in your passport, so passengers pass through in seconds instead of queuing at a desk, and some banking apps use it to confirm it is really you before a payment.

In medical diagnosis, a system compares a patient's symptoms, test results and history with thousands of earlier cases and suggests conditions the doctor may not have considered. It gives a second opinion; the doctor still examines the patient and decides. For drug development, AI predicts which chemical compounds are likely to work against a disease, so scientists test the most promising ones first instead of trying hundreds at random, which saves years of work.

Driving, checking and hiring

Self-driving cars use sensors to steer, brake and follow a route without a driver. Driver-assistance systems in ordinary cars brake for you if you do not react, keep you in lane and hold a safe gap in traffic. A mistake here can injure someone, so the driver must stay alert and ready to take over.

Fraud detection software learns what a normal insurance claim or tax return looks like, then flags the few that do not fit so staff investigate those rather than reading every one. In recruitment, AI can rank CVs against the job description and have a shortlist ready in minutes, saving the hiring team hours, though a person should check it because a ranking is only as fair as the CVs it learned from.

Artificial intelligence (AI) is software that carries out tasks that normally need human intelligence. It does not think or understand like a person: it follows patterns in data, and it can work quickly without getting tired or bored. Used well, AI saves time, reduces effort and helps more people use a service.

Speed, accuracy and scale

AI can work through large amounts of material quickly and accurately, and it does not get tired or bored. A museum could use AI to read 30,000 handwritten visitor cards and pull out the date, rating and main comment. Staff would take weeks to do the same, and their attention would fade near the end; the AI keeps the same routine for every card. It is not perfect, so a sample still needs checking, but it can make a large job practical.

Patterns and automation

AI can find patterns in huge sets of data that people would struggle to compare. A school sports team could feed in every training session and injury logged over two years, and AI might show that players report fewer injuries when hard sessions are followed by a rest day. The coach can then plan training around that pattern.

AI can also automate repetitive jobs, such as a camera that spots bruised fruit on a packing line so it can be removed, or dangerous jobs, such as a robot crawling through a narrow drainage pipe to film cracks, so engineers do not have to enter. This frees people for work that needs judgement and keeps them away from harm.

Accessibility

AI can improve accessibility. Speech-to-text can put a speaker's words on screen as live captions; image descriptions can tell a blind user what a photo contains; and translation can turn a letter into a family's home language. These tools help people understand information and take part without always relying on someone else.

Many AI systems depend on large amounts of example data, often called training data. The quality of that data strongly affects how reliable and fair the AI's outputs are.

When the data is poor

If the example data is incomplete, the AI has not seen some situations and may handle them badly. If it is outdated, the AI can give answers that were once true but are no longer correct. If it is biased, it over-represents some people or cases and under-represents others, so its decisions can be unfair.

Bias is not always added on purpose. It can come from the examples chosen: if one group appears far more often in the data, the AI picks up that group's patterns as normal and treats other patterns as unusual.

Wrong answers given confidently

AI can produce an answer that looks clear, detailed and certain but is still wrong. It does not automatically check facts and cannot tell when its data does not cover the question. A generative AI tool asked about a town's old cinema might invent an opening date and present it neatly as fact. Treat an AI output as a first draft to check against a reliable source, not as proof.

Transparency means being able to see and understand how a decision was reached. Many AI systems are not transparent: they give an answer without being able to say which factors produced it, and even the people running the system may not be able to explain a single result. That causes problems, and AI can also be misused deliberately.

Why hidden decisions cause problems

When a decision cannot be explained, an error is hard to spot and hard to challenge. The person affected may never find out that something went wrong, and the organisation using the system cannot check that it is behaving correctly. That makes it unclear who is accountable when harm is done, because nobody can point to the step where the mistake happened. A hidden fault is also likely to repeat for everyone in the same situation before anyone notices.

For example, a council uses AI to put housing repair reports into priority order. A tenant reports a leaking roof and the system rates it low priority. Staff can only say the system decided it, and two weeks later the ceiling collapses. Because the reasoning was never visible, she could not challenge the rating and the council could not see its priorities were wrong.

Misuse of AI

AI can also be used deliberately to cause harm:

• Invented material – AI can produce realistic pictures, video clips or voice recordings of events that never happened, quickly and at almost no cost.
• Misinformation – an AI-written notice claiming a town's water supply is unsafe can be shared hundreds of times, spreading panic and tying up the water company's phone lines before anyone checks it.
• Fraud and scams – a fake online shop with AI-made photographs of trainers at half price can take payments for goods that do not exist.
• Harassment – AI can generate hundreds of insulting messages or fake one-star reviews aimed at one person or small business, far faster than a person could type them.

The harm lands on real people: the resident who is not believed, the buyer who loses money, the shop owner whose reputation suffers.

Artificial intelligence does not have its own set of laws. Laws written for everyone still apply to AI, but two things make this awkward: laws are written slowly and lag behind what AI can do, and different countries take different approaches, so a feature one country restricts may be allowed in another.

Copyright of training data and of AI output

Copyright protects creative work such as writing, pictures and music, and gives the creator control over how it is copied and used. Two questions about AI are disputed. First, whether it is lawful to train an AI on other people's work without their permission: a small furniture maker might find that its measured drawings of chair parts were used to train a design tool that now produces similar cutting plans. Second, nobody has settled who owns work an AI produces: the person who gave the instructions, the company that made the tool, or no one at all. Because the law has not caught up, such cases end up argued in court.

Existing laws still apply to AI

The Data Protection Act does not stop AI using personal data; it controls how it is used. In the same way, misuse of AI is illegal because of the harm it does, not because AI itself is banned.

AI systems improve when they have plenty of examples to learn from, so companies and public bodies have a strong reason to collect more personal data than they really need, and to keep it for longer. Details that look harmless when you hand them over — quiz scores, step counts, the time you tap a bus pass — become valuable once software can spot patterns in them. This is a privacy issue rather than a security one: nothing has been hacked, because the data was given away.

Facial recognition

Facial recognition software compares a face in a camera image with photos held in a database, so a person can be identified and their movements logged without being asked or told. A shop, a stadium or a school gate can run it on everyone who walks past. You cannot change your face like a password, so once it is in a database it can be matched again and again, long after you agreed to it.

Conversations that are kept

Typing to a chatbot or speaking to a voice assistant feels as private as thinking aloud, but each conversation is stored on the company's servers. It can be read by staff and used as training data for later versions of the system, so a question about a health worry or a family argument can be reused years after you asked it. You usually accepted this in the terms and conditions, which is why keeping it can be legal and still make people uneasy.

Guessing what you never said

AI can infer sensitive facts from data that looks dull on its own, because it is good at joining clues together. You never chose to reveal the sensitive fact, so you cannot correct it or take it back.

AI is part of everyday culture now: it changes how people work, how they learn, what counts as creative work and how they find things out. Because it changes habits rather than just tools, its effects reach people who never chose to use it.

Work, learning and creating

In work, AI takes over routine parts of a job, such as counting stock on a shelf or predicting how many lunches to prepare, so staff can spend longer on the parts that need judgement. That can make a small team feel bigger, but it changes which skills are valued.

A museum tablet that changes how much it explains about an exhibit based on a visitor's questions helps people learn at the right level, and text read aloud in a natural voice helps people who find print difficult.

Creative work raises harder questions. A community pantomime that once paid a choreographer to create a routine can now use a tool that suggests steps from a style and tempo for very little, so the cast perform a routine made without a choreographer. Nothing is obviously copied, but it asks what human creativity is worth and puts pressure on the people who earn a living that way.

Trust, reliance and access

Convincing false content is cheap to make. A generated review of a bowling alley can claim its lanes are unsafe and spread through local groups in an afternoon, and when the owner posts genuine safety certificates some customers doubt those too, so trust in all online content weakens.

There is also a risk of leaning on AI too heavily. If a tool always works out how much paint a room needs, new decorators never learn to measure walls and cannot spot a wrong estimate, and an AI answer about charity volunteering rules can be confident but wrong, so answers that matter still need checking. Skills fade when they are never practised.

Access is uneven too. Paid tools, fast broadband, spare devices and confidence all vary, and many tools work best in English, so some people gain a head start while others fall further behind.

Artificial intelligence is software, but it runs on physical computers. Training a model and then using it both need electricity, and the longer and more often the computers work, the more electricity is used.

That electricity has to be generated. The carbon footprint is higher when the power comes from coal or gas and lower when it comes from wind, solar or hydro. The machines also get very hot, so cooling them uses large amounts of water as well as electricity.

AI needs hardware too. Chips and servers contain metals and other materials that must be mined, and when equipment is replaced it becomes e-waste, which can pollute soil and water if it is not recycled safely. AI can even speed up replacement: if a feature only works on the newest phone, some people upgrade sooner than they need to. If it runs on older phones, devices stay in use for longer and fewer new ones are needed.

Where AI can reduce harm

A water company can use AI to spot the pattern of pressure changes that comes before a pipe bursts, so leaks are repaired early and less water is lost. A recycling plant can use AI to recognise different plastics on a conveyor and sort them, so more material is reused instead of being made from scratch. A school canteen can forecast how many meals it will sell each day, so less food is cooked and thrown away.

These gains are not automatic. A forecasting model uses electricity itself, but if it saves many meals each week the net benefit can be large. Whether AI helps or harms the environment depends on what it is used for and how much computing it needs.