Artificial general intelligence
Artificial General Intelligence: What It Is and What It Means for Your Business
Artificial general intelligence, usually shortened to AGI, is the name for a computer system that could think and learn across almost any subject the way a person can, rather than being built to do one job. It does not exist yet. The AI that businesses use today, including ChatGPT, is very good at particular things, but it is not general in the way a person is. This page explains in plain English what artificial general intelligence is, how it differs from the AI you can use now, how close the experts think it is, and what AI can already do for your business while the world waits to find out.
The meaning
What is artificial general intelligence?
Artificial general intelligence is an AI that could understand, learn and do any intellectual task that a human being can. A person who has learned to drive can also learn to cook, read a contract, comfort a friend and work out why the boiler has stopped. Nobody has to rebuild their brain for each new job. They take what they already know, apply it to something new and work it out. An AGI would be able to do the same: given a task it has never seen before, it would work out how to do it, the way a sensible new member of staff would.
The phrase was first used in the late 1990s and became common in the early 2000s, when a group of researchers wanted a name that set this goal apart from the narrower kinds of AI being built at the time. The companies building the largest AI systems today each describe it in their own way. OpenAI, the company behind ChatGPT, defines AGI in its charter as highly autonomous systems that outperform humans at most economically valuable work
. Google DeepMind published a paper in November 2023 that sets out levels of AGI, from “emerging”, where a system is as good as an unskilled person at a wide range of tasks, up to “superhuman”, where it is better than every person at all of them.
- Level 1EmergingAs good as an unskilled person
- Level 2CompetentAs good as an average skilled adult
- Level 3ExpertBetter than 9 in 10 skilled adults
- Level 4VirtuosoBetter than 99 in 100 skilled adults
- Level 5SuperhumanBetter than every person
Google DeepMind’s levels of AGI, from its paper of November 2023, for an AI that can do a wide range of tasks.
What all the definitions share is the word general. The test is not whether a machine can do one clever thing, but whether it can do almost anything it is asked to.
Narrow AI and general AI
How AGI differs from the AI you use today
Every AI system in use today, however impressive, is what researchers call narrow AI. It has been built and trained for a particular kind of work. The program that recommends films on Netflix cannot drive a car. The system that reads number plates at a car park cannot write a letter. Even the AI that beats the world champion at chess cannot play draughts unless it is built again for draughts.
Large language models such as ChatGPT, Claude and Gemini are much broader than anything that came before them. They can write, summarise, translate, answer questions on almost any subject and write computer code, which is why some people say they are the first step on the way to AGI. But they still have clear limits. They learn once, during their training, and do not keep learning from what happens to them afterwards. They can state something false with complete confidence. They can get a hard exam question right and then fail at a simple task that a ten-year-old would manage. And they do not plan and act in the world on their own over weeks and months, the way a person running a business does.
Today’s AI
Narrow AI is built for one kind of job. It is very good at that job, it cannot carry what it knows over to a different one, and it is what every business uses today.
Artificial general intelligence
AGI would be able to learn any kind of job. It would take what it knows in one area and use it in another, and work out new problems on its own. It does not exist yet.
There is also a third term you will see: artificial superintelligence, often shortened to ASI. That is an AI that would be far cleverer than the cleverest people in every field. The philosopher Nick Bostrom made the idea widely known with his 2014 book Superintelligence. AGI would match people; superintelligence would go well beyond them. Many researchers think that if AGI arrives, superintelligence could follow it quickly, because an AI as capable as the people who designed it could help to improve itself.
The abilities
What would make an AI “general”?
There is no single test that everyone agrees on, but researchers tend to look for the same abilities.
- 01
Learning something new without being rebuilt
A general AI would pick up a new skill from a handful of examples or a set of instructions, the way a person does in their first week in a new job, rather than needing millions of examples and months of training.
- 02
Reasoning
It would work through a problem it has never met before, step by step, and notice when its own answer does not add up.
- 03
Using what it knows in one area in another
Someone who understands how water flows through pipes can make a fair guess at how traffic flows through a town. Researchers call this transfer, and today’s AI is still weak at it.
- 04
Common sense
Knowing that a glass dropped on a stone floor will probably break, that a shop is likely to be shut on Christmas Day, and a million other things nobody ever bothers to write down.
- 05
Planning and acting over a long time
Setting itself a goal, breaking it into steps, carrying them out over days or weeks, and changing course when something goes wrong.
- 06
Understanding the physical world
Many researchers think a general intelligence also has to understand how real objects behave and move, not only how words fit together.
Alan Turing proposed the most famous test of machine intelligence in 1950. In what he called the imitation game, a person holds typed conversations with a hidden human and a hidden machine, and if they cannot reliably tell which is which, the machine has passed. Today’s chatbots can hold a conversation good enough to fool many people for a few minutes, which is why most researchers now say the Turing test is not enough on its own. Being able to chat like a person is not the same as being able to do everything a person can.
The timeline
How close are we to artificial general intelligence?
- 1950
Alan Turing asks whether machines can think, and proposes the imitation game.
- 1956
A summer workshop at Dartmouth College in the United States gives the field its name, artificial intelligence.
- 1997
IBM’s Deep Blue computer beats Garry Kasparov, the world chess champion.
- 2016
Google DeepMind’s AlphaGo beats Lee Sedol, one of the best players in the world at Go, a game far harder for computers than chess.
- 2022
OpenAI releases ChatGPT, which was reported to have reached 100 million users within two months.
- 2023
Google DeepMind sets out its levels of AGI, and the UK holds the first international AI Safety Summit at Bletchley Park.
- Today
AI that can write, reason through problems, look at pictures, speak and use a computer, but nothing that the experts agree is general.
The honest answer is that nobody knows, and the people closest to the work disagree sharply.
The heads of the big AI companies are the most confident. Sam Altman, who runs OpenAI, wrote in January 2025 that the company is now confident it knows how to build AGI. Demis Hassabis, who runs Google DeepMind and shared the 2024 Nobel Prize in Chemistry for using AI to work out the shapes of proteins, said in 2025 that he expects AGI within five to ten years. Dario Amodei, who runs Anthropic, the company behind Claude, wrote in 2024 that AI as capable as the top experts in most fields could arrive as soon as 2026 or 2027.
AI researchers as a whole are more cautious. In a survey of 2,778 researchers published in early 2024, the average answer gave an even chance that machines would be able to do every task better and more cheaply than people by 2047. That is thirteen years sooner than the same survey found only a year earlier, which shows how quickly opinion has moved.
Some of the most respected names in the field think today’s approach will not get there at all. Yann LeCun, who led AI research at Meta for over ten years and is one of the pioneers of the neural networks behind today’s AI, has argued for years that large language models, however big they are made, will not lead to human-level intelligence, because they learn from text rather than from the world.
Both sides
The case for AGI arriving soon, and the case against
The case for it arriving soon
- The systems have improved far faster than almost anyone predicted. In 2019 the best language models could just about write a few sensible paragraphs; by 2023 they were passing professional exams, including the American bar exam that lawyers sit.
- Each larger model, trained on more data with more computing power, has been more capable than the one before it, and the AI companies are now spending hundreds of billions of dollars on data centres to keep that going.
- The newest models do not only answer. They reason through a problem step by step before they reply, and they can use software and carry out tasks on a computer on their own.
- AI is starting to help with AI research itself, which could speed everything up.
The case against
- Today’s AI still makes mistakes that no person would make, and states false things with confidence.
- It does not keep learning once it has been trained, which every person does every day of their life.
- Doing well in tests is not the same as doing a job. An AI that passes the bar exam still cannot run a solicitor’s practice.
- Researchers have warned that the supply of good-quality writing to learn from is running out, and the cost of building each new model is rising steeply.
- Many researchers think something important is still missing, such as a proper understanding of the physical world, and that no amount of extra size will supply it.
Where that leaves a business owner is this. AGI might arrive in a few years, in a few decades, or not in the form people imagine. Nobody can plan a business around a date the experts cannot agree on. What you can plan around is the AI that is already here, which is further down this page.
For business
What artificial general intelligence could change for businesses
If AGI does arrive, it would be the biggest change to how work is done since computers came into offices, and probably bigger. An AI that could do any office task as well as a capable person could take on whole jobs rather than parts of them: answering every customer, doing the books, writing and checking contracts, planning a marketing campaign, ordering the stock and chasing the invoices, all at once and at any hour.
For a small business that would cut both ways. On one side, a business of five people could do the work of fifty, and a sole trader could offer the kind of service that only a large firm can afford today. On the other, the things that set a business apart now, such as knowing the trade, answering quickly and writing well, would be available to everyone. What would count then is what the AI cannot copy: the relationships, the reputation and the judgement of the people who run the business.
Economists are divided on what it would mean for jobs. Some expect new kinds of work to appear, as they did after electricity and the internet. Others think this change is different, because it is the first that could do almost any job a person can. What almost everybody agrees on is that the businesses best placed for it will be the ones already using AI well, because they will have learned what it is good at, what it gets wrong and how to check it.
Safety
The risks, and why safety matters
A machine as capable as a person, or more capable, raises questions that the people building it take seriously. The main worries are these.
Control
Making sure a very capable AI does what we actually want, and not something we asked for clumsily. Researchers call this alignment.
Misuse
Powerful AI in the wrong hands could be used for fraud, cyber attacks or worse.
Jobs and money
If AI can do most work, who earns a living, and who owns the AI that does the earning?
Truth
AI can already produce convincing fake writing, pictures, voices and video, which makes it harder to know what is real.
Geoffrey Hinton, whose work in the 1980s led to the neural networks behind today’s AI and who shared the 2024 Nobel Prize in Physics for it, left Google in May 2023 so that he could speak freely about these risks. In November 2023 the UK government brought world leaders and the AI companies together at Bletchley Park, where the wartime codebreakers worked, for the first international AI Safety Summit, and it set up its own institute to test the most advanced AI systems.
For an ordinary business, the same care applies on a much smaller scale today. Any AI that speaks to your customers should answer only from information you have given it, should pass the customer to a person when it is out of its depth, and should keep a record of everything it says so that you can check it. That is how we build ours.
AI you can use now
What AI can already do for your business today
You do not need to wait for artificial general intelligence to put AI to work. The narrow AI available now is already good enough to take real work off a business, as long as each tool is set up for one job, trained on that business’s own information and checked by a person. These are the AI products we build and run for businesses.
01
An AI chatbot for your website
A chatbot is a small chat window that sits in the corner of your website. When a customer types a question into it, it answers them straight away, at any time of the day or night, using what is written on your own website: your services, your prices and your opening hours. When a customer wants to get in touch, it takes their name and details. Every chat is emailed to you and kept in your login, so you can read them all. We build it, train it on your website, put it on your website for you and keep it running. It costs £39 a month plus VAT, with nothing to pay up front, no limit on the number of chats and no contract.
02
An AI receptionist for your phone
An AI receptionist is a person-like voice called Sarah who answers your business phone when you cannot. She picks up straight away, any time of day or night, finds out what the caller needs, takes their name and number, and writes the whole call down for you, so you get an email the moment the call ends. We set her up for you and give her a UK phone number. There are two packages: Light, with 5 hours of calls a month for £40 a month, and Pro, with 8 hours of calls a month for £65 a month. Both are plus VAT, with nothing to pay up front and a 12 month minimum contract.
03
AI automation
A lot of office time goes on work that follows the same steps every time: typing the details from an invoice into the accounts, sorting the morning’s emails, putting together the same report every week. AI automation takes those jobs on. We look at how the work is done now, build the AI to do it, connect it to the systems you already use, and test it on real work before it goes live, so the hours your team spent on it go back into the business.
04
AI that answers from your own documents
Most businesses have years of knowledge sitting in manuals, price lists, policies, contracts and old emails. We build AI that searches your own documents and gives an answer together with the document it came from, so your staff can find what they need in seconds and check it for themselves. This is called a RAG system, and because it answers only from your own information, it does not make things up. Where the information is sensitive, we can run the AI on your own servers so that nothing leaves your business.
Each of these is narrow AI doing one job well, which is exactly what today’s AI is good at. Everything else we build is on our page about AI development for business.
Our process
How we build AI for your business
We are Mediatopia, a web design and development company, and we have been building websites and software for businesses for 16 years, with more than 3,000 websites delivered.
A free consultation
We talk about your business, your systems and what you want AI to do, and we tell you honestly what AI can do for you and what it will cost.
Design
We plan how the AI will work, what it will say and which of your systems it connects to, and we agree all of it with you before we build anything.
Build and test
We build the AI, train it on your business’s information, connect it to your systems, and test it with real situations before it goes live.
Launch and improve
Once it is live, we watch how it performs, read what it says to your customers, and keep improving it.
Common questions
Questions people ask about artificial general intelligence
Does artificial general intelligence exist yet?
Is ChatGPT artificial general intelligence?
What is the difference between AI and AGI?
Which type of AI is still only an idea?
What is the difference between AGI and superintelligence?
Who is building AGI?
Will artificial general intelligence take people’s jobs?
How can my business use AI now?
Talk to us about
AI for your business.
Artificial general intelligence may be years away, but the AI that can answer your customers, take your calls and do your repetitive work is ready now. Ring us on 0117 299 0348 or send us your details, and we will tell you honestly what AI can do for your business and what it will cost.