Artificial Intelligence

Artificial intelligence is not a single technology, model or piece of software. It is a broad field of computer science containing many different approaches to making machines perform tasks that appear intelligent. Modern AI encompasses technologies including machine learning, neural networks, generative AI, computer vision, language processing and autonomous systems, alongside earlier approaches built around logic, rules, search and symbolic reasoning.

AI now operates throughout everyday digital life, often without being immediately visible. Search engines use it to interpret queries, banks use it to detect unusual transactions, navigation systems use it to predict traffic, streaming platforms use it to recommend content, manufacturers use it to detect faults and technology companies use increasingly capable AI systems to understand and generate text, images, audio, software and video.

The rapid development of generative AI has made artificial intelligence far more visible, but systems such as ChatGPT represent only one part of a much larger technological field.

What is artificial intelligence?

Artificial intelligence is a branch of computer science concerned with creating machines and software that can perform tasks requiring capabilities we associate with intelligence.

Exactly what should qualify as intelligence has never had a universally accepted definition. The boundaries of AI have consequently shifted throughout its history. Capabilities such as playing chess, recognising speech or translating languages were once regarded as major demonstrations of machine intelligence. As those technologies became commonplace, they increasingly became regarded simply as features of modern computing.

A useful way to understand artificial intelligence is therefore not as a particular product or algorithm, but as an objective: to develop computational systems capable of perceiving information, processing it and producing useful behaviour that would otherwise require some degree of human intelligence or judgement.

That behaviour can take many forms. An AI system might determine whether a medical image contains an abnormality, estimate how much electricity will be required tomorrow, recognise a person's voice, recommend a film, calculate an efficient delivery route or respond to a question written in natural language.

Different systems can accomplish those tasks through very different methods. This is why artificial intelligence should not be used interchangeably with terms such as machine learning, deep learning or generative AI. They describe particular approaches or categories within the much broader field of AI.

What does AI stand for?

AI stands for artificial intelligence.

The term combines artificial, meaning created rather than naturally occurring, with intelligence, a concept associated with capabilities such as reasoning, learning, adapting, planning and solving problems.

Calling a system artificial intelligence does not necessarily mean that it possesses human-like intelligence, consciousness or understanding. An AI system can perform an intelligent-seeming task using mathematical processes that are fundamentally different from human thought.

This distinction has become particularly important as modern AI systems have become capable of producing fluent language, realistic images and increasingly sophisticated reasoning-like behaviour. The quality of an output alone does not tell us whether the process producing it resembles human cognition.

Who invented artificial intelligence?

Artificial intelligence was not invented by a single person. Its development emerged from work across mathematics, computer science, logic, neuroscience, psychology and engineering.

The intellectual foundations of AI predate modern computers. Philosophers and mathematicians had long explored whether reasoning itself could be represented through formal systems, while early computer scientists began asking whether machines might eventually imitate aspects of human thought.

British mathematician Alan Turing made one of the most influential early contributions. In his 1950 paper Computing Machinery and Intelligence, Turing proposed approaching the question of machine intelligence through observable behaviour rather than attempting first to define thinking itself. The idea later became associated with what is commonly called the Turing Test.

The term artificial intelligence was introduced by American computer scientist John McCarthy in the 1950s. McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon subsequently proposed the 1956 Dartmouth Summer Research Project on Artificial Intelligence, which is commonly regarded as a foundational moment in establishing AI as a formal research field.

The proposal reflected an ambitious belief that aspects of learning and intelligence could be described precisely enough for machines to simulate them. Nearly seven decades later, the fundamental ambition remains recognisable, even though the technologies being used to pursue it have changed dramatically.

How does artificial intelligence work?

There is no single mechanism through which artificial intelligence works.

AI is an umbrella term covering many different computational approaches. An artificial intelligence system may use programmed rules, statistical techniques, search algorithms, optimisation methods, machine learning or combinations of several different technologies.

Much of contemporary AI is based on systems that identify relationships within data. Instead of a programmer specifying every possible rule governing how the software should behave, a mathematical model can be adjusted using examples so that it becomes increasingly capable of producing useful outputs.

Other forms of AI can operate differently. A rule-based expert system may encode specialist knowledge explicitly. A planning system may examine possible sequences of actions before choosing a path towards an objective. An optimisation system may evaluate an enormous number of potential configurations to identify one that best satisfies a set of constraints.

What connects these approaches is not a particular algorithm but the broader objective of enabling computational systems to perform behaviours associated with intelligence.

What makes a system artificial intelligence?

There is no universal checklist that determines whether a piece of software should be described as AI. However, artificial intelligence systems frequently demonstrate one or more capabilities associated with perception, prediction, reasoning, learning, planning, language or autonomous action.

A conventional calculator, for example, follows precisely defined mathematical operations and would not normally be described as AI. A system capable of interpreting an ambiguous handwritten equation, understanding its meaning and selecting an appropriate way to solve it might reasonably be classified as artificial intelligence.

The boundary is not always clear.

As technologies become established, the label AI sometimes disappears from them. Optical character recognition, automated route planning and certain forms of recommendation technology have become such normal components of software that users may no longer think of them as artificial intelligence, even though their development emerged partly from AI research.

This phenomenon is sometimes described informally as the AI effect: once machines can reliably perform something previously considered intelligent, people begin to regard the capability as ordinary computing.

What are the main capabilities of artificial intelligence?

Artificial intelligence can perform a remarkably broad range of tasks, but its capabilities can be understood through several recurring categories.

Perception allows machines to extract useful information from sensory or environmental data. AI can recognise objects in photographs, identify speech within audio, interpret handwriting or analyse patterns generated by physical sensors.

Language capabilities allow machines to process human communication. AI systems can translate languages, summarise documents, answer questions, interpret speech and generate new written or spoken material.

Prediction involves using available information to estimate an unknown or future outcome. This can include predicting customer demand, weather conditions, equipment failures, financial risk or the likelihood that a transaction is fraudulent.

Reasoning and problem-solving describe systems capable of evaluating information in order to reach conclusions or determine possible solutions. The extent to which current AI genuinely 'reasons' remains debated, particularly when the term is applied to large language models.

Planning and decision-making allow systems to compare possible actions and determine which is most likely to achieve a specified objective.

Generation enables AI to create new outputs such as language, images, music, software, molecular structures or video.

Action extends AI beyond analysing information into interacting with an environment. This might involve controlling a robot, operating software, adjusting industrial machinery or carrying out a sequence of digital tasks.

Not every AI system possesses all of these capabilities, and demonstrating one does not imply that a system possesses the others.

What are the different types of artificial intelligence?

Artificial intelligence can be categorised in several ways, and no single classification system is universally accepted.

One of the best-known divides AI according to the breadth of capability a system possesses.

Artificial Narrow Intelligence

Artificial Narrow Intelligence, or ANI, refers to AI designed to operate within a restricted task or domain.

Most artificial intelligence in practical use today falls broadly into this category. A system may be extraordinarily capable at recognising images, predicting demand, playing a game or generating language while having no ability to independently apply that capability to unrelated tasks.

The word 'narrow' therefore does not mean unsophisticated. Some narrow AI systems considerably exceed human performance within specialised domains.

Artificial General Intelligence

Artificial General Intelligence, or AGI, is a proposed form of artificial intelligence capable of operating competently across a broad range of intellectual tasks rather than being restricted to a particular function.

There is no universally accepted technical definition of AGI and no agreed test for determining whether it has been achieved.

The rapid expansion of capabilities demonstrated by general-purpose AI models has intensified discussion around the concept, but claims about AGI should be treated carefully because different researchers and organisations use the term in different ways.

Artificial Superintelligence

Artificial Superintelligence, usually abbreviated to ASI, describes a hypothetical form of AI that would substantially exceed human intellectual capabilities across a broad range of domains.

ASI does not currently exist. It is primarily discussed within theoretical work concerning the long-term trajectory, governance and safety of advanced artificial intelligence.

ANI, AGI and ASI are therefore better understood as a framework for discussing levels of capability than as three existing technologies.

What is the difference between AI and automation?

Artificial intelligence and automation overlap, but they are not the same thing.

Automation describes the use of technology to carry out a process with reduced human involvement. The automated process may be completely deterministic. A system that sends an invoice automatically on the first day of every month is automated, but there is nothing inherently intelligent about the task.

Artificial intelligence becomes relevant when software must perform capabilities such as interpretation, prediction, learning, reasoning or decision-making.

The two are increasingly combined. An AI system might interpret an incoming document, determine what it contains and decide what should happen next, while an automation system then carries out the resulting workflow.

This combination is important because AI expands the range of activities that can potentially be automated. Traditional automation works particularly well when rules are predictable. AI can help software operate where inputs are less structured and decisions cannot easily be reduced to a fixed series of instructions.

Is artificial intelligence the same as machine learning?

No. Machine learning is one approach within artificial intelligence.

Artificial intelligence is the broader field concerned with developing intelligent computational systems. Machine learning focuses specifically on methods through which computers can identify patterns or improve performance using data.

The distinction matters because AI existed before modern machine learning and includes approaches that do not necessarily learn from datasets.

Machine learning has nevertheless become one of the dominant methods behind contemporary artificial intelligence, particularly following advances in computing power, access to large datasets and the development of increasingly capable neural-network architectures.

Is generative AI the same as artificial intelligence?

No. Generative AI is a category within artificial intelligence concerned with systems capable of generating new content or data.

Generative systems can produce text, photographs, illustrations, speech, music, video, computer code and other outputs. The rapid growth of systems such as ChatGPT has caused AI and generative AI to become almost interchangeable in some popular discussion, but that is technically incorrect.

Many important AI systems do not generate anything.

Fraud detection, industrial forecasting, image classification, autonomous control, recommendation systems, demand prediction and optimisation can all involve artificial intelligence without being forms of generative AI.

Understanding this distinction is important because the recent generative AI boom represents one chapter in a much larger history of artificial intelligence.

Examples of artificial intelligence

Artificial intelligence already operates across consumer technology, business, science and critical infrastructure.

Search engines use AI to understand language and determine which information is likely to answer a query. Email systems identify spam and suspicious messages. Banks use AI to detect unusual transaction patterns. Online retailers and streaming platforms recommend products and media based on predicted preferences.

Healthcare researchers use artificial intelligence to analyse medical images, investigate biological data and support drug discovery. Manufacturers use it to identify defects and anticipate equipment failures. Energy systems use AI for forecasting and optimisation, while logistics businesses use it to predict demand and improve the movement of goods.

Modern generative systems have added another visible category of applications, allowing people to create and manipulate language, software, images, speech and video through increasingly natural interfaces.

These examples illustrate an important feature of AI: some of its most economically significant uses are almost invisible. Artificial intelligence does not need to resemble a chatbot or humanoid robot to have a substantial impact.

Why is artificial intelligence important?

Artificial intelligence changes the kinds of problems computers can address.

Traditional software is extremely effective when a problem can be expressed through precise instructions. Many real-world tasks, however, involve ambiguity, enormous numbers of variables or patterns that are difficult for humans to describe explicitly.

It is relatively straightforward to write software that calculates VAT according to a known percentage. It is far more difficult to specify every rule needed for a computer to recognise a dog in any possible photograph, understand a sentence written in ordinary language or predict when a component inside a complex machine is likely to fail.

AI provides methods for approaching these less deterministic problems.

That capability means artificial intelligence increasingly sits inside the infrastructure of modern organisations rather than existing solely as a standalone technology. It can influence how information is discovered, how products are designed, how demand is forecast, how scientific hypotheses are explored and how decisions are made.

Its importance therefore comes not simply from the creation of new AI products, but from the prospect of intelligence becoming a general capability embedded throughout software.

What are the limitations of artificial intelligence?

Artificial intelligence can produce impressive results without being consistently reliable.

AI systems can make incorrect predictions, misinterpret unfamiliar situations and reproduce patterns or biases contained within their data. Generative systems can confidently produce information that is false. Models that perform well during testing can behave differently when deployed into environments where conditions change.

Some AI systems are also difficult to interpret. The complexity that enables a large neural network to capture subtle relationships within data can make it difficult to establish exactly why a particular conclusion was reached.

There are broader limitations too. Artificial intelligence can require substantial computing infrastructure and energy. Its development raises questions surrounding copyright, privacy, security, employment, accountability and access to data. Increasingly autonomous systems also create questions around who remains responsible when software takes an action rather than simply providing information.

None of these limitations means AI cannot be useful. They mean its capabilities need to be understood alongside the conditions under which those capabilities can fail.

Is artificial intelligence intelligent?

The answer depends partly on what is meant by intelligence.

Artificial intelligence systems clearly demonstrate abilities that would be considered signs of intelligence when performed by humans, including language use, perception, planning, pattern recognition and problem-solving.

Whether those capabilities constitute intelligence in a deeper sense is a much harder question.

A system can produce an intelligent output without necessarily arriving at it in the same way a person would. Modern AI systems are computational systems. Their internal mechanisms do not need to reproduce the biology or subjective experience of a human brain in order to perform useful tasks.

For this reason, discussions about AI benefit from separating capability from consciousness. A machine becoming increasingly capable does not automatically demonstrate that it possesses awareness, emotions, beliefs or a subjective experience of the world.

Is artificial intelligence conscious?

There is currently no established scientific evidence that today's AI systems are conscious.

Modern models can generate statements about emotions, personal experiences or their own existence because they are capable of producing highly sophisticated language. Such responses should not in themselves be interpreted as proof that the system experiences the concepts it describes.

Machine consciousness remains a subject of philosophical and scientific debate, and consciousness itself is not yet sufficiently understood for researchers to possess a universally agreed test that could simply be applied to an artificial system.

Claims that an AI is conscious should therefore be distinguished from evidence that it has become more capable.

What is the future of artificial intelligence?

Artificial intelligence is moving towards systems that can operate across more types of information, perform more complex tasks and interact more directly with the digital and physical world.

The boundaries between previously separate AI capabilities are already becoming less distinct. Systems that began primarily as language models can increasingly interpret images, process speech, create video, use external tools and interact with software. AI is simultaneously becoming embedded into conventional products, meaning the distinction between 'AI software' and software generally may become less meaningful over time.

Another major challenge concerns adaptation. Many AI systems are still trained and then deployed as relatively static models. The ability for artificial intelligence to learn safely from new experience, respond to changing conditions and improve over time remains an important area of development.

Research will also continue around efficiency, reasoning, autonomy, robotics, interpretability, safety and the possibility of more general forms of machine intelligence.

Exactly where that development leads cannot be stated with certainty. What is already clear is that artificial intelligence is no longer a specialist area sitting at the edge of computing. It is becoming one of the technologies through which computing itself is being redefined.