What is artificial intelligence and how do you use it?

Artificial intelligence (AI) is no longer a topic confined to research labs. Marketing teams use it to speed up content production, customer service teams deploy it for first-line answers, and finance departments let models assist with forecasts. At the same time, there is a lot of confusion about what AI actually is, how it works and where its limits lie.

This page explains what artificial intelligence is, how the technology has developed, which forms exist and how organisations get started with it responsibly.

What is artificial intelligence?

Artificial intelligence is the umbrella term for technology that performs tasks normally requiring human intelligence: understanding language, recognising patterns, making decisions, generating predictions. It is not consciousness and not a "thinking machine" in the human sense — it is software that produces output resembling intelligent behaviour, based on data and mathematical models.

Important distinctions:

• AI is the broad field: all techniques that mimic human intelligence.
• Machine learning is a subfield of AI in which systems learn from data rather than being explicitly programmed.
• Deep learning is a subfield of machine learning that uses neural networks with multiple layers.
• Generative AI is an application of deep learning that generates new content: text, images, audio, code.

In short: AI is the umbrella, machine learning and deep learning are the underlying technique, and generative AI is a concrete application that has become the most visible in recent years.

A brief history of AI

The idea of artificial intelligence is older than most people think.

• 1940s-1950s — the foundations: in 1950, Alan Turing raised the question of whether machines can "think" and introduced a test to assess machine intelligence, later known as the Turing test.
• 1956 — the Dartmouth Conference: at this conference the term "artificial intelligence" was formally introduced. This moment is regarded as the official starting point of AI as a field.
• AI winters: in the following decades, several periods followed in which expectations exceeded what the technology could deliver. Funding dried up and research stagnated — known as "AI winters".
• The rise of machine learning: from the 1990s and 2000s onward, growing availability of data and computing power made statistical learning methods practically usable.
• The breakthrough of deep learning: in the 2010s, multi-layer neural networks led to major leaps in image recognition and speech processing.
• Transformers and generative AI: the introduction of the transformer architecture made it possible to train large language models capable of generating text, code and later also images and audio. This took AI from a research field to widely used technology within a few years.

Meaning: the current wave of AI is not a standalone phenomenon, but the result of decades of research coming together with sufficient data and computing power.

How does artificial intelligence work?

Broadly speaking, AI systems learn to recognise patterns in data and apply those patterns to new situations. The main learning methods within machine learning are:

• Supervised learning: the model learns from labelled examples, such as thousands of emails marked as spam or not spam.
• Unsupervised learning: the model finds structure in unlabelled data itself, for example clustering customers with similar behaviour.
• Reinforcement learning: the model learns through trial and error based on reward and punishment, often used in robotics and game environments.

Neural networks and large language models

A neural network is a mathematical model loosely inspired by the working of brain cells: layers of connected "nodes" that pass on and adjust signals during training. Large language models (LLMs), such as the models behind many chatbots, are neural networks trained on huge amounts of text to understand and generate language.

Training data and hallucinations

The quality and composition of training data largely determine what a model can and cannot do. Two related concerns:

• A model reproduces patterns from its training data, including any errors, biases or outdated information.
• Large language models can hallucinate: confidently present information that is factually incorrect or fabricated. This happens because the model generates language based on probability, not on verified facts.

Meaning: AI output should always be checked, especially for factual claims, figures, and legal or medical information.

Types of AI: narrow, general and super

• Narrow AI: specialised in a single task, such as generating text, recognising images or optimising routes. All AI that exists and is used today falls into this category — including advanced language models.
• General AI (AGI): a hypothetical system able to reason at a human level across a wide range of tasks, comparable to human versatility. This does not currently exist.
• Super AI: an even further-reaching, theoretical concept of intelligence that surpasses human intelligence in every respect. This remains speculative for now.

In short: when businesses talk about "AI", they almost always mean narrow AI tailored to a specific task.

AI applications by business function

Function	Typical application
Marketing	content generation, personalisation, campaign optimisation, data analysis
Sales	lead scoring, predictive analytics, automated follow-up
Customer service	chatbots, automatic ticket classification, summarising conversations
Finance	fraud detection, forecasting, invoicing automation
Operations	planning and logistics, predictive maintenance, quality control
HR	CV screening, candidate matching, drafting job postings

AI in marketing

Within marketing, AI is used among other things for generating and optimising content, segmenting audiences and automating repetitive tasks. This increasingly happens through AI agents: systems that independently carry out a series of steps, from research to draft. The quality of the output still depends on a well-considered content strategy — AI speeds up execution, but does not replace strategy.

Risks and points of attention

• Bias: if training data contains skewed or incomplete patterns, the model adopts them. This can lead to unfair outcomes, for example in CV screening.
• Privacy and data protection: processing personal data in AI systems must comply with data protection regulation such as the GDPR. This applies both to training data and to input users provide to an AI tool.
• Copyright: the legal status of training data and generated content is still evolving and differs by jurisdiction.
• Reliability: as noted, models can hallucinate; human review remains necessary.
• Energy consumption: training and running large models requires significant computing power and therefore energy.
• Dependency: relying too heavily on a single vendor or model can leave an organisation vulnerable to price changes, outages or policy changes.

Regulation in Europe: the AI Act

The European Union has introduced a legal framework for AI systems with the AI Act. Broadly, the law uses a risk-based classification:

• Unacceptable risk: applications that are prohibited, such as certain forms of social scoring.
• High risk: applications with strict requirements around documentation, transparency and human oversight, for example in recruitment or credit assessment.
• Limited risk: applications with transparency obligations, such as making clear that a user is interacting with a chatbot.
• Minimal risk: most current AI applications, without additional obligations.

The regulation is being introduced in phases, with obligations for the highest-risk categories taking effect earlier than the broader rules. Organisations using AI are well advised to determine, per application, which risk category it falls into and which obligations follow from that.

Meaning: not every AI application falls under strict obligations, but it is advisable to assess in advance which category a specific application belongs to.

How to get started with AI as an organisation

1. Choose a concrete use case. Start with a recurring, time-consuming problem rather than with the technology itself.
2. Start with a pilot. Test on a small scale, with clear success criteria, before scaling up.
3. Set internal policy. Define which data may and may not be used, who is responsible for reviewing output, and how privacy-sensitive information is handled.
4. Invest in training. Employees need to know what a tool can and cannot do, and how to assess its output.
5. Evaluate and scale. Measure results, adjust, and only expand to other departments once the pilot demonstrably adds value.

The role of people

Despite the progress, human involvement remains essential. People set the goals, assess the quality of AI output, take responsibility for decisions and provide the context a model itself lacks. AI is a tool that speeds up and supports tasks — final responsibility and judgement remain with people.

Frequently asked questions

What is the difference between AI and machine learning?

AI is the broader field of technology that mimics human intelligence. Machine learning is a specific approach within AI in which systems learn from data instead of being explicitly programmed.

What is generative AI?

Generative AI is AI that generates new content, such as text, images, audio or code, based on patterns it has learned from training data.

What is an AI hallucination?

A hallucination is when an AI model confidently presents information that is factually incorrect or fabricated. This happens because the model generates language based on probability, not on verified facts.

Does general AI (AGI) already exist?

No. All current AI applications are narrow AI: specialised in specific tasks. General, human-like intelligence does not currently exist.

What does the European AI Act regulate?

The AI Act classifies AI applications by risk level and sets requirements accordingly, ranging from a ban for unacceptable risk to transparency obligations for limited risk. The rules are being phased in over time.

Should my business start using AI now?

That depends on the organisation and the use case. A focused pilot around a concrete problem is usually a wiser start than rolling out AI broadly without a clear goal.