Skip to content
Knowledge Hub/Article
Article

What Is AI? How It Works, Types and Examples

We cover official definitions of AI, how it differs from rule-based software, its types, enterprise examples, risks and where a business should start.

Oğuzhan Gerçek··9 min read
What Is AI? How It Works, Types and Examples

Short answer: Artificial intelligence (AI) is the general name for machine-based systems that infer, from the input they receive, how to generate outputs such as predictions, content, recommendations or decisions. In conventional software, people write the rules; an AI system learns much of its behavior from example data. Every system in use today is narrow AI, designed for specific tasks; general AI, which could take on any task the way a person can, does not exist yet. Customer service, fraud detection and predictive maintenance are typical enterprise examples, and before starting, a business should pin down the goal, the data and who owns the risk.

What is artificial intelligence?

In IBM's definition, artificial intelligence (AI) is technology that enables computers and machines to simulate human learning, comprehension, problem solving and decision making.

In the OECD's definition, an AI system is a machine-based system that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations or decisions that can influence physical or virtual environments. According to a November 29, 2023 explainer on OECD.AI, adding "content" to the outputs in the update clarifies that the definition also covers generative AI.

Article 3 of the EU Artificial Intelligence Act (EU AI Act) uses the same core, and the Act's recitals say the notion should be closely aligned with the work of international organisations. NIST's AI RMF 1.0 adapts its definition from the OECD and from ISO/IEC 22989, the terminology standard published in July 2022. Our glossary has a short definition of the term.

A short history: Turing's question and the Dartmouth proposal

In "Computing Machinery and Intelligence", published in Mind in October 1950, Alan Turing replaced the question "Can machines think?" with a test he called the imitation game. The Dartmouth proposal by McCarthy, Minsky, Rochester and Shannon, dated August 31, 1955, called for a two-month, ten-person study of artificial intelligence in the summer of 1956. Its conjecture was that every aspect of learning or any other feature of intelligence could in principle be described so precisely that a machine could be made to simulate it.

How does AI work?

In conventional software, the rules a programmer writes determine the behavior. The AI Act's recitals exclude systems that execute operations based solely on rules defined by people; a key characteristic of AI systems, they say, is their capability to infer. Google Cloud sums up the difference: teaching a computer by showing it a million examples instead of writing a million rules. Methods that learn from examples go by the name machine learning, and we cover the types of learning in our machine learning article. The work happens in two stages:

  • Training: The model learns the patterns in example data; according to Google Cloud, the quality and quantity of that data are crucial to performance.
  • Inference: In IBM's definition, using a trained model to make predictions on new data.

We look at the hardware needs of the two stages in our GPU article.

Types of AI

Three distinctions are common:

  • By capability: Google Cloud distinguishes artificial narrow intelligence (ANI), artificial general intelligence (AGI) and artificial superintelligence (ASI). Narrow AI is designed for a single task and is the only kind that exists today; AGI does not exist yet, and ASI is theoretical. IBM draws the same line between weak and strong AI.
  • By function: The OECD definition gives predictions, content, recommendations and decisions as examples of outputs. Google Cloud adds computer vision and natural language processing (NLP).
  • By method: According to IBM, the concepts are nested: machine learning sits directly beneath AI, deep learning is a subset of machine learning, and generative AI is built on deep learning models.

Generative AI, agents and RAG

In NIST's Generative AI Profile, published in July 2024, generative AI is defined as the class of AI models that emulate the structure and characteristics of input data in order to generate derived synthetic content, such as text, images, audio and video. We explain large language models and hallucinations in our LLM article, and our glossary has a short definition. Two patterns build on generative models:

  • Agents: According to Google Cloud, AI agents are software systems that use AI to pursue goals and complete tasks on behalf of users; they operate with more autonomy than assistants (AI agent article).
  • Answers from company data: As AWS describes it, retrieval-augmented generation (RAG) has the model reference an authoritative knowledge base outside its training data before it generates a response (RAG article).

Examples and uses of AI

Google Cloud's everyday examples are a smartphone's virtual assistant, recommendations on streaming services, email spam filters and navigation apps. IBM's enterprise examples include:

  • Customer service: Chatbots and virtual assistants that handle customer inquiries and support tickets.
  • Fraud detection: Algorithms that flag anomalies such as unusual spending or login locations.
  • Predictive maintenance: Models that predict equipment failures before they occur.
  • Recruitment and software development: Screening resumes and matching candidates; code generation tools.

Adoption figures depend on who is measured. Stanford HAI's 2026 AI Index, drawing on McKinsey's survey, puts the share of respondents whose organization uses AI in at least one business function at 88% for 2025, up from 78% in 2024. According to Eurostat, 20.0% of EU enterprises with 10 or more employees used AI in 2025, up from 13.5% in 2024; the most common use was analysing written language (11.8%).

AI in businesses and SMEs: where to start

According to the OECD's 2026 D4SME survey of more than 2,000 SMEs in 12 countries, most businesses use off-the-shelf AI products, and some are experimenting with more tailored applications, including AI agents. The sample is not representative, but the barriers to using digital tools are familiar: time constraints, maintenance costs and skills gaps. A sensible order for getting started:

  1. Write down the goal and the metric: Be clear about which task should get faster or which error should become rarer.
  2. Keep an inventory: Record which team uses which tool, with which data.
  3. Classify your data: Decide which data may go into an off-the-shelf tool; personal data brings the obligations of the KVKK (Türkiye's personal data protection law) into play (KVKK article).
  4. Keep a human in the loop: Have a person check any output that touches customers, money or individuals.
  5. Plan for operations: Data preparation and monitoring the model in production are separate work items; we cover the first in our data platform article and the second in our MLOps article.

For organizations that will run models on their own infrastructure, we cover this layer in our AI stack design and engineering service, and AI in IT operations on our AIOps and intelligent operations page.

AI risks and governance

NIST's Generative AI Profile lists 12 risks, among them confabulation (confidently stated but false content), data privacy, harmful bias and information security. The main frameworks are:

  • NIST AI RMF 1.0: A voluntary framework released on January 26, 2023; its core is made up of four functions: GOVERN, MAP, MEASURE and MANAGE. As of October 2026, NIST says it is being revised.
  • OECD AI Principles: Adopted in May 2019 and updated in May 2024; transparency, robustness and accountability are among the five principles.
  • EU AI Act: According to the European Commission, it entered into force on August 1, 2024, became applicable on August 2, 2026, with some exceptions, and defines four levels of risk. Practices with unacceptable risk, such as social scoring, are banned; CV-sorting software for recruitment is high-risk; chatbots carry transparency obligations; and minimal-risk systems such as spam filters face no rules. Rules for high-risk uses in sensitive areas will apply from December 2, 2027.

We cover manipulation of training data in our data poisoning article, the management system standard in our ISO/IEC 42001 article and the governance agenda in Türkiye in our AI governance article. We handle the policy and risk framework in our governance, risk and compliance service.

Frequently asked questions

What does AI mean? AI stands for artificial intelligence. It describes systems that can perform tasks that typically require human intelligence, such as learning, reasoning and problem solving.

How does AI work? A model is trained on example data and learns its patterns; at inference time it applies them to new inputs to produce predictions, recommendations or content.

What are the types of AI? By capability, AI is divided into narrow, general and super AI, and only narrow AI exists today. By function, the categories include vision, language, recommendation and decision systems, and generative models.

What are some examples of AI, and where is it used? Virtual assistants, spam filters, recommendation systems and navigation apps are everyday examples. Enterprise examples include customer service, fraud detection, predictive maintenance and recruitment.

What is generative AI? It is the name for models that emulate the structure of input data to generate new text, images, audio or video. Large language models (LLMs) are the members of this class that generate text.

Sources