The expression artificial intelligence combines artificial, meaning produced by humans rather than occurring naturally, and intelligence, broadly referring to the capacity to learn, reason, solve problems, adapt and achieve goals.

There is no single universally accepted definition of intelligence or artificial intelligence. Different definitions emphasize different characteristics, including:

• learning from experience
• reasoning
• problem-solving
• perception
• language
• planning
• adaptation
• decision-making
• autonomy
• goal-directed behaviour

John McCarthy, who coined the term, described AI as the science and engineering of making intelligent machines.

The modern OECD definition places greater emphasis on what AI systems do: they receive inputs, infer how to produce outputs and may generate predictions, recommendations, content or decisions that affect physical or virtual environments.

The intellectual roots of artificial intelligence extend well beyond modern computers.

Philosophers have debated the nature of intelligence, reasoning and mechanical thought for centuries. The development of formal logic, probability, mathematics, neuroscience and computing during the nineteenth and twentieth centuries eventually created the technical foundations for AI research.

Early mechanical ideas

The idea of creating artificial beings or mechanical minds appears in myths and philosophical traditions from many cultures.

Ancient Greek mythology included artificial beings such as Talos, while later European traditions produced stories of mechanical humans, automata and artificial creatures.

These were primarily philosophical, religious or literary concepts rather than scientific implementations.

The development of mechanical calculators and automata during the seventeenth through nineteenth centuries demonstrated that certain forms of human calculation could be mechanized.

The emergence of electronic computing during the twentieth century transformed discussions about machine intelligence.

Mathematicians and computer scientists began investigating whether machines could perform operations traditionally associated with human reasoning.

One of the most influential figures was Alan Turing.

In 1950, Turing published his famous paper Computing Machinery and Intelligence, which asked whether machines could think and introduced what later became known as the Turing test.

Rather than attempting to define “thinking” directly, the test proposed evaluating whether a machine could produce conversational behaviour sufficiently similar to that of a human.

Turing’s work became an important intellectual foundation for AI.

The modern academic field of artificial intelligence is generally associated with the Dartmouth Summer Research Project on Artificial Intelligence, held in 1956.

The workshop was organized by John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon.

The proposal argued that aspects of learning and intelligence could potentially be described precisely enough for machines to simulate them.

Dartmouth itself describes the 1956 meeting as the birthplace of AI as a formal field of research.

The participants investigated questions including whether machines could:

• use language
• form concepts
• solve problems
• improve themselves
• reason about information
• simulate aspects of human intelligence

The optimism surrounding the early field was considerable.

During the 1950s and 1960s, researchers developed programs capable of performing tasks that appeared surprisingly intelligent for their time.

Early AI systems demonstrated abilities such as:

• mathematical problem-solving
• theorem proving
• playing games
• symbolic reasoning
• language manipulation
• solving constrained problems

One important characteristic of early AI was its reliance on symbolic representation.

Researchers attempted to represent knowledge explicitly through:

• symbols
• rules
• logical statements
• search procedures
• manually encoded relationships

This approach became known as symbolic AI or good old-fashioned AI (GOFAI).

During the 1970s and 1980s, expert systems became an important branch of AI.

An expert system attempted to reproduce aspects of human specialist knowledge using a knowledge base and inference rules.

A typical architecture contained:

Knowledge base → inference engine → conclusion

Expert systems were developed for domains such as:

• medicine
• chemistry
• engineering
• finance
• geological exploration

Although some achieved commercial success, they had important limitations.

Knowledge had to be explicitly encoded, systems were often difficult to maintain, and they generally struggled when confronted with situations outside their predefined rules.

AI research has experienced periods known as AI winters, during which enthusiasm, funding and expectations declined.

These periods occurred partly because early systems failed to meet ambitious expectations.

Researchers discovered that problems that appeared simple for humans—such as understanding ordinary language, recognizing objects in unpredictable environments or reasoning with incomplete information—could be extraordinarily difficult for computers.

Nevertheless, research continued.

The field gradually shifted toward methods capable of learning from data rather than relying exclusively on manually written rules.

Machine learning (ML) is a major branch of AI in which computer systems learn patterns or relationships from data rather than being programmed with explicit instructions for every situation.

A simplified machine-learning process can be represented as:

Data → training algorithm → model → new input → prediction/output

For example, instead of manually programming every characteristic of a cat, a machine-learning system can be trained using many labeled examples of cats and non-cats.

The system learns statistical patterns that allow it to classify previously unseen images.

Machine learning has become one of the dominant approaches to modern AI.

Supervised learning

In supervised learning, a model is trained using examples containing inputs and known outputs.

Examples include:

• email → spam/not spam
• photograph → object category
• patient data → predicted risk
• house characteristics → predicted price

Supervised learning is widely used for classification and regression.

Unsupervised learning

In unsupervised learning, the system receives data without predefined labels and attempts to discover structure within it.

Examples include:

• clustering customers
• identifying unusual transactions
• discovering topics
• reducing dimensionality
• finding hidden patterns

Semi-supervised learning

Semi-supervised learning combines relatively small quantities of labeled data with larger quantities of unlabeled data.

This approach can be useful when obtaining human-generated labels is expensive or time-consuming.

Self-supervised learning

Self-supervised learning allows models to generate learning signals from the structure of the data itself.

It has become particularly important in modern language and multimodal models.

For example, a language model can learn by predicting missing or subsequent tokens in large collections of text.

Reinforcement learning

In reinforcement learning, an agent interacts with an environment and learns through feedback.

A simplified process is:

State → action → reward/penalty → updated strategy

Reinforcement learning has been used in:

• robotics
• games
• control systems
• optimization
• recommendation
• autonomous systems

Artificial neural networks are computational models inspired loosely by biological nervous systems.

They consist of interconnected computational units commonly organized into layers.

A simplified neural network contains:

Input layer → hidden layers → output layer

Neural networks can learn complex relationships between inputs and outputs by adjusting numerical parameters during training.

They became particularly powerful when combined with:

• large datasets
• powerful GPUs and specialized AI hardware
• improved algorithms
• distributed computing

Deep learning refers broadly to machine-learning methods based on neural networks with multiple layers.

Deep learning dramatically improved performance in areas such as:

• image recognition
• speech recognition
• natural-language processing
• machine translation
• game playing
• scientific modeling
• generative AI

Modern AI systems frequently contain millions, billions or even larger numbers of learned parameters.

Computer vision is the branch of AI concerned with enabling machines to interpret visual information.

Applications include:

• image classification
• object detection
• facial recognition
• medical imaging
• optical character recognition
• autonomous vehicles
• satellite imagery
• industrial inspection
• surveillance
• augmented reality

A computer-vision model may transform raw pixels into increasingly abstract representations and ultimately produce a classification, detection or other output.

Natural-language processing (NLP) concerns the interaction between computers and human language.

Traditional NLP included techniques for:

• text classification
• information extraction
• machine translation
• sentiment analysis
• speech-related applications
• question answering

Modern NLP is heavily influenced by large neural networks and foundation models.

AI systems can convert spoken language into text through automatic speech recognition (ASR).

They can also generate spoken language through text-to-speech (TTS) systems.

These technologies enable:

• voice assistants
• automated transcription
• accessibility tools
• call-centre automation
• language translation
• voice-controlled devices

Generative artificial intelligence, or generative AI, refers to AI systems capable of generating new content.

Depending on the model, generated content may include:

• text
• images
• audio
• music
• video
• computer code
• 3D assets
• scientific structures

Generative AI has become one of the most visible developments in the field.

Large language models and multimodal foundation models are among its most prominent examples.

A large language model (LLM) is a neural-network-based model trained on large quantities of text and, in many cases, additional types of data.

LLMs can perform tasks including:

• answering questions
• summarizing documents
• translating languages
• writing
• classification
• extracting information
• generating code
• reasoning through problems
• conversational interaction

Modern systems may also use external tools, retrieval systems, databases and software applications.

The transformer architecture became one of the most influential developments in modern AI.

Transformers use mechanisms including attention to model relationships between elements of an input sequence.

The architecture has been applied to:

• language
• images
• audio
• video
• multimodal systems
• biological sequences

Transformers helped make large-scale foundation models practical and became central to the modern generative-AI ecosystem. The OECD identifies transformer-based models among important developments in contemporary AI.

A foundation model is a large model trained on broad datasets and subsequently adapted for multiple applications.

Instead of building a separate model from scratch for every task, developers can use a foundation model as a general-purpose base.

Applications may include:

• text generation
• image generation
• coding
• translation
• question answering
• analysis
• search
• scientific research

This represents an important shift from task-specific AI toward more general-purpose AI systems.

Traditional AI systems often specialized in a single type of information.

Multimodal AI can work with multiple modalities, such as:

• text
• images
• audio
• video
• speech
• structured data

A multimodal system might, for example, analyze an image, read accompanying text and respond using natural language.

An AI agent is an AI system designed to pursue objectives through a sequence of actions rather than merely producing a single response.

Agentic systems may be able to:

1. interpret a goal
2. create a plan
3. use tools
4. execute actions
5. observe results
6. revise the plan
7. repeat the process

Modern AI agents can potentially interact with:

• web browsers
• databases
• APIs
• software applications
• coding environments
• enterprise systems
• physical devices

The concept of agentic AI is increasingly important as AI systems move from answering questions to performing tasks.

Artificial general intelligence (AGI) is a hypothetical or aspirational form of AI capable of learning, reasoning and applying knowledge across a very broad range of tasks at approximately human-level or beyond-human capability.

Unlike narrow AI, AGI would not be restricted to one particular domain.

For example, a hypothetical AGI might be able to:

• learn mathematics
• understand literature
• conduct scientific research
• operate software
• learn new physical skills
• reason about unfamiliar situations
• communicate naturally
• transfer knowledge between domains

There is no universally accepted test for AGI, and experts disagree about whether current systems should be considered early forms of general intelligence or sophisticated narrow/general-purpose tools.

Artificial superintelligence (ASI) refers to a hypothetical AI system whose general intellectual capabilities substantially exceed those of humans.

ASI remains a theoretical concept.

Debates concerning ASI involve questions about:

• technological control
• alignment
• safety
• governance
• economic power
• human autonomy
• existential risk

These questions remain subjects of active research and philosophical debate.

Although AI systems differ substantially, many modern systems follow a general pipeline.

1. Data collection

Data may come from:

• documents
• images
• audio
• video
• sensors
• databases
• human interactions
• scientific experiments

2. Data preparation

Data may be:

• cleaned
• labeled
• transformed
• filtered
• deduplicated
• normalized

3. Model development

Developers select an architecture and learning method.

4. Training

The model adjusts its parameters to reduce errors according to an objective or loss function.

5. Evaluation

The system is tested against datasets, benchmarks and real-world scenarios.

6. Deployment

The trained model is integrated into an application or service.

7. Inference

The deployed model receives new inputs and generates outputs.

8. Monitoring

Developers monitor performance, safety, security and changes in real-world data.

This last stage is particularly important because an AI system can perform well during development but behave differently after deployment.

Data is one of the fundamental resources of modern AI.

The quality of an AI system depends heavily on the quality and relevance of its training and operational data.

Problems in datasets can produce:

• bias
• inaccurate predictions
• unfair outcomes
• privacy violations
• poor generalization
• security vulnerabilities

The famous principle “garbage in, garbage out” applies particularly strongly to data-driven systems.

Two fundamental stages of many AI systems are training and inference.

Training

During training, the model learns parameters from data.

This process can require enormous computational resources for frontier models.

Inference

During inference, the trained model receives new input and produces an output.

For example:

Training: millions of examples → learn language patterns

Inference: user prompt → generate response

The computational requirements of training and inference differ, and optimization of both has become an important area of AI engineering.

Modern AI development depends heavily on specialized computing hardware.

Important technologies include:

• CPUs
• GPUs
• tensor-processing accelerators
• AI inference chips
• high-speed networking
• memory systems
• data centres

Large AI models may require thousands of interconnected processors during training.

Consequently, AI has become closely connected to semiconductor manufacturing, cloud computing and data-centre infrastructure.

Artificial intelligence has applications across a remarkably broad range of human activity.

Healthcare

AI is increasingly used for:

• medical image analysis
• drug discovery
• clinical decision support
• patient monitoring
• medical documentation
• genomics
• personalized medicine
• administrative automation

AI can assist medical professionals, but it does not eliminate the need for clinical judgment, medical responsibility or appropriate validation.

Scientific research

AI is increasingly used as a scientific tool.

Applications include:

• protein structure prediction
• materials discovery
• weather forecasting
• climate modeling
• astronomy
• particle physics
• genomics
• mathematical research

AI can accelerate parts of the scientific process by identifying patterns in datasets too large or complex for conventional analysis.

Education

AI can support:

• tutoring
• personalized learning
• language learning
• content generation
• automated feedback
• accessibility
• curriculum development
• educational administration

At the same time, educational institutions face questions about:

• academic integrity
• student dependence
• assessment design
• misinformation
• privacy
• unequal access

Stanford’s 2026 AI Index treats education as a major area of AI’s social impact and examines implications for learning and career readiness.

Finance

Financial institutions use AI for:

• fraud detection
• credit assessment
• risk modeling
• algorithmic trading
• customer service
• anti-money-laundering systems
• document processing

Financial AI systems must operate within regulatory and risk-management frameworks because errors can have significant economic consequences.

Manufacturing

Industrial AI is used for:

• predictive maintenance
• quality inspection
• robotics
• process optimization
• demand forecasting
• supply-chain management

Computer vision can automatically identify manufacturing defects at high speed.

Agriculture

AI can assist farmers through:

• crop monitoring
• disease detection
• yield prediction
• irrigation optimization
• soil analysis
• autonomous machinery
• weather forecasting

Combining AI with satellite imagery and agricultural sensors has created new possibilities for precision agriculture.

Transportation

AI contributes to:

• navigation
• traffic prediction
• logistics
• route optimization
• driver assistance
• autonomous vehicles
• fleet management

Autonomous driving remains a particularly challenging AI application because vehicles must operate safely in complex, unpredictable environments.

Robotics

AI-powered robotics combines computational intelligence with physical machines.

Applications include:

• industrial robots
• warehouse robots
• surgical robots
• agricultural robots
• drones
• service robots
• exploration robots

Robotics introduces additional challenges because AI systems must interact with the physical world.

Cybersecurity

AI can be used to detect:

• unusual network behaviour
• malware
• phishing
• fraud
• account compromise
• system anomalies

However, AI can also be used by attackers to automate and improve malicious activities, creating an ongoing technological arms race.

Entertainment

AI is increasingly used in:

• film production
• visual effects
• music
• video games
• recommendation systems
• animation
• script development
• personalized content

Generative AI has significantly expanded the ability to create synthetic media.

Search and recommendation

Search engines and recommendation platforms use AI to determine which information or content is most relevant to users.

Examples include recommendations for:

• films
• music
• products
• news
• websites
• social-media content

These systems can improve personalization but can also influence attention, public discourse and user behaviour.

Business

Businesses use AI for:

• customer support
• sales forecasting
• marketing
• document analysis
• software development
• recruitment
• accounting
• supply chains
• decision support
• process automation

AI is increasingly becoming a general-purpose business technology rather than a specialized research tool.

Government

Governments may use AI for:

• public-service delivery
• fraud detection
• traffic management
• disaster response
• tax administration
• translation
• document processing
• policy analysis

Government deployment raises particularly important questions concerning transparency, due process, accountability and citizens’ rights.

Defence and military applications

AI has military applications including:

• intelligence analysis
• logistics
• surveillance
• autonomous systems
• cyber operations
• simulation
• decision support
• target recognition

The use of AI in warfare is controversial, particularly where systems may influence decisions involving the use of lethal force.

Questions surrounding human control, accountability and autonomous weapons remain major international policy issues.

AI has the potential to provide substantial benefits.

Productivity

AI can automate repetitive cognitive tasks and assist workers with complex activities.

Accessibility

AI-powered speech recognition, translation, image description and assistive technologies can help people with disabilities.

Scientific discovery

AI can analyze enormous datasets and identify relationships that may be difficult for humans to detect.

Personalization

AI can tailor education, recommendations and services to individual users.

Automation

AI can perform tasks continuously and at high speed.

Decision support

AI can provide predictions and analyses that help humans make informed decisions.

However, AI outputs are not automatically correct. Effective systems require appropriate validation, human oversight and monitoring.

Limitations

Artificial intelligence has significant limitations.

AI systems may:

• make factual errors
• generate fabricated information
• misunderstand context
• reproduce biases
• fail unexpectedly
• be vulnerable to adversarial attacks
• perform poorly outside training conditions
• lack common-sense understanding
• produce confident but incorrect answers

One important characteristic of modern generative AI is that fluent language does not necessarily imply factual accuracy or human-like understanding.

Hallucination

In generative AI, a hallucination generally refers to an output that appears plausible but is unsupported, incorrect or fabricated.

Examples include:

• invented citations
• nonexistent books
• incorrect historical claims
• fabricated quotations
• false statistics
• imaginary people or organizations

Hallucination remains an important technical challenge for generative AI.

Methods such as retrieval-augmented generation, tool use, verification and structured evaluation can reduce—but not necessarily eliminate—the problem.

Bias and fairness

AI systems can reproduce or amplify biases present in their training data, design choices or deployment environments.

Potential sources include:

• incomplete datasets
• historical discrimination
• unequal representation
• biased labels
• flawed objectives
• measurement errors

AI bias can affect areas such as:

• employment
• lending
• insurance
• policing
• healthcare
• education
• facial recognition

Consequently, fairness, accountability and transparency have become major areas of AI research and policy.

Privacy

AI systems can process enormous quantities of personal information.

Potential privacy concerns include:

• collection of personal data
• surveillance
• facial recognition
• voice identification
• profiling
• inference of sensitive characteristics
• data leakage
• unauthorized use of training data

Privacy-preserving techniques include:

• data minimization
• anonymization
• differential privacy
• federated learning
• access controls
• encryption

Security

AI systems themselves can become targets of attack.

Potential threats include:

• prompt injection
• data poisoning
• adversarial examples
• model theft
• model manipulation
• jailbreaks
• privacy attacks
• supply-chain attacks

As AI becomes integrated into critical infrastructure, AI security becomes increasingly important.

Misinformation and deepfakes

Generative AI can produce highly convincing synthetic:

• photographs
• videos
• voices
• documents
• websites
• news-like content

Such material can be used to create deepfakes or other forms of synthetic misinformation.

Potential consequences include:

• political manipulation
• financial fraud
• identity theft
• reputational damage
• social unrest

The growing ability to generate realistic synthetic media has increased the importance of provenance, authentication and media literacy.

Copyright and intellectual property

AI has generated major debates concerning intellectual property.

Questions include:

• Can copyrighted material be used to train AI models?
• Who owns AI-generated content?
• Can an AI reproduce an artist’s distinctive style?
• Who is responsible when AI-generated content infringes copyright?
• How should creators be compensated?

Different jurisdictions have adopted different approaches, and the legal landscape continues to evolve.

Employment

AI may automate certain tasks while creating new ones.

Potentially affected occupations include work involving:

• data processing
• customer support
• translation
• administrative tasks
• content production
• software development
• analysis

At the same time, AI is generating demand for skills involving:

• AI engineering
• data science
• model evaluation
• cybersecurity
• AI governance
• human-AI interaction
• domain-specific AI implementation

The likely economic effect is therefore not simply “jobs disappear” but a more complicated combination of automation, augmentation, job transformation and new job creation.

Human-AI collaboration

One increasingly important model is human-AI collaboration.

Rather than replacing people completely, AI can function as:

human expertise + machine computation + human judgment

For example:

• doctors can use AI to examine medical images
• scientists can use AI to identify research hypotheses
• programmers can use AI coding assistants
• teachers can use AI to prepare learning materials
• lawyers can use AI for document analysis

The effectiveness of such systems depends heavily on how responsibilities are divided between humans and machines.

AI alignment

AI alignment concerns the challenge of ensuring that AI systems behave in accordance with intended human goals, values and constraints.

A system can be highly capable while still pursuing an objective in an undesirable way.

Alignment research therefore considers:

• goal specification
• preference learning
• human feedback
• interpretability
• robustness
• safety
• oversight
• controllability

Alignment becomes increasingly important as systems become more autonomous.

AI safety

AI safety is the broader field concerned with preventing AI systems from causing unacceptable harm.

It encompasses both immediate and long-term risks.

Areas include:

• reliability
• robustness
• misuse prevention
• cybersecurity
• evaluation
• human oversight
• model behaviour
• autonomous-system safety
• catastrophic-risk reduction

Explainability and interpretability

Some AI systems are difficult to understand internally.

This has led to research into explainable AI (XAI) and mechanistic interpretability.

The goal is to better understand:

• why a model produced an output
• which information influenced the decision
• what representations exist inside the model
• how errors arise
• whether a system is following intended behaviour

Interpretability is particularly important in high-stakes applications.

AI governance

As AI becomes increasingly influential, governments and international organizations are developing regulatory and governance frameworks.

AI governance addresses issues such as:

• safety
• privacy
• transparency
• accountability
• discrimination
• copyright
• competition
• national security
• consumer protection

The OECD, European Union and other institutions have developed frameworks intended to encourage responsible AI development and deployment. The OECD’s updated AI definition is explicitly designed to support policy and regulatory applications.

Open-source and closed AI

AI models may be released under different degrees of openness.

Some models provide:

• openly available weights
• source code
• training information
• documentation

Others are distributed as proprietary services with limited public access to their internal models.

The debate between open and closed AI concerns:

• innovation
• safety
• transparency
• competition
• security
• concentration of power
• accessibility

Neither approach automatically solves all AI-related problems.

AI is becoming an important general-purpose technology.

Its economic impact extends beyond technology companies into:

• manufacturing
• healthcare
• finance
• education
• agriculture
• retail
• logistics
• professional services

Stanford’s 2026 AI Index reports that AI capabilities, investment and adoption continue to expand, while emphasizing a widening gap between technological progress and society’s ability to evaluate and govern increasingly powerful systems.

The modern AI industry includes several interconnected layers:

Hardware

• semiconductor manufacturers
• GPU and accelerator companies
• memory manufacturers
• networking companies

Infrastructure

• cloud providers
• data centres
• energy systems
• networking infrastructure

Foundation models

Companies and research organizations develop large models for language, vision and multimodal applications.

Applications

Thousands of companies build specialized AI products on top of existing models.

This creates an emerging technology stack:

Energy → chips → data centres → foundation models → AI infrastructure → applications → users

Large-scale AI requires substantial computing infrastructure.

Environmental considerations include:

• electricity consumption
• water use for cooling
• semiconductor manufacturing
• data-centre construction
• electronic waste
• carbon emissions

At the same time, AI may potentially help reduce environmental impact through:

• energy optimization
• improved weather forecasting
• climate modeling
• smart grids
• industrial efficiency
• precision agriculture
• materials discovery

The environmental balance depends on how AI is developed, powered and deployed.

AI has significantly changed discussions about creativity.

Generative systems can produce:

• paintings
• illustrations
• music
• poetry
• stories
• video
• designs
• software

This raises a fundamental question:

Is creativity exclusively a human capability, or can machines participate in creative processes?

One practical view is that AI can function as a creative instrument, much like a camera, musical instrument or digital editing system.

Another view emphasizes differences between statistical generation and human experiences involving intention, consciousness, emotion and cultural context.

The question remains philosophically and scientifically unsettled.

Artificial intelligence should not be assumed to be equivalent to human intelligence.

Human intelligence involves a complex combination of:

• perception
• memory
• reasoning
• emotion
• social cognition
• embodied experience
• motivation
• cultural knowledge
• consciousness
• adaptation

AI systems can outperform humans in specific tasks while remaining unreliable in others.

For example, a system may perform exceptionally well on a difficult mathematical benchmark while failing at a seemingly ordinary real-world task requiring contextual understanding.

This phenomenon is sometimes described as brittleness or the jagged nature of AI capabilities.

Whether an AI system can be conscious is an unresolved philosophical and scientific question.

Current AI systems can produce sophisticated language and behaviour, but behavioural sophistication alone does not establish subjective experience.

Important unanswered questions include:

• What constitutes consciousness?
• Can consciousness exist in a machine?
• Is biological embodiment necessary?
• Can subjective experience be measured externally?
• Could an artificial system possess genuine self-awareness?

These questions lie at the intersection of AI, neuroscience, philosophy of mind and cognitive science.

The future development of AI may involve increasingly capable systems that combine:

• language
• vision
• reasoning
• memory
• planning
• tool use
• robotics
• real-time interaction
• scientific discovery

AI systems may increasingly move from passive software tools toward active digital agents capable of performing multi-step tasks.

At the same time, future development will depend not only on technical progress but also on:

• regulation
• public trust
• economic incentives
• energy availability
• hardware development
• international competition
• safety research
• social adaptation

Stanford’s 2026 AI Index emphasizes that measuring and understanding these developments has become increasingly important as AI capabilities spread throughout the economy and society.

Some of the most important concepts encountered in modern AI include:

Algorithm — a procedure for solving a problem.

Model — a computational representation learned or constructed to perform a task.

Dataset — a collection of data used for training, evaluation or operation.

Parameter — a numerical value learned by many machine-learning models.

Training — the process through which a model learns from data.

Inference — using a trained model to produce an output.

Neural network — a computational model consisting of interconnected layers or units.

Deep learning — machine learning based primarily on multi-layer neural networks.

Transformer — a neural architecture based heavily on attention mechanisms.

Foundation model — a broadly trained model that can support many downstream applications.

Generative AI — AI capable of producing new content.

LLM — large language model.

AGI — artificial general intelligence.

AI agent — an AI system capable of pursuing goals through sequences of actions.

Hallucination — an incorrect or unsupported generated output.

Prompt — an instruction or input supplied to a generative AI system.

Fine-tuning — adapting a pretrained model to a particular task or domain.

RAG — retrieval-augmented generation, in which a model retrieves external information before generating an answer.

For many people, AI is no longer an abstract laboratory technology.

People encounter AI through:

• smartphone assistants
• search engines
• social-media recommendations
• navigation apps
• online shopping
• spam filters
• translation tools
• camera software
• streaming services
• banking security
• customer-service chatbots
• generative AI applications

In many cases, users interact with AI without explicitly realizing that AI is involved.

As AI becomes widespread, AI literacy is increasingly important.

AI literacy involves understanding:

• what AI can do
• what AI cannot reliably do
• how AI systems learn
• how to evaluate AI-generated information
• how to protect personal data
• how to identify synthetic media
• how to use AI responsibly
• when human expertise is necessary

AI literacy is becoming analogous to digital literacy: a basic competence needed to navigate an increasingly technology-mediated society.

Artificial intelligence represents a fundamental change in the relationship between humans and machines.

Earlier computers primarily followed explicit instructions.

Modern AI systems can increasingly:

learn → predict → generate → reason → plan → act

This transition has enormous implications.

AI may become one of the most powerful tools ever developed for extending human intellectual capabilities. It may help humans solve scientific problems, improve medicine, personalize education, automate dangerous work and create new forms of art and communication.

But the same technology can also amplify misinformation, surveillance, discrimination, cyberattacks and other forms of harm.

The central question is therefore no longer simply:

“Can machines become intelligent?”

It has become a broader question:

“How should humanity build, use, govern and live alongside increasingly capable intelligent machines?”

Artificial intelligence is one of the most consequential technological and scientific developments of the modern era.

Born as an ambitious research question in the middle of the twentieth century, AI has evolved from symbolic reasoning programs and expert systems into sophisticated machine-learning systems capable of processing language, images, audio and other forms of information.

The emergence of deep learning, transformers, foundation models, generative AI and increasingly agentic systems has dramatically expanded the range of tasks machines can perform.

Yet AI is not simply a story of technological capability. It is simultaneously a story about human intelligence, economics, creativity, science, work, privacy, governance, security and philosophy.

Its ultimate significance will depend not merely on how intelligent machines become, but on how wisely humans choose to design, deploy and govern them.

In that sense, artificial intelligence is not only a technology for building intelligent machines. It is also a technology forcing humanity to reconsider one of its oldest questions:

What does it actually mean to be intelligent?