Artificial intelligence (AI) is the field of computer science that studies and builds computational systems able to perform tasks associated with human intelligence. These tasks include reasoning, learning from experience, recognizing patterns, understanding and producing language, perceiving the environment, and making decisions. The term refers both to the academic discipline and to the systems it produces. These range from narrow programs built for a single task, such as playing chess or classifying images, to general-purpose models that handle many kinds of problems. Modern AI relies mainly on machine learning, and especially on deep learning, which uses layered artificial neural networks trained on large datasets.
Definitions and goals
AI has no single accepted definition. Some definitions stress behavior: a system is intelligent if it acts in ways that would be called intelligent if a human did the same thing. In 1950, Alan Turing proposed what became known as the Turing test. In this test, a machine counts as intelligent if, in text conversation, an interrogator cannot reliably tell it apart from a human. Other definitions focus on rational agents, meaning systems that perceive their surroundings and act to achieve goals. Researchers often distinguish narrow AI, which is built for specific tasks, from artificial general intelligence (AGI), a hypothetical system that could match or exceed human ability across nearly all cognitive tasks. Whether machines can truly "think" or have understanding is still debated in philosophy and cognitive science.
History
The idea of artificial beings appears in ancient myths and mechanical automata. The modern field began after the invention of the electronic computer. John McCarthy coined the name "artificial intelligence" in his 1955 proposal for the Dartmouth Summer Research Project, which took place in 1956 and is usually treated as the founding event of the field. McCarthy organized it with Marvin Minsky, Nathaniel Rochester and Claude Shannon. The proposal rested on the conjecture that every aspect of learning or intelligence could be described precisely enough for a machine to simulate it.
Early research focused on symbolic methods, which manipulated explicit rules and representations based on formal logic. Notable programs included the Logic Theorist, which proved mathematical theorems, the General Problem Solver, and the chatbot ELIZA (1966). Researchers had underestimated how hard their goals were. Funding fell sharply in the 1970s and again in the late 1980s, and these downturns are known as AI winters. In between, expert systems had commercial success in the 1980s. These systems encoded the knowledge of human specialists as large sets of if–then rules.
From the 1990s onward, statistical approaches based on probability and statistics gradually replaced hand-written rules. Three things drove the shift: more powerful hardware, large datasets, and rigorous mathematical methods. In 2012 the deep convolutional network AlexNet, developed by Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton, won the ImageNet image recognition challenge by a wide margin. It was trained on graphics processing units (GPUs), and its success started the deep learning era. In March 2016, DeepMind's AlphaGo beat the top Go player Lee Sedol four games to one. Many researchers had expected that milestone to be decades away.
The transformer architecture, introduced in 2017, made large language models (LLMs) possible. Public releases of chatbots such as ChatGPT and Claude in the 2020s led to a surge of investment often called the AI boom. In 2024, two Nobel Prizes recognized AI-related work. John Hopfield and Geoffrey Hinton received the Physics prize for foundational work on machine learning with artificial neural networks. Demis Hassabis and John Jumper shared half of the Chemistry prize for AlphaFold, which predicts the three-dimensional structure of a protein from its amino-acid sequence.
Approaches and techniques
- Symbolic AI represents knowledge with explicit symbols and rules and draws conclusions by logical inference. Its reasoning is transparent, but it struggles with ambiguous or noisy real-world data.
- Machine learning lets a system infer patterns from data instead of following hand-written rules. There are three main forms. In supervised learning, the system learns from labeled examples. In unsupervised learning, it finds structure in unlabeled data. In reinforcement learning, it learns by trial and error from rewards.
- Neural networks are loosely inspired by neurons in the brain. They consist of layers of simple units whose connection weights are adjusted during training, usually by backpropagation. Deep networks with many layers dominate modern applications.
- Search and optimization methods explore spaces of possible solutions. They are used in planning, game playing and scheduling.
- Probabilistic models, such as Bayesian networks, represent uncertainty explicitly.
Subfields and applications
Major subfields include natural language processing, computer vision, speech recognition, robotics, knowledge representation and automated planning. Generative artificial intelligence produces new text, images, audio, video and computer code. AI is used in web search, recommendation systems, machine translation, fraud detection, and driver-assistance and self-driving vehicles. In medicine, it supports medical imaging and clinical decision-making. In science, it is used for protein structure prediction, materials discovery and data analysis. Training and running large models depends on specialized semiconductor chips, data centers and cloud computing infrastructure.
Limitations and risks
Current AI systems have several known limitations. Learned models are often hard to interpret. They can absorb biases present in their training data. They can also fail unpredictably when given inputs unlike anything in their training. Language models sometimes produce fluent but false statements, a problem known as AI hallucination. Wider concerns include:
- privacy and surveillance;
- copyright in training data;
- misinformation and synthetic media, including deepfakes;
- the effects of automation on jobs;
- misuse in cybersecurity attacks;
- the energy demand of large data centers.
Some researchers, including several pioneers of the field, have warned about risks from increasingly capable systems. These warnings have spurred work on AI alignment, which aims to make AI systems act in line with human intentions and values. These issues are central to the ethics of AI.
Governance
Governments and international bodies have developed laws and guidelines for AI. The European Union's AI Act is widely described as the first comprehensive legal framework for AI. It entered into force on 1 August 2024 and takes effect in stages. Bans on certain practices, such as social scoring, applied from February 2025, and rules for general-purpose AI models from August 2025. Most remaining provisions, including transparency duties such as labeling AI-generated content, applied from 2 August 2026. However, a 2026 amendment package moved the main obligations for high-risk systems to dates in 2027 and 2028. Other jurisdictions use a mix of binding laws, voluntary frameworks and technical standards.
References
- Dartmouth and the Dawn of AIadmissions.dartmouth.edu
- Artificial Intelligence (AI) Coined at Dartmouthhome.dartmouth.edu
- Deep Learning in Sciencearxiv.org
- History of artificial intelligenceen.wikipedia.org
- Internet of Intelligence: A Survey on the Enabling Technologies, Applications, and Challengesarxiv.org
- The History of Artificial Intelligence: A Timeline from Turing to Todayswisscyberinstitute.com
- Artificial Intelligence awarded two Nobel Prizes for innovations that will shape the future of medicinenature.com
- The Nobel Prize in Chemistry 2024 - Popular information - NobelPrize.orgnobelprize.org
- Mathematical theory of deep learningarxiv.org
- AlexNet: The Deep Learning Breakthrough That Reshaped Google’s AI Strategydejan.ai
- AlphaGo versus Lee Sedolen.wikipedia.org
- AI Actdigital-strategy.ec.europa.eu
- EU AI Act Enforcement: August 2026 Rules and Deadlinesbrightdefense.com
- A comprehensive EU AI Act Summary [August 2026 update] - SIGsoftwareimprovementgroup.com
- EU AI Act Timeline 2026: Aug 2 Milestone & Next Deadlinesalicelabs.ai