Artificial intelligence

Artificial intelligence is an umbrella term rather than one single technique, and boundaries between AI and other software are not universally fixed. The OECD describes an AI system as a machine-based system that, in light of explicit or implicit objectives, infers from inputs how to generate outputs—such as predictions, content, recommendations or decisions—that may influence physical or virtual environments. Systems vary in their autonomy and ability to adapt after deployment.

AI encompasses both knowledge-based approaches, which use representations or rules, and machine-learning methods that derive patterns from data. Machine learning includes deep-learning approaches built around multilayer neural networks. A system may classify an image, transcribe speech, translate text, recommend an item or help control a robot. Generative AI is a group of systems that produces new outputs, including text, images or other media, in response to inputs. These capabilities are task-specific: success on one task does not by itself establish broad human-like understanding or competence.

AI is used across areas such as science, health, education, finance, industry and public services. Some tools assist people with analysis or routine tasks; others support or automate parts of a decision process. Their behavior depends on design choices, training or reference data, objectives, and the context in which they are used. The Stanford Institute for Human-Centered AI's AI Index tracks research, technical performance, economic effects and societal impacts, reflecting the field's rapid development and breadth.

AI systems can produce incorrect, incomplete or misleading outputs, and their performance can vary across populations and settings. Data and design choices can reproduce or amplify harmful biases; automation may also affect privacy, security, accountability and access to opportunities. Risks depend on the system and its use, so an evaluation appropriate for a low-stakes application may not be sufficient for a consequential one. Human review, testing and clear responsibility are important safeguards, though oversight alone cannot remove every risk.

Responsible AI requires attention throughout design, development, deployment and use. NIST's voluntary AI Risk Management Framework organizes risk work around governing, mapping, measuring and managing risks, and identifies characteristics of trustworthy systems such as validity, reliability, safety, security, transparency, explainability, privacy and fairness. UNESCO's 2021 Recommendation on the Ethics of Artificial Intelligence places human rights and human dignity at its core and emphasizes principles including transparency, fairness, environmental sustainability and human oversight. These frameworks offer guidance; they do not make every AI system trustworthy automatically, and implementation must account for context and trade-offs.