Learn AI / A–Z
A–Z of AI terms
All 307 published terms, alphabetically. Acronyms are filed under the full term, but search finds them either way.
A
- Access control
- Accessibility
- Accountability
- Accuracy
- Activation function
- Adam optimiser
- Adversarial attack
- Adversarial example
- Agentic AI
- AI agent
- AI assistant
- AI audit and assurance
- AI divide
- AI ethics
- AI governance
- AI safety
- AI security
- Algorithm
- Algorithmic bias
- Alignment
- Annotation and data labelling
- Anthropomorphism
- API (API)
- Artificial general intelligence (AGI)
- Artificial intelligence (AI)
- Artificial neuron
- Attention
- Attention score
- Audio and speech generation
- Authentication and authorisation
- Automation
- Automation bias
- Autoregressive model
B
C
- Calibration
- Capacity-building
- Chain rule
- Chatbot
- Checkpoint
- Chunking
- Citation (in AI answers)
- Class imbalance
- Classification
- Cloud computing
- Clustering
- Code generation
- Completion
- Compute
- Computer use
- Computer vision (CV)
- Confidentiality
- Confusion matrix
- Connector
- Consensus
- Consent
- Content moderation
- Context
- Context window
- Convergence
- Convolutional neural network (CNN)
- Copilot
- Corpus
- Cosine similarity
- Cost per token
- Cross-validation
D
- Data
- Data cleaning
- Data distribution
- Data exfiltration
- Data governance
- Data leakage
- Data poisoning
- Data security
- Data sovereignty and localisation
- Dataset
- Deep learning
- Deepfake
- Deployment
- Deterministic output
- Diffusion model
- Digital divide
- Digital public infrastructure and digital public goods (DPI)
- Distillation
- Distribution shift and data drift
- Document symbol
- Dropout
- Dual use
E
F
G
H
I
- Image captioning
- Image classification
- Image generation
- Image segmentation
- Impact assessment
- In-context learning
- Inclusion
- Independent International Scientific Panel on AI
- Indexing
- Indirect prompt injection
- Inference
- Input, hidden and output layers
- Instruction tuning
- Intent detection
- International standard
- Interoperability
- Interpretability
J
K
L
M
- Machine consciousness
- Machine learning (ML)
- Machine translation
- Mandate
- Meaningful human control
- Member State
- Memorisation
- Metadata
- Misinformation and disinformation
- Missing data
- Model
- Model Context Protocol (MCP)
- Model extraction and inversion
- Model hosting
- Model version
- Model weights (as a released artefact)
- Momentum
- Multi-agent system
- Multi-head attention
- Multi-stakeholder governance
- Multilateralism
- Multimodal model
N
O
P
Q
R
- Ranking and reranking
- Rate limit
- Reasoning model
- Recall
- Recorded vote
- Recurrent neural network (RNN)
- Red teaming
- Regression
- Regularisation
- Regulation
- Regulatory sandbox
- Reinforcement learning (RL)
- Reinforcement learning from human feedback (RLHF)
- Reliability
- Resolution
- Responsible AI
- Retrieval
- Retrieval-augmented generation (RAG)
- Risk-based regulation
- Robustness
- Rule-based system
S
- Sampling (choosing the next token)
- Sampling (of data)
- Sandboxing
- Scalability
- Scaling laws
- Secretariat
- Secrets management
- Security Council (UNSC)
- Self-attention
- Self-supervised learning
- Semantic search
- Semantic similarity
- Semi-supervised learning
- Sentiment analysis
- Shadow AI
- Short-term memory (in AI systems)
- Softmax
- Software
- Speech-to-text
- Structured data
- Summarisation
- Supervised fine-tuning (SFT)
- Supervised learning
- Sustainable Development Goals (SDGs)
- Synthetic data
- Synthetic media
- System prompt