Reference

Glossary of intelligence terms

A growing reference of the vocabulary you need to think clearly about AGI, AI, neuroscience, and cognitive science. Each entry links to the hub where it is explored in depth.

AGI (Artificial General Intelligence): A hypothesised AI system that can match or exceed competent human adults at the full range of cognitive tasks, transferring skills across novel domains.
AI Agent: A system that perceives an environment, plans, takes actions, and pursues goals over time, typically with tool use and memory.
AI Alignment: The research problem of ensuring advanced AI systems pursue goals and values that are genuinely beneficial to humans.
AI Safety: The broader field studying how to develop and deploy AI without unacceptable risks - including alignment, robustness, interpretability, and governance.
Attention (mechanism): A neural network operation that learns to weight which inputs matter most for a given output. The core of the transformer architecture.
Augmented Intelligence: A design philosophy that uses AI to amplify human judgement rather than replace it.
Backpropagation: The algorithm for computing gradients in neural networks, enabling them to learn from errors.
Brain-Computer Interface (BCI): Hardware that establishes a direct communication pathway between the brain and an external device.
Chain-of-Thought: A prompting and training technique where a model produces intermediate reasoning steps before its final answer.
Cognitive Architecture: An integrated framework specifying the components and processes of a mind - memory, attention, learning, control.
Cognitive Capital: The accumulated productive cognitive capacity of a society - human skills, machine reasoning, and the tools that augment both.
Cognitive Copilot: An always-available AI assistant designed to support knowledge work and decision-making.
Cognitive Neuroscience: The branch of neuroscience that studies how brain activity gives rise to perception, memory, language, and reasoning.
Cognitive Science: The interdisciplinary study of mind and intelligence across psychology, neuroscience, linguistics, philosophy, and CS.
Consciousness: Subjective experience - the felt quality of being a perceiving, thinking system.
Crystallized Intelligence: Accumulated knowledge and verbal skill built through experience and learning; pairs with fluid intelligence in Cattell's model.
Deep Learning: Machine learning using multi-layer neural networks that learn hierarchical representations.
Embodiment: The view that intelligence is shaped by, and partly constituted by, having a body that interacts with the world.
Emergent Capability: A behaviour that appears in larger models but is absent or weak in smaller ones, often unpredictably.
Executive Function: Higher-order cognitive control including planning, inhibition, working memory, and task switching.
Fine-tuning: Adapting a pre-trained model to a specific task or domain using additional, targeted training.
Fluid Intelligence: The ability to reason about novel problems independently of prior knowledge; one half of the Cattell-Horn-Carroll model of intelligence.
Foundation Model: A large model pre-trained on broad data and intended to be adapted to many downstream tasks.
Frontier AI: Highly capable general-purpose AI models at the cutting edge of capability; the regulatory category targeted by AI Safety Institutes.
General-Purpose AI (GPAI): The EU AI Act's legal term for models with broad capability across tasks; carries specific transparency and risk obligations.
Generative AI: Models that produce new content - text, images, audio, video, code - rather than only classifying.
Governance (AI / AGI): Laws, standards, and institutions that shape how AI and AGI are built and deployed - from the EU AI Act to the Frontier Model Forum.
Hallucination: Confident but incorrect output from a language model, typically presenting fiction as fact.
Human-in-the-Loop: A system design where humans review, approve, or correct model outputs before action is taken.
Hybrid Intelligence: Teams in which humans and AI agents share tasks, context, and accountability to outperform either alone.
Intelligence Economy: The economic regime in which reasoning, expertise, and judgment are increasingly produced by machines as well as people.
Large Language Model (LLM): A neural network trained on large text corpora to predict tokens, capable of fluent language and emergent reasoning.
Machine Learning (ML): AI in which systems learn patterns from data rather than being explicitly programmed.
Machine Reasoning: The capacity of an AI system to plan, decompose problems, and verify intermediate steps - the defining research story of 2024-2026.
Metacognition: Thinking about thinking - monitoring and regulating one's own cognitive processes.
Multimodal Model: A model that processes and generates across multiple modalities - text, image, audio, video.
Neural Decoding: The machine-learning problem of translating brain signals into intent, language, or motor commands; the foundation of modern BCIs.
Neural Network: A computational system loosely inspired by biological neurons, learning by adjusting weighted connections.
Neuroplasticity: The brain's capacity to change its structure and function in response to experience.
Neurotechnology: Engineering tools that record, stimulate, or interface with the nervous system - from EEG to high-density cortical implants.
Reasoning Model: A model trained with explicit chain-of-thought and reinforcement learning to improve multi-step problem solving.
Reinforcement Learning (RL): Learning by interaction with an environment, guided by reward signals.
Responsible AI: Operational standards for fairness, privacy, security, and accountability in deployed AI systems.
RLHF: Reinforcement Learning from Human Feedback - training models using human preference data.
Scaling Laws: Empirical regularities relating model performance to compute, parameters, and data.
Self-Supervised Learning: Training in which the learning signal is generated from the data itself, without external labels.
Superintelligence (ASI): A hypothetical system whose general cognitive performance substantially exceeds the best human minds.
Synthetic Mind: A hypothetical fully integrated artificial cognitive system - more than a language model, less defined than AGI.
Token: The basic unit a language model reads and writes - usually a sub-word fragment.
Transfer Learning: Applying knowledge gained in one task or domain to a different, related one.
Transformer: A neural network architecture based on self-attention; the basis of modern LLMs and multimodal models.
Trustworthy AI: Practitioner-facing label for AI systems built to the standards of frameworks like NIST AI RMF and ISO/IEC 42001.
Working Memory: A limited-capacity cognitive workspace where information is actively held and manipulated.