The AI Dictionary that you didn't realise you needed

Most AI glossaries explain how machines work. This one explains what humans need to know. Every term here is defined through a single lens: does this help you stay the author of your own decisions?

A

Absent User

The person affected by a decision but not present when it is made. Naming the absent user keeps systems humane and prevents harm by default.

Adaptive Intelligence

The combined strength of human judgement and machine speed working together, with humans retaining responsibility.

Agency

The capacity to shape circumstances rather than be shaped by systems, habits, or automation.

Agentic AI

AI systems that take autonomous action toward goals with limited supervision. As agency increases, human guardrails become essential.

Algorithm

A set of instructions that turns inputs into outputs. In modern systems, algorithms increasingly influence decisions once made by people.

Algorithmic Drift

The gradual transfer of judgement from humans to automated systems through defaults and convenience.

Alignment (AI Alignment / Value Alignment)

Ensuring AI systems pursue outcomes humans actually want, rather than optimising narrowly in harmful ways.

Anthropomorphism

Treating AI systems as if they think or intend, leading to misplaced trust.

Artificial General Intelligence (AGI)

A hypothetical form of AI capable of learning and reasoning across domains at or beyond human level.

Artificial Intelligence (AI)

Systems that learn patterns from data to predict, classify, or generate outputs. Powerful tools, not moral agents.

Attention (Self-Attention)

The mechanism that allows AI to weight which inputs matter most. Humans do this naturally; machines must be trained.

Attention Economy

The competition to capture and monetise human focus, often at the expense of depth or wellbeing.

Augmented Mindset

The disposition to treat AI as an extension of capability, not a replacement for judgement.

Authorship

Taking visible responsibility for decisions and meaning rather than deferring to systems.

Automation

Delegating tasks to systems that operate without continuous human input. Valuable only with accountability.

B

Bearings

Your moral and cognitive orientation when speed increases and systems accelerate.

Benchmark

A test used to compare system performance. What gets measured gets optimised.

Bias (Algorithmic Bias)

Systematic distortion caused by skewed data or assumptions.

Black Box

A system whose internal reasoning cannot be easily inspected or explained.

Boundary Erosion

The quiet expansion of what systems decide without explicit permission.

C

Cadence

A steady rhythm of work and reflection that replaces urgency with progress.

Capability Decay

The slow loss of relevance when skills are not practised or updated.

Chain-of-Thought

Prompting AI to show intermediate reasoning steps. Helpful for inspection, not proof of understanding.

Change Readiness

The ability to move without panic when conditions shift.

Cognitive Exoskeleton

Tools that extend human thinking without replacing judgement.

Cognitive Offloading

Delegating mental effort to tools. Useful when deliberate, risky when unconscious.

Compute

The processing power required to train and run AI. Access shapes who can build and scale.

Confabulation (Hallucination)

When AI generates confident but false outputs because it predicts plausibility, not truth.

Context Window

The amount of information an AI system can consider at once.

Curiosity Dividend

The compounding return from asking better questions over time.

Curiosity Loop

See → Seek → Shift. A repeatable learning cycle.

D

Dark Pattern

Design that nudges people into choices they would not make with full clarity.

Data Gravity

The tendency for power, talent, and decisions to cluster around data holders.

Decision Debt

Choices made without reflection that later require costly correction.

Deep Learning

Layered pattern-learning systems. Powerful at recognition, weak at meaning.

Defaults

Pre-set choices that run unless humans intervene.

Deskilling

Loss of human capability when practice is replaced by automation.

Design

Conscious authorship. Choosing direction instead of accepting drift.

Design Debt

The cost accumulated when speed overrides care.

Digital Drift

The quiet surrender of time and attention to algorithmic curation.

Diffusion Model

AI that generates images by refining noise into form.

Drift

Unconscious compliance with systems or habits that decide for you.

Drift Debt

The accumulated cost of uncorrected small shortcuts.

Drift vs Design Matrix

A lens for diagnosing whether action is deliberate or default.

E

Edge (Edge Computing)

Processing done locally rather than remotely. Keeps data closer to people.

Embedding

A numerical representation of meaning. Proximity is not understanding.

Emergent Behaviour

Capabilities that appear at scale without being explicitly designed.

Empathy

Understanding others well enough to act with their reality in mind.

Empathy Gap

The distance between what systems optimise and what humans feel.

Epistemic Humility

Knowing the limits of what you or any system can know.

Ethical Pre-Mortem

Imagining how harm could occur, then preventing it.

Explainable AI (XAI)

Making AI decisions understandable to humans.

Exponential Change

Acceleration that feels slow until it overwhelms existing systems.

F

February 29 Cycle

Spot, Sense, Shift, Shape. An adaptability loop for disruption.

Feedback Loop

When outputs become inputs. Loops can amplify wisdom or error.

Fine-Tuning

Adapting a general model to a specific context. Values are encoded here.

Five Loops

A weekly rhythm sustaining agency through learning and leverage.

Fragility Index

How dependent a system is on one person or assumption.

Friction (Productive Friction)

Deliberate slowness that prevents harm or preserves dignity.

Frontier Model

The most capable AI systems available at a given moment.

G

Generative AI

AI systems that create new content from learned patterns.

Global Adaptability

Thriving across cultures and constraints.

Global Mirror

A trusted local check on intent and impact.

Grounding

Connecting outputs to verifiable sources.

Guardrails

Non-negotiable limits protecting safety and accountability.

H

Half-Life of Skills

The time it takes for half of what you know to become outdated.

Hand–Head–Hours

What stayed human, what was learned, and where time went.

Hidden Labour

Invisible human work inside AI systems.

Human Dividend

Value created when judgement and care remain human.

Human-in-the-Loop

Humans retaining accountability in automated systems.

Human Override

The ability to interrupt or halt automated action.

I

Illusion of Understanding

False confidence created by fluent but shallow output.

Inference

Using a trained system in real-world conditions.

Integrity Loop

Drawing the line, building visibly, repairing publicly.

Integrity Premium

Trust earned through lived ethics.

Intent Gap

The distance between what was meant and what was optimised.

Interpretability

How well humans can understand system reasoning.

J

Jailbreaking

Attempts to bypass AI safeguards.

Judgement

Deciding what matters and why. Begins where metrics end.

Judgement Debt

Delegating decisions without building evaluative capacity.

K

Knowledge Cutoff

The point beyond which an AI has no training data.

Knowledge Debt

Repeated action without increasing understanding.

Knowledge Work

Work centred on thinking and decision-making.

L

Large Language Model (LLM)

AI trained to predict the next unit of text, enabling language generation.

Latent Space

The internal map AI uses to organise concepts.

Learning Velocity

The speed at which insight becomes behaviour.

Legibility

How readable a system’s logic is to those affected.

Leverage Ladder

I do it → We do it → Systems do it, with stewardship.

Local Context

Situational factors generic systems often miss.

M

Machine Partnering

Working with AI without surrendering authorship.

Meaning Debt

Exhaustion when output outpaces purpose.

Meta-Skill

A skill that strengthens other skills.

Middle Seat

Where most people work, between power and precarity.

Model Collapse

Degradation from training AI on AI-generated content.

Moral Speed Limit

The point where speed outpaces ethics.

Multimodal

AI that processes multiple input types.

N

Negative Space

What a system omits.

Never Twice

Fix the system, not the person.

Noise

Information that obscures rather than clarifies.

O

Opacity

Inability to see how a system reasons.

Optimisation Target

What a system is tuned to maximise.

Overfitting

Expertise without adaptability.

P

Parameter

A learned internal value shaping model behaviour.

Personalisation Trap

When tailoring removes challenge and growth.

Platform Shift

A change in where value is created.

Post-Training

Work after initial training where values are embedded.

Predictive Model

Forecasting from patterns, not causes.

Principled Innovation

Progress that protects what matters.

Principled Pause

A moment to check alignment before acting.

Proof of Humanity

Evidence that judgement and care remain human.

Prompt

Input that shapes AI output.

Prompt Engineering

Crafting inputs to guide system behaviour.

R

Reality Advantage

Learning from evidence before rivals.

Red Teaming

Deliberately probing systems for failure modes.

Reflection Debt

Lessons not learned due to speed.

Regression to the Mean

Extremes tend to normalise over time.

Repair Note

A public record of what broke and was fixed.

Responsible AI

Practised commitment to safety and accountability.

Retrieval-Augmented Generation (RAG)

Grounding AI outputs in retrieved information.

RLHF

Training AI using human preference signals.

S

Scaffolds, Not Cages

Support structures that do not constrain thought.

Signal-to-Noise Ratio

Separating insight from distraction.

Sludge

Friction designed to obstruct.

Solutionism

Assuming every problem has a technical fix.

Stochastic Parrot

Fluency without understanding.

Stillness Paradox

Adaptability rooted in what does not change.

SuperSkills

Meta-capabilities that preserve human agency.

SuperSkills Ladder

The evolution from survival to human mastery.

Synthetic Data

Artificial data used when real data is limited.

Synthetic Seniority

The illusion of experience produced by AI answers.

System Prompt

Hidden instructions shaping behaviour.

T

Task Unbundling

Breaking roles into automatable and human tasks.

Temperature

Controls predictability versus creativity in outputs.

Technical Debt

Shortcuts that create long-term cost.

Token

The basic unit of text processed by language models.

Training Data

The examples a system learns from.

Transfer Learning

Applying learned patterns across domains.

Transformer

The architecture behind most modern language models.

Transparency

Ability to inspect system operation.

Trust Flywheel

Ownership and openness that compound trust.

Trust Scar

Lasting damage from breach.

U

Uncertainty Quantification

Measuring how confident a model is.

Undrift

Reclaiming authorship from automation.

Unlearning

Removing specific knowledge from a system.

V

Validation

Testing performance on unfamiliar data.

Vector

A numerical position in meaning-space.

W

Watermarking

Marking AI-generated content for provenance.

Weight

A learned value shaping outputs.

Wisdom Loop

Turning data into insight, then action.

Work-as-Design

Treating work as improvable, not fixed.

World Model

An internal representation of relationships.

“Who wrote your last decision?”

A diagnostic for authorship.

Z

Zero-Shot

Performing tasks without examples. Impressive, but fragile.

If this is what you are grappling with in your organisation, the fastest starting point is the Sprint.

Workforce Readiness Sprint

Kick off a Workshop

Win hearts and minds with a Keynote

or Hire Rahim Hirji on Retainer