-01.png)
ABOUT AIECB · KNOWLEDGE HUB
Knowledge Hub
ON THIS PAGE
Glossary
Twenty terms UAICS builds on — written so a parent or teacher with no technical background can follow
every one.
Showing the first 20 entries. The full glossary grows alongside the standard itself.
Media Literacy
Evaluating sources and recognising manipulation. In an era
when AI mass-produces convincing content, it has moved
from an enrichment to basic protection.
Digital Footprint
The traceable record left by online activity. Managing
digital identity and privacy settings is formally assessed
content in the UAICS citizenship domain.
Metacognition
Awareness and regulation of one’s own thinking. “Is this my
idea, or the AI’s?” is the defining metacognitive question of
the era.
Cognitive Load
Working memory processes only so much at once.
Judging what to think through oneself and what to hand to
a tool begins with knowing one’s own load — part of
UAICS’s taught-not-tested content.
Human Agency
The retained capacity and right to understand, question
and overrule an AI’s suggestion. The heart of UNESCO’s
framework — and the permanent question: who, in the
end, made this decision?
Word Embedding
Words mapped to numerical vectors that reflect closeness
of meaning — the foundation of natural language
processing: similar meanings, similar numbers.
Feature Vector
An object’s important attributes turned into a list of
numbers, so computers can calculate and compare.
“Turning concepts into numbers” is a key step in
understanding how AI sees the world.
Supervised Learning
Training on examples that carry correct answers — one of
machine learning’s three paradigms, alongside
unsupervised learning (finding hidden structure) and
reinforcement learning (learning from trial and feedback).
Model Card
A document recording a model’s purpose, training data,
limits and proper scope — an industry accountability
practice. UAICS has children write Model Cards for models
they train themselves: owning what one has made.
Abstraction
Judging which details matter and which can be dropped.
Selecting features for a model and deciding what context
to give an AI draw on exactly this one capability.
Computational Thinking
Decomposing problems, recognising patterns, abstracting,
designing steps. It is not coding; in the AI era it is the
underlying craft of making a problem explicit.
Prompt Engineering
The craft of stating a task so AI can act on it: goal, context,
constraints. In advanced form, decomposing a large task
into staged prompts with checking between — which is
structured thinking in training.
Hallucination
Fluent, plausible output that is wrong or invented. The
cause lies in probabilistic continuation itself — and
understanding the cause serves a child far better than
memorising “AI lies.”
Generative AI
AI that produces new content — text, images, sound —
unlike classification AI, which only judges. It generates text
by probabilistic continuation, word by word; understanding
this explains the particular kind of error it makes.
Backpropagation
The method by which the signal of “how wrong we were”
travels backwards through a network, adjusting
parameters layer by layer. Think of it as the model’s
mechanism for learning from its own errors.
Neural Network
A model built from many layers of simple units, each
transforming input into more abstract representations.
UAICS teaches its working intuition at the conceptual level;
no mathematics is required.
Algorithmic Bias
When training data under-represents or skews against
certain groups, a model performs systematically worse for
them — speech recognition faltering on strong accents, for
example. Bias mostly enters through data, not machine
malice.
Training Data & Labels
The examples used to teach a model; labels are each
example’s correct answer. Data quantity and quality decide
performance — “rubbish in, rubbish out” begins here as a
judgement children can own.
Classifier
A machine-learning model that sorts inputs into categories
— cat or dog, for instance. Training one and inspecting
where it errs is the best possible introduction to the idea
that AI makes mistakes, and that mistakes can be found.
Machine Learning
A way for computers to find patterns from many examples,
rather than following rules written one by one. A child’s
first UAICS practical is exactly this: feeding examples to
train a small model by hand.
National & International Framework Digests
The three frameworks UAICS integrates — what each one covers, its core claim, and exactly where it
shows up in UAICS.
UNESCO — AI Competency Framework for Students
PUBLISHED 2024
STRUCTURE
Four dimensions (human-centred mindset, ethics of AI, AI techniques and applications, AI system design) by three
levels (Understand, Apply, Create) — twelve competency blocks for school-age learners, deliberately unbanded by
age.
CORE CLAIM
Student AI competency must centre on human agency and social responsibility — technical understanding in
service of judgement.
UAICS CORRESPONDENCE
Domain D3 is built on the human-centred mindset dimension; UAICS cognitive-level tags adopt its three levels; the
A4/E4 capstones correspond to Create.
AI4K12 — Five Big Ideas in Artificial Intelligence
AAAI & CSTA, SINCE 2018
STRUCTURE
Five Big Ideas — Perception; Representation & Reasoning; Learning; Natural Interaction; Societal Impact — each
across four grade bands (K-2 / 3-5 / 6-8 / 9-12).
CORE CLAIM
Students should form a balanced view of AI — neither fear nor blind trust — by understanding how it actually
works.
UAICS CORRESPONDENCE
The principal source for domain D1’s throughline and progression; the SEARCH family maps to Big Idea 2, the ML
and NN families to Big Idea 3.
ISTE Standards for Students
CONTINUALLY REVISED
STRUCTURE
Seven standards — Empowered Learner, Digital Citizen, Knowledge Constructor, Innovative Designer,
Computational Thinker, Creative Communicator, Global Collaborator — described by capability rather than age.
CORE CLAIM
The heart of technological literacy is learner agency — safe, responsible, creative use.
UAICS CORRESPONDENCE
The principal source for domain D5; the D4-CT family maps to Computational Thinker; the Assumed Foundations
Statement references Empowered Learner among others.
ABOUT AIECB · KNOWLEDGE HUB
Knowledge Hub
AIECB
About AIECB
UAICS
People
Insights & News
Contact
Glossary
Twenty terms UAICS builds on — written so a parent or teacher with no technical
background can follow every one.
Showing the first 20 entries. The full glossary grows alongside the standard itself.
Media Literacy
Evaluating sources and recognising
manipulation. In an era when AI mass-
produces convincing content, it has moved
from an enrichment to basic protection.
Digital Footprint
The traceable record left by online activity.
Managing digital identity and privacy
settings is formally assessed content in the
UAICS citizenship domain.
Metacognition
Awareness and regulation of one’s own
thinking. “Is this my idea, or the AI’s?” is the
defining metacognitive question of the era.
Cognitive Load
Working memory processes only so much at
once. Judging what to think through oneself
and what to hand to a tool begins with
knowing one’s own load — part of UAICS’s
taught-not-tested content.
Human Agency
The retained capacity and right to
understand, question and overrule an AI’s
suggestion. The heart of UNESCO’s
framework — and the permanent question:
who, in the end, made this decision?
Word Embedding
Words mapped to numerical vectors that
reflect closeness of meaning — the
foundation of natural language processing:
similar meanings, similar numbers.
Feature Vector
An object’s important attributes turned into a
list of numbers, so computers can calculate
and compare. “Turning concepts into
numbers” is a key step in understanding how
AI sees the world.
Supervised Learning
Training on examples that carry correct
answers — one of machine learning’s three
paradigms, alongside unsupervised learning
(finding hidden structure) and reinforcement
learning (learning from trial and feedback).
Model Card
A document recording a model’s purpose,
training data, limits and proper scope — an
industry accountability practice. UAICS has
children write Model Cards for models they
train themselves: owning what one has
made.
Abstraction
Judging which details matter and which can
be dropped. Selecting features for a model
and deciding what context to give an AI draw
on exactly this one capability.
Computational Thinking
Decomposing problems, recognising
patterns, abstracting, designing steps. It is
not coding; in the AI era it is the underlying
craft of making a problem explicit.
Prompt Engineering
The craft of stating a task so AI can act on it:
goal, context, constraints. In advanced form,
decomposing a large task into staged
prompts with checking between — which is
structured thinking in training.
Hallucination
Fluent, plausible output that is wrong or
invented. The cause lies in probabilistic
continuation itself — and understanding the
cause serves a child far better than
memorising “AI lies.”
Generative AI
AI that produces new content — text,
images, sound — unlike classification AI,
which only judges. It generates text by
probabilistic continuation, word by word;
understanding this explains the particular
kind of error it makes.
Backpropagation
The method by which the signal of “how
wrong we were” travels backwards through
a network, adjusting parameters layer by
layer. Think of it as the model’s mechanism
for learning from its own errors.
Neural Network
A model built from many layers of simple
units, each transforming input into more
abstract representations. UAICS teaches its
working intuition at the conceptual level; no
mathematics is required.
Algorithmic Bias
When training data under-represents or
skews against certain groups, a model
performs systematically worse for them —
speech recognition faltering on strong
accents, for example. Bias mostly enters
through data, not machine malice.
Training Data & Labels
The examples used to teach a model; labels
are each example’s correct answer. Data
quantity and quality decide performance —
“rubbish in, rubbish out” begins here as a
judgement children can own.
Classifier
A machine-learning model that sorts inputs
into categories — cat or dog, for instance.
Training one and inspecting where it errs is
the best possible introduction to the idea
that AI makes mistakes, and that mistakes
can be found.
Machine Learning
A way for computers to find patterns from
many examples, rather than following rules
written one by one. A child’s first UAICS
practical is exactly this: feeding examples to
train a small model by hand.
National & International Framework Digests
The three frameworks UAICS integrates — what each one covers, its core claim,
and exactly where it shows up in UAICS.
UNESCO — AI Competency Framework for
Students
PUBLISHED 2024
STRUCTURE
Four dimensions (human-centred mindset, ethics of AI, AI techniques and applications,
AI system design) by three levels (Understand, Apply, Create) — twelve competency
blocks for school-age learners, deliberately unbanded by age.
CORE CLAIM
Student AI competency must centre on human agency and social responsibility —
technical understanding in service of judgement.
UAICS CORRESPONDENCE
Domain D3 is built on the human-centred mindset dimension; UAICS cognitive-level
tags adopt its three levels; the A4/E4 capstones correspond to Create.
AI4K12 — Five Big Ideas in Artificial
Intelligence
AAAI & CSTA, SINCE 2018
STRUCTURE
Five Big Ideas — Perception; Representation & Reasoning; Learning; Natural Interaction;
Societal Impact — each across four grade bands (K-2 / 3-5 / 6-8 / 9-12).
CORE CLAIM
Students should form a balanced view of AI — neither fear nor blind trust — by
understanding how it actually works.
UAICS CORRESPONDENCE
The principal source for domain D1’s throughline and progression; the SEARCH family
maps to Big Idea 2, the ML and NN families to Big Idea 3.
ISTE Standards for Students
CONTINUALLY REVISED
STRUCTURE
Seven standards — Empowered Learner, Digital Citizen, Knowledge Constructor,
Innovative Designer, Computational Thinker, Creative Communicator, Global
Collaborator — described by capability rather than age.
CORE CLAIM
The heart of technological literacy is learner agency — safe, responsible, creative
use.
UAICS CORRESPONDENCE
The principal source for domain D5; the D4-CT family maps to Computational Thinker;
the Assumed Foundations Statement references Empowered Learner among others.
ABOUT AIECB · KNOWLEDGE HUB
Knowledge Hub
Glossary
Twenty terms UAICS builds on — written
so a parent or teacher with no technical
background can follow every one.
Showing the first 20 entries. The full
glossary grows alongside the standard
itself.
Machine Learning
A way for computers to find patterns from many
examples, rather than following rules written
one by one. A child’s first UAICS practical is
exactly this: feeding examples to train a small
model by hand.
Classifier
A machine-learning model that sorts inputs into
categories — cat or dog, for instance. Training
one and inspecting where it errs is the best
possible introduction to the idea that AI makes
mistakes, and that mistakes can be found.
Training Data & Labels
The examples used to teach a model; labels are
each example’s correct answer. Data quantity
and quality decide performance — “rubbish in,
rubbish out” begins here as a judgement
children can own.
Algorithmic Bias
When training data under-represents or skews
against certain groups, a model performs
systematically worse for them — speech
recognition faltering on strong accents, for
example. Bias mostly enters through data, not
machine malice.
Neural Network
A model built from many layers of simple units,
each transforming input into more abstract
representations. UAICS teaches its working
intuition at the conceptual level; no mathematics
is required.
Backpropagation
The method by which the signal of “how wrong
we were” travels backwards through a network,
adjusting parameters layer by layer. Think of it
as the model’s mechanism for learning from its
own errors.
Generative AI
AI that produces new content — text, images,
sound — unlike classification AI, which only
judges. It generates text by probabilistic
continuation, word by word; understanding this
explains the particular kind of error it makes.
Hallucination
Fluent, plausible output that is wrong or
invented. The cause lies in probabilistic
continuation itself — and understanding the
cause serves a child far better than memorising
“AI lies.”
Prompt Engineering
The craft of stating a task so AI can act on it:
goal, context, constraints. In advanced form,
decomposing a large task into staged prompts
with checking between — which is structured
thinking in training.
Computational Thinking
Decomposing problems, recognising patterns,
abstracting, designing steps. It is not coding; in
the AI era it is the underlying craft of making a
problem explicit.
Abstraction
Judging which details matter and which can be
dropped. Selecting features for a model and
deciding what context to give an AI draw on
exactly this one capability.
Model Card
A document recording a model’s purpose,
training data, limits and proper scope — an
industry accountability practice. UAICS has
children write Model Cards for models they train
themselves: owning what one has made.
Supervised Learning
Training on examples that carry correct
answers — one of machine learning’s three
paradigms, alongside unsupervised learning
(finding hidden structure) and reinforcement
learning (learning from trial and feedback).
Feature Vector
An object’s important attributes turned into a list
of numbers, so computers can calculate and
compare. “Turning concepts into numbers” is a
key step in understanding how AI sees the
world.
Word Embedding
Words mapped to numerical vectors that reflect
closeness of meaning — the foundation of
natural language processing: similar meanings,
similar numbers.
Human Agency
The retained capacity and right to understand,
question and overrule an AI’s suggestion. The
heart of UNESCO’s framework — and the
permanent question: who, in the end, made this
decision?
Cognitive Load
Working memory processes only so much at
once. Judging what to think through oneself
and what to hand to a tool begins with knowing
one’s own load — part of UAICS’s taught-not-
tested content.
Metacognition
Awareness and regulation of one’s own thinking.
“Is this my idea, or the AI’s?” is the defining
metacognitive question of the era.
Digital Footprint
The traceable record left by online activity.
Managing digital identity and privacy settings is
formally assessed content in the UAICS
citizenship domain.
Media Literacy
Evaluating sources and recognising
manipulation. In an era when AI mass-produces
convincing content, it has moved from an
enrichment to basic protection.
National & International
Framework Digests
The three frameworks UAICS integrates —
what each one covers, its core claim, and
exactly where it shows up in UAICS.
UNESCO — AI Competency
Framework for Students
PUBLISHED 2024
STRUCTURE
Four dimensions (human-centred
mindset, ethics of AI, AI techniques and
applications, AI system design) by three
levels (Understand, Apply, Create) —
twelve competency blocks for school-
age learners, deliberately unbanded by
age.
CORE CLAIM
Student AI competency must centre
on human agency and social
responsibility — technical
understanding in service of
judgement.
UAICS CORRESPONDENCE
Domain D3 is built on the human-centred
mindset dimension; UAICS cognitive-
level tags adopt its three levels; the
A4/E4 capstones correspond to Create.
UNESCO official site ↗
AI4K12 — Five Big Ideas in
Artificial Intelligence
AAAI & CSTA, SINCE 2018
STRUCTURE
Five Big Ideas — Perception;
Representation & Reasoning; Learning;
Natural Interaction; Societal Impact —
each across four grade bands (K-2 / 3-5
/ 6-8 / 9-12).
CORE CLAIM
Students should form a balanced
view of AI — neither fear nor blind
trust — by understanding how it
actually works.
UAICS CORRESPONDENCE
The principal source for domain D1’s
throughline and progression; the
SEARCH family maps to Big Idea 2, the
ML and NN families to Big Idea 3.
ISTE Standards for Students
CONTINUALLY REVISED
STRUCTURE
Seven standards — Empowered Learner,
Digital Citizen, Knowledge Constructor,
Innovative Designer, Computational
Thinker, Creative Communicator, Global
Collaborator — described by capability
rather than age.
CORE CLAIM
The heart of technological literacy
is learner agency — safe,
responsible, creative use.
UAICS CORRESPONDENCE
The principal source for domain D5; the
D4-CT family maps to Computational
Thinker; the Assumed Foundations
Statement references Empowered
Learner among others.