Bit by Bit
Glossary

175 terms in plain language

Every term used in the Bit by Bit lessons, explained for a complete beginner.

accountability
Clear responsibility for a system's outcomes, including someone who answers for mistakes and can fix them.Taught in AI Ethics and Society
activation function
The final step in a neuron, which turns its total into an output. The simplest one gives 1 if the total is above zero and 0 otherwise.Taught in How Machines Learn
aggregation
Combining the updates from all clients into one change to the shared model.Taught in Federated Learning
algorithm
A step-by-step method for solving a problem. A recipe is an algorithm for cooking a dish.Taught in AI from Zero
anonymisation
Removing information that identifies people from a dataset.Taught in AI Ethics and Society
artificial general intelligence
A hypothetical AI that could learn and perform any intellectual task a person can. Experts disagree on whether and when it will exist. Often shortened to AGI.Taught in AI from Zero
artificial intelligence
Computer systems that carry out tasks which normally need human thinking, such as understanding language, recognising pictures or making decisions. Often shortened to AI.Taught in AI from Zero
assignment
Storing a value under a name with the = sign. age = 20 means "let age now refer to 20".Taught in Python First Steps
backpropagation
The method that works out, for every weight in a network, how much it contributed to the error. It passes the error backwards from the output layer to the input layer.Taught in Neural Networks
bar chart
A chart that compares amounts across categories using bars of different lengths.Taught in Working with Data
bias
A consistent unfairness in a model's answers, usually copied from unbalanced or unfair training data.Taught in AI from Zero
bias term
A parameter added to a neuron's total. It shifts the point at which the neuron switches on. It has nothing to do with unfairness, which is a different meaning of the word bias.Taught in How Machines Learn
big o notation
A short way of describing how an algorithm's work grows with the size of its input, such as O(n) or O(log n).Taught in Thinking in Algorithms
binary
A way of writing numbers using only the digits 0 and 1.Taught in How Computers Work
bit
The smallest piece of information: a single 0 or 1.Taught in How Computers Work
boolean
A value that is either True or False.Taught in Python First Steps
bubble sort
A simple sorting method that repeatedly compares neighbouring items and swaps them if they are in the wrong order.Taught in Thinking in Algorithms
bug
A mistake in a program that makes it behave wrongly.Taught in Python First Steps
byte
A group of eight bits. One byte can hold any number from 0 to 255.Taught in How Computers Work
categorical data
Data that names a group, such as a country or a colour. Averaging it makes no sense.Taught in Working with Data
causation
When one thing actually makes another happen. Correlation alone does not prove causation.Taught in Working with Data
character encoding
An agreed table that gives every letter and symbol a number, so that text can be stored as numbers. ASCII and Unicode are the best known.Taught in How Computers Work
chat assistant
A program you talk to in everyday language, built on a large language model.Taught in Using AI Tools Wisely
classification
Predicting a category, such as "spam" or "not spam".Taught in How Machines Learn
client
A device or organisation that holds its own data and does its own training, such as a phone, a hospital or a sensor.Taught in Federated Learning
column
One vertical line in a table. In a dataset, usually one property measured for every example, such as age or price.Taught in Working with Data
comment
A note for human readers that Python ignores. It starts with the # sign.Taught in Python First Steps
comparison operator
A symbol that compares two values, such as > (greater than), < (less than), == (equal to) and != (not equal to).Taught in Python First Steps
computer program
A list of exact instructions that a computer follows, one step after another.Taught in AI from Zero
condition
An expression that is either True or False, used to make a decision.Taught in Python First Steps
context window
The amount of text a language model can take into account at once, measured in tokens. Anything beyond it is not seen.Taught in Using AI Tools Wisely
convergence
The point where more training steps no longer change the loss in any meaningful way.Taught in How Machines Learn
coordinator
The server that sends out the shared model and combines the changes the clients send back. Also called the server or aggregator.Taught in Federated Learning
correlation
A pattern in which two quantities tend to rise and fall together, or in opposite directions.Taught in Working with Data
csv
Comma-separated values: a plain text file where each line is a row and commas separate the columns. A common way to store tables.Taught in Working with Data
data
Recorded facts that a computer can store: numbers, words, pictures, sounds, clicks.Taught in AI from Zero
data cleaning
Finding and fixing problems in data, such as missing values, duplicates and typing mistakes, before using it.Taught in Working with Data
dataset
A collection of examples, usually arranged as a table with one row per example.Taught in How Machines Learn
decision boundary
The line that separates the inputs a model puts in one category from those it puts in another.Taught in How Machines Learn
decomposition
Breaking a big problem into smaller problems that are each easy to solve.Taught in Thinking in Algorithms
deep learning
Machine learning that uses neural networks with many layers.Taught in AI from Zero
deepfake
An image, video or voice recording made or changed by AI to show something that did not happen.Taught in Using AI Tools Wisely
duplicate
The same record appearing more than once in a dataset.Taught in Working with Data
efficiency
How much time or memory an algorithm needs, especially as the input grows.Taught in Thinking in Algorithms
epoch
One full pass through all the training examples.Taught in Neural Networks
error
The gap between a model's prediction and the correct answer.Taught in How Machines Learn
ethics
Careful thinking about what is right and wrong, and how we should act.Taught in AI Ethics and Society
example
One item a computer learns from, such as one photo or one email, usually with the correct answer attached.Taught in AI from Zero
explainability
The ability to give a reason for a particular AI decision that a person can understand.Taught in AI Ethics and Society
expression
A piece of code that works out to a value, such as 2 + 3 or price * 2.Taught in Python First Steps
fairness
In AI, the idea that a system should not treat people worse because of who they are. There are several precise definitions, and they can conflict.Taught in AI Ethics and Society
false negative
An item that should have been flagged and was missed.Taught in Federated Learning
false positive
A normal item that the model wrongly flags.Taught in Federated Learning
feature
One measurable property of an example, such as its length, its colour or its price.Taught in AI from Zero
federated averaging
The standard aggregation method, in which each client's update is weighted by its sample count. Often shortened to FedAvg.Taught in Federated Learning
federated learning
A way to train a shared model in which several clients each train on their own data and send only their changes to a coordinator. The raw data never leaves the clients.Taught in Federated Learning
fetch-decode-execute cycle
The three steps a processor repeats without end: fetch the next instruction, decode what it means, execute it.Taught in How Computers Work
few-shot prompting
Showing a model a few examples of what you want inside the prompt, so it can copy the pattern.Taught in Using AI Tools Wisely
float
A number with a decimal point, such as 3.14.Taught in Python First Steps
forward pass
Running an input through the network, layer by layer, to get a prediction.Taught in Neural Networks
function
A named, reusable piece of code. You give it inputs and it returns a result.Taught in Neural Networks
generalisation
How well a model performs on examples it has never seen. This is the real goal of training.Taught in How Machines Learn
generative ai
AI that creates new content, such as text, pictures, music or computer code.Taught in AI from Zero
global model
The shared model that the coordinator maintains and every client receives.Taught in Federated Learning
gradient
The slope of the loss at the current parameter values. It points in the direction that makes the loss grow fastest.Taught in How Machines Learn
gradient descent
A training method: repeatedly nudge each parameter a little in the direction that lowers the loss.Taught in How Machines Learn
hallucination
When a generative AI states something false in a fluent, confident way.Taught in AI from Zero
hardware
The physical parts of a computer that you can touch: the chips, the screen, the keyboard.Taught in How Computers Work
hidden layer
A layer between the input and the output. Its neurons find useful intermediate patterns.Taught in Neural Networks
histogram
A chart showing how often values fall into each range, which reveals the shape of the data.Taught in Working with Data
historical bias
Unfairness already present in past decisions, which a model copies when it learns from them.Taught in AI Ethics and Society
honest client
A client that sends the update it really learned.Taught in Federated Learning
human oversight
Keeping people able to check, override or stop an AI system's decisions.Taught in AI Ethics and Society
if statement
Code that runs only when a condition is True. An optional else part runs when it is False.Taught in Python First Steps
indentation
The spaces at the start of a line. Python uses them to know which lines belong inside an if or a loop.Taught in Python First Steps
input
Anything that goes into a computer, such as a key press, a tap, a photo or a sound.Taught in How Computers Work
input layer
The first layer of a network. It simply holds the input values, one per feature.Taught in Neural Networks
instruction
One very small command that a processor can carry out, such as "add these two numbers".Taught in How Computers Work
integer
A whole number, such as 7 or −3.Taught in Python First Steps
intercept
The prediction when the input is zero. It sets where the line starts.Taught in How Machines Learn
iteration
One repeat of a process. In training, one update of the parameters.Taught in How Machines Learn
knowledge cutoff
The date after which a model's training data stops. It may not know about later events unless it can search.Taught in Using AI Tools Wisely
label
The correct answer attached to an example, such as "apple" or "spam".Taught in AI from Zero
large language model
A very large neural network trained on huge amounts of text to predict which word comes next. Chat assistants are built on these. Often shortened to LLM.Taught in AI from Zero
layer
A group of neurons that all receive the same inputs and work side by side.Taught in Neural Networks
learning rate
A number that sets how big each gradient descent step is.Taught in How Machines Learn
linear model
A model that predicts with a straight line: prediction = slope × input + intercept.Taught in How Machines Learn
linear time
Work that grows in step with the input: twice the data, twice the work. Written O(n).Taught in Thinking in Algorithms
linearly separable
Describes two groups of points that a single straight line can separate.Taught in Neural Networks
list
An ordered collection of values kept under one name, written in square brackets, such as [3, 8, 5].Taught in Python First Steps
local training
Training that a client carries out on its own data, on its own machine.Taught in Federated Learning
logarithmic time
Work that grows very slowly: doubling the data adds only one more step. Written O(log n).Taught in Thinking in Algorithms
logic gate
A small circuit, built from transistors, that takes one or two bits in and gives one bit out according to a fixed rule.Taught in How Computers Work
loop
Code that repeats. A for loop runs once for each item in a collection.Taught in Python First Steps
loss
A single number that says how wrong a model is across all its examples. Lower is better.Taught in How Machines Learn
loss curve
A chart of the loss as training goes on. A healthy curve falls and then flattens.Taught in Neural Networks
machine code
Instructions written as binary numbers, the only language a processor understands directly.Taught in How Computers Work
machine learning
A way of building programs in which the computer works out the rules itself by studying examples, instead of a person writing the rules.Taught in AI from Zero
malicious update
An update made deliberately harmful in order to damage the shared model. Sending such updates is called a poisoning attack.Taught in Federated Learning
mean
The ordinary average: add the values and divide by how many there are.Taught in Federated Learning
mean squared error
A common loss. Square each error, then take the average. Often shortened to MSE.Taught in How Machines Learn
measurement bias
When the thing measured is a poor stand-in for what we actually care about, and the gap differs between groups.Taught in AI Ethics and Society
median
The middle value after sorting. Half the values are below it and half are above.Taught in Federated Learning
memory
Fast, temporary working space for whatever the computer is doing right now. It is emptied when the power goes off. Also called RAM.Taught in How Computers Work
missing value
An empty spot in a dataset where a value should be.Taught in Working with Data
mode
The value that appears most often.Taught in Working with Data
model
The result of training: a program that takes an input and gives an answer, using the patterns it found in the examples.Taught in AI from Zero
narrow ai
AI built for one kind of task, such as recognising faces or recommending films.Taught in AI from Zero
nearest neighbour
A simple learning method. To label something new, find the most similar example you already know and copy its label.Taught in AI from Zero
neural network
A model built from many small calculating units, loosely inspired by brain cells, connected in layers.Taught in AI from Zero
neuron
The basic unit of a neural network. It multiplies each input by a weight, adds the results and a bias term, then applies an activation function.Taught in How Machines Learn
noise
Random variation in data that carries no real pattern, such as measurement mistakes.Taught in How Machines Learn
non-linearity
Anything that is not a straight line. Activation functions add it, which lets a network learn curved boundaries.Taught in Neural Networks
numerical data
Data that is a number you can do arithmetic with, such as height or price.Taught in Working with Data
operating system
The software that manages the hardware and lets other programs run, such as Windows, Android, iOS or Linux.Taught in How Computers Work
outlier
A value far away from the others. It may be a mistake or a real but unusual case.Taught in Working with Data
output
Anything a computer gives back, such as a picture on the screen, a sound or a printed page.Taught in How Computers Work
output layer
The last layer of a network. It produces the prediction.Taught in Neural Networks
overfitting
When a model learns the training examples too closely, noise included, and so does badly on new examples.Taught in How Machines Learn
parameter
An adjustable number inside a model. Training means finding good values for the parameters.Taught in How Machines Learn
pattern
A regularity that repeats in data, such as "emails that mention a prize are often spam".Taught in AI from Zero
personal data
Information about an identifiable person, such as a name, phone number, address or ID number.Taught in Using AI Tools Wisely
pixel
One tiny dot of colour in a picture or on a screen.Taught in How Computers Work
precision
Of everything the model flagged, the share that deserved it.Taught in Federated Learning
prediction
The answer a model gives for an input it has not seen before.Taught in AI from Zero
primary source
The original source of a fact, such as the official document, the research paper or the law itself.Taught in Using AI Tools Wisely
privacy
People's ability to control information about themselves.Taught in AI Ethics and Society
processor
The chip that carries out instructions. It is the part that does the actual computing. Also called the CPU (central processing unit).Taught in How Computers Work
programming language
A language that people can read and write, such as Python, which is translated into machine code for the processor.Taught in How Computers Work
prompt
The message or instructions you give an AI tool.Taught in Using AI Tools Wisely
pseudocode
Steps written in plain language, laid out like a program but not in any real programming language.Taught in Thinking in Algorithms
Python
A popular programming language known for being easy to read. It is the main language of AI.Taught in How Computers Work
quadratic time
Work that grows with the square of the input: twice the data, four times the work. Written O(n²).Taught in Thinking in Algorithms
range
The largest value minus the smallest value. A quick measure of how spread out data is.Taught in Working with Data
re-identification
Working out who someone is in a supposedly anonymous dataset by combining it with other information.Taught in AI Ethics and Society
recall
Of everything that deserved a flag, the share the model caught.Taught in Federated Learning
regression
Predicting a number, such as a price or a temperature.Taught in How Machines Learn
ReLU
An activation function that outputs 0 for negative inputs and passes positive inputs through unchanged. Short for rectified linear unit.Taught in Neural Networks
robust aggregation
Aggregation designed so that a small number of bad updates cannot drag the result far. The median is one example.Taught in Federated Learning
round
One full cycle of federated learning: send the model out, train locally, send updates back, combine them.Taught in Federated Learning
row
One horizontal line in a table. In a dataset, usually one example, such as one person or one sale.Taught in Working with Data
rule-based system
A program in which people write every rule by hand, in the form "if this happens, do that".Taught in AI from Zero
sample
One record in a client's data. The sample count is how many records the client trained on.Taught in Federated Learning
sampling bias
When the data collected does not represent everyone the system will be used on.Taught in AI Ethics and Society
scatter plot
A chart with one dot per example, placed by two numerical values, to show whether they are related.Taught in Working with Data
score
In this lab, a number from 0 to 100 describing how unusual a network record looks.Taught in Federated Learning
search algorithm
A method for finding an item in a collection.Taught in Thinking in Algorithms
sigmoid
An activation function that squeezes any number smoothly into the range 0 to 1.Taught in Neural Networks
slope
How much the prediction rises when the input goes up by one. It sets how steep the line is.Taught in How Machines Learn
software
The programs that tell the hardware what to do. You cannot touch software, only run it.Taught in How Computers Work
sorting
Putting items in order, such as smallest to largest or A to Z.Taught in Thinking in Algorithms
stakeholder
Anyone affected by a decision or a system, not only the people who build or buy it.Taught in AI Ethics and Society
statement
One complete line of instruction in a program.Taught in Python First Steps
storage
Long-term space where files and programs are kept, even when the power is off.Taught in How Computers Work
string
A piece of text in a program, written between quotes, such as "hello".Taught in Python First Steps
supervised learning
Machine learning from examples that come with the correct answer (a label) attached.Taught in How Machines Learn
surveillance
Watching or recording people's activities, often at large scale.Taught in AI Ethics and Society
tanh
An activation function that squeezes any number smoothly into the range −1 to 1.Taught in Neural Networks
test set
Examples kept hidden during training and used afterwards to check the model honestly.Taught in How Machines Learn
threshold
A cut-off value. In this lab the model flags every record whose score is at or above the threshold.Taught in Federated Learning
token
A small piece of text, often part of a word, that a language model reads and writes one at a time.Taught in Using AI Tools Wisely
training
Showing examples to a learning program so that it becomes better at its task.Taught in AI from Zero
training set
The examples a model is allowed to learn from.Taught in How Machines Learn
transistor
A microscopic electronic switch with no moving parts. It is either on (1) or off (0).Taught in How Computers Work
transparency
Being open about when AI is used, what data it uses and how it is meant to work.Taught in AI Ethics and Society
truth table
A table listing every possible input to a logic gate together with the output it gives.Taught in How Computers Work
underfitting
When a model is too simple to capture the real pattern, so it does badly even on its training examples.Taught in How Machines Learn
update
The change a client sends to the coordinator after local training.Taught in Federated Learning
variable
A name that refers to a stored value, so that the value can be used and changed later.Taught in Python First Steps
verification
Checking a claim against a reliable, independent source before relying on it.Taught in Using AI Tools Wisely
weight
A parameter that sets how strongly one input influences a neuron.Taught in How Machines Learn
weighted mean
An average in which some values count more than others. Each value is multiplied by its share of influence before adding.Taught in Federated Learning