when something is free from bias and gives equal consideration and treatment to all parts of a group or situation

when the data does not properly reflect the whole group or situation it is meant to describe

when data includes a wide range of examples that fairly represent different parts of the whole group or situation

Bias and accuracy in machine learning

Computing

AI in our world today

Uses of predictive and generative AI

Chatbot applications and other LLMs

Getting the most out of an LLM

Understanding image generators

Deepfakes and trust

Risk mitigation of AI

In this unit, pupils will explore how data science and AI tools are having an impact on our daily lives. They will develop an understanding of how to use these tools effectively as well as an awareness of the issues relating to trust, bias and misinformation.

Using data science and AI tools effectively and safely

ML models are always neutral and never make mistakes.

ML models automatically correct any unfairness in data.

ML models can reflect or repeat bias found in their training data.

Which statement about machine learning (ML) models is correct?

Put these steps in order for training an ML model fairly.

to make sure the model works reliably on new data

Why do ML models need both training data and separate test data?

It does not reflect the whole group or situation.

What does it mean if data is “unrepresentative”?

What is one effect of bias in an ML model?

when results unfairly favour or exclude certain groups

extra information that helps a model interpret your input

treating all groups equally, without favouritism

when data is used to help a model get better at recognising patterns

data used to check if a model works well on new information

What do we call the message or instruction you give to a large language model to get a response?

It helps the model to determine exactly what you want.

Why is it important to make your prompt specific when using an LLM?

The model may give a general or unrelated response.

The model will always give a perfect answer.

What can happen if your prompt to an LLM is too vague?

extra information to help the LLM understand your prompt

using knowledge to design or improve things

Bias can make answers unreliable or unfair.

Why is it important to check LLM responses for bias?

because the model always repeats the same answer

because the responses might reflect bias from the training data

Why should you critically evaluate the responses from an LLM?

Starter Quiz

What do we call the message or instruction you give to a large language model to get a response?