Lesson 4 - ethiscs of AI

Lesson 4:
Ethics and the
Responsible Use of AI

2024 Junior Program

DAYOFAIAUSTRALIA.COM | 3

The Big Ideas of AI

Societal Impact is the 5th “Big
Idea of AI.”

We hope that the AI tools that we
create will be used fairly and
honestly - making sure that they
don't have a negative impact on
society. We call this an “ethical”
or responsible use of a tool.

Lets play a simple game of real or generated by AI.

​

Real or fake?

Multiple Choice

Is this a real image or AI generated?

• real image

• AI generated

Poll

Is this a real image or AI generated?

• Real

• AI

Word Cloud

How did you know these images were AI?

Multiple Choice

Was this created by an artist or AI generated?

• created by an artist

• AI generated

Look at the process behind this image.

How many hours do you think went into making this image?

https://www.youtube.com/watch?v=6gHmKBym2pc

Multiple Choice

Was this created by an artist or AI generated?

• created by an artist

• AI generated

Multiple Choice

Was this created by an artist or AI generated?

• created by an artist

• AI generated?

Open Ended

How could AI be dangerous for artists?

DAYOFAIAUSTRALIA.COM | 4

A Simple Model of Ethics

Ethics is about understanding right and wrong

We're going to build a simple model of ethics to figure out
if an AI system should be used or not. The building blocks
of this system will be two ways the AI system could affect
people:

☝

Harm is something that results in a negative effect
for a person, or a group of people.

☝

Benefit is something that results in a positive effect
for a person, or a group of people.

DAYOFAIAUSTRALIA.COM | 5DAYOFAIAUSTRALIA.COM | 5

Understanding Harm & Benefit

Let’s look at a simple example

of harm and benefit.

DAYOFAIAUSTRALIA.COM | 6

Let's say a local
museum hosts an art
exhibition and uses
AI to generate the
artwork instead of
hiring local artists.

Effect: The museum saves time
and money by getting quick results and beautiful artwork.
This is a benefit for the museum.

Effect:Local artists don’t get
paid for their work and their
work is not seen by visitors. This
is harmful for the local artists.

DISCUSSION
QUESTIONS

DAYOFAIAUSTRALIA.COM | 7

Understanding Harm & Benefit

Applying your understanding

In your same pairs, discuss:
“Do you think that the way you use something (such as technology) makes a difference to whether it is harmful or beneficial?”

Some examples!
Cars are a great benefit
when driven safely, but
speeding is harmful.
Phones are a very helpful
tool in life, but can be
distracting in class.

DAYOFAIAUSTRALIA.COM | 8

Perspective Taking

Putting yourself in other people's shoes

From the plagiarism example, we can see how one decision can have harms and benefits for different groups of people. It's important to consider all perspectives that might be affected by a decision.

In ethics, we call the people who might be affected by something stakeholders.

Have a look at this image.

DAYOFAIAUSTRALIA.COM | 9

Here’s a similar one.

DAYOFAIAUSTRALIA.COM | 10

One of these images is made with AI.

DAYOFAIAUSTRALIA.COM | 11

Poll

Created by an artist or AI generated?

• Created by an artist

• AI generated

Poll

Created by an artist or AI generated?

• Created by an artist

• AI generated

Here’s the answer!

DAYOFAIAUSTRALIA.COM | 11

​Crystalline Maples,
Erin Hanson 2021

​AI-Generated Art, Stable Diffusion AI

DISCUSSION
QUESTIONS

DAYOFAIAUSTRALIA.COM | 14

Understanding Harm & Benefit

Plagiarism. Taking someone else’s work or ideas and passing them off as your own.

AI art like the one made with Stable Diffusion has to be trained using real artwork made by humans.

☝

Do you think AI-generated art like this one is
plagiarism? Why, or why not?

☝

Does your answer change if the artist knows their work
is being used to train AI?

Human-made art

AI-generated art

DAYOFAIAUSTRALIA.COM | 15

A Biased Point Of View

What happens if one perspective is favoured or ignored?

There's a problem with our simple model of ethics based on
benefit and harm. What if we simply chose to favour or ignore some perspectives over others?

“Bias ia a preference or a

prejudice towards one or more

groups, especially in a way
considered to be unfair or

​This leads to bias.

DAYOFAIAUSTRALIA.COM | 16

Understanding Data Bias

When our training data is biased, things get worse.

If humans become biased, we can change our minds in
response to new information, and learn from it. AI can’t do
that as easily.

AI can become biased when the data used to train it is
biased. We call this data bias. It’s not always obvious that the data was biased in the first place, and an AI will need to be re-trained with new data to remove the bias.

DAYOFAIAUSTRALIA.COM | 17

An Example of Data Bias

A large, fair and representative dataset is
important. For example:imagine an AI system that generates pictures of people.

Lots of people wear glasses. If an AI system is
trained on a dataset that mostly includes people without glasses, then the AI is not likely to generate a picture of someone with glasses.

This system would be biased, because the data used to train it was biased.

Generated by DALL.E, OpenAI

DAYOFAIAUSTRALIA.COM | 18

An Example of Data Bias

To fix this problem, we should re-train the AI
with a more diverse dataset, that shows lots of
people with glasses, as well as lots of people
without glasses.

This way, the dataset is larger, fairer, and
represents a broader range of people. We can apply the same concept when thinking about other biases like race, culture and gender!

Generated by DALL.E, OpenAI

DAYOFAIAUSTRALIA.COM | 19DAYOFAIAUSTRALIA.COM | 19

Meeting the Stakeholders:

People in the Arts

DAYOFAIAUSTRALIA.COM | 20

Who will be affected by this AI technology?

Let's meet the stakeholders
affected by this AI technology
from the art industry.

DAYOFAIAUSTRALIA.COM | 21

Felicity

High school student, art fan

“Hi, I’m Felicity! I’m in Year 10 and my favourite subject is
visual arts. I like to watch YouTube videos and play around
with AI models in my spare time. My favourite artists are
Joy Buolamwini and Tega Brain, both of them work with AI!

My parents are worried I spend too much time on my
computer and not drawing as much. I wish they could
understand that I’m still creating art, just in a different way through AI engines! It’s so wonderful seeing ideas brought to life through technology.”

Images generated by DALL.E and stakeholder statements by

ChatGPT,OpenAI

Word Cloud

High school student, art fan.

What do they stand to benefit (positively) from AI technology?

Word Cloud

High school student, art fan.

What might be a potential harm (negative) from the AI technology?

DAYOFAIAUSTRALIA.COM | 22

Caleb

Local artist

“Hey, I’m Caleb. I have been in the industry for ten years and
have a university degree in Visual Arts with a major in
Photography. My favourite subject matter to photograph is
nature.

It has been hard getting work recently, often people will tell me they have seen better work online. Some of the
photographs are not even of real life, instead they’re AI, but look so realistic!”

Images generated by DALL.E and stakeholder statements by

ChatGPT,OpenAI

Word Cloud

Local artist

What do they stand to benefit (positively) from AI technology?

Word Cloud

Local artist

What might be a potential harm (negative) from the AI technology?

Kelly

Local art museum owner

“Hello, I’m Kelly and the proud owner of the local art
museum. I’ve been running my business for over three
decades and have always had lots of people coming to the
museum. I like to display a large variety of art, from
drawings to digital pieces!

Lately, I have had many more artists come to me which is
great! However, I had several artists with very similar
looking artworks. It has been hard to decide what artwork to
display as I don’t always know if the art is made with AI or
by hand.”

DAYOFAIAUSTRALIA.COM | 23

Images generated by DALL.E and stakeholder statements by

ChatGPT,OpenAI

Open Ended

How do you think that AI could be worrying to Kelly the museum owner?

In March the Sony World Photography Awards announced the winning entry in their creative photo category: a black-and-white image of an older woman embracing a younger one, entitled PSEUDOMNESIA: The Electrician. The press release announcing the win describes the photograph as “haunting” and “reminiscent of the visual language of 1940s family portraits.”

But the artist, Berlin-based Boris Eldagsen, turned down the award. His photograph was not a photograph at all, he announced: he had crafted it through creative prompting of DALL-E 2, an artificial intelligence image generator.

​

​AI photography

Poll

Do you think AI art should be able to win prizes?

• Yes

• No

• Maybe, it depends on the art work.

STUDENT-LED

ACTIVITY

DAYOFAIAUSTRALIA.COM | 27

Discussion

☝

Do you think the video is real?
If so, what features appear real?

☝

Do you think the video is AI?
If so, what features seem artificially generated?

STUDENT-LED

ACTIVITY

Sora: a new AI model

Sora is a new generative AI model created by OpenAI that can generate video based on a text prompt.

E.g. The video on the right used this prompt:

Prompt: A litter of golden retriever puppies
playing in the snow. Their heads pop out of the
snow, covered in.

Watch this for examples of what Sora can generate: https://www.youtube.com/watch?v=HK6y8DAPN_0

DAYOFAIAUSTRALIA.COM | 29

Generated by Sora, OpenAI

CASE

STUDIES

DAYOFAIAUSTRALIA.COM | 32

In Australia, AI technology
is all around us.

Discuss:
How do you know what sources of
information to trust?

At the University of Queensland,
AI is being used to fight fake news.

CASE

STUDIES

DAYOFAIAUSTRALIA.COM | 33

AI can be used to create and spread false information through online social networks and use data to profile and micro-target individuals with customised false content.

Fortunately, AI and data-driven methods can be used to detect and control misinformation in the same way they are used to spread it.

CASE

STUDIES

DAYOFAIAUSTRALIA.COM | 34

AI can be used to create and spread false information through online social networks and use data to profile and micro-target individuals with customised false content.

The University of Queensland’s AI Research Network
studies how people make decisions (e.g. whether to
trust a piece of information) and how AI systems can
empower people to spot misinformation and how
this information can be removed from online
platforms.

CASE

STUDIES

DAYOFAIAUSTRALIA.COM | 35

Researchers at UQ have
developed a system that
uses experts together with AI (this is called hybrid intelligence) to refine an algorithm that spots misinformation online.

CASE

STUDIES

DAYOFAIAUSTRALIA.COM | 36

Here’s how it works:

Human Expert Input

Datasets

(e.g. RMIT ABC Fact Check)

Misleading

Not Misleading

AI Model

The AI takes human input and trusted datasets, and categorises new information it sees as being

misleading or not misleading. Using AI this way means we can process huge amounts of information

very quickly, much faster than humans working alone!

DISCUSSION
QUESTIONS

DAYOFAIAUSTRALIA.COM | 37

Discuss in your groups:

Misinformation. False or inaccurate information, especially that which is

deliberately intended to deceive.

Personal Reflection

What do you think?

Do you think AI generated media made with tools like Sora, Dall-E or ChatGPT is misinformation?

DISCUSSION
QUESTIONS

DAYOFAIAUSTRALIA.COM | 38

What are the
limits of AI?

Open Ended

What are the
limits of AI?

● All forms of AI, including image and video

generators, are ultimately just guessing at what
you want. Sometimes they’re right, sometimes
they’re not.

● They don’t really “know” or understand the

things they’re creating or saying.

DAYOFAIAUSTRALIA.COM | 39

The limits of AI

AI models are just
making predictions

● Even if you give it the exact same prompt!

● The AI generates a new prediction every

single time.

● Generating two identical responses is

extremely unlikely.

DAYOFAIAUSTRALIA.COM | 40

The limits of AI

AI models do not always

give the same answer

● Bias is a preference or prejudice towards one or

more groups, especially in a way considered to be
unfair or unrepresentative.

● For example, an AI trained to generate images of

people will never generate an image of someone
with glasses, if none of the training samples
showed people with glasses.

● We should always be aware of potential bias in the

AI tools we use.

DAYOFAIAUSTRALIA.COM | 41

The limits of AI
AI models can

be biased