1 of 130

The Challenges and Opportunities for Education of Artificial Intelligence

2 of 130

Introduction to AI in Education

Navigating the New Frontier

3 of 130

What is Artificial Intelligence?

AI involves machines or software mimicking human intelligence to perform tasks and improve themselves �based on the information they collect

4 of 130

Brain vs AI

Human Brain

�Complexity�Adaptability�Creativity

Artificial Intelligence

�Data Processing�Consistency

Learning Efficiency

Synergy & Differences��Complementary Strengths�Interdependence

5 of 130

Pre 2012 Artifical Intelligence

6 of 130

Post 2012 Artifical Intelligence

7 of 130

Post 2012 Artifical Intelligence

Learnt Processes

8 of 130

A Brief History of AI in Education

9 of 130

1960s-1970s Programmed instruction and branching tutorials�

1980s-1990s Intelligent tutoring systems with customized feedback and adaptive learning paths based on individual student interactions�

2000s Introduction of data mining techniques for assessment analysis, retention targeting, and personalised learning experiences

�2010s Adaptive learning technologies

Present AI-driven content creation and administrative automation

10 of 130

Administrative Automation

Personalised Learning

Content Creation

11 of 130

Personalised Learning

Adaptive Learning Platforms

Intelligent Tutoring Systems

AccommodateLearning Pace

Improve Engagement & Outcomes

12 of 130

Needs�Abilities

Preferences

24/7�Realtime�Analysis

Patterns�Insights

Customisation

Feedback�Motivation�DIagnosis

13 of 130

Data Collection��Student Interactions�Performance Metrics�Preferences & Feedback

Analysis��Identifying Patterns�Strengths & Weaknesses�Learning Preferences

Adaptation��Customising Content�Pacing Adjustments�Learner Adaptation�

14 of 130

Transforming Learning

DreamBox Learning

�Adaptive Mathematics�Adjusting Curriculum�Improved Maths Scores

Khan Academy��Practice Exercises�Personalised Pathways�Insights into Progress

Carnegie Learning��Intelligent Tutoring�Realtime Feedback�Meeting Student Needs

15 of 130

Catering to Every Learner

Text

�Summarisation�Analysis�Synthesis

Sound��Speak Aloud�Translation

Music

Images��Visuals�Diagrams

Video

16 of 130

Content Creation

Dynamic

�Texts�Activities�Audiovisuals

Simulations��Understanding�Practice

Exploration

Diverse��Special Needs

Personalisation

Extension

17 of 130

Lesson Planning

18 of 130

Automating Administrative Tasks

Grading

�Objective�Subjective�Feedback

Attendance��Identity Verification�Location Tracking

Activity Monitoring

Scheduling��Dynamic & Responsive�Personalised

Optimised

19 of 130

BEd Study Planner

20 of 130

Efficiency

�Routine�Repetitive�Communication

Decision Making��Allocations and Focus�Strategic Analysis

Predictions & Scenarios

Management��Budgeting�Personalisation

Optimisation

21 of 130

Enhancing Accessibility

Disabilities

Text-to-Speech-Text�Hearing Aids�Adaptive Inputs

Language��Real-time Translation

Language Learning

Automated Signing

Visualisation��VR and AR�Data Visuals

22 of 130

Data Privacy

Importance

�Sensitivity�Misuse�Trust

Challenges��Consent�Breaches

Discrimination

Solutions��Transparency and Control�Minimisation and Anonymisation

Regulations and Standards

23 of 130

Bias and Fairness

Understanding Bias

�Sources�Types�Impact

Promoting Fairness��Metrics & Models

Diverse & Inclusive Data Sets

Transparency & Explainability

Mitigation��Auditing and Testing�Frameworks

24 of 130

Trust and Transparency

Importance

�Confidence�Decision-Making�Ethical Compliance

Challenges��Complexity�Tradeoffs

Risks

Strategies��Explainable AI�Standards

Communication

25 of 130

Introduction to Machine Learning

26 of 130

Supervised Learning

27 of 130

Teachable Machine

28 of 130

Map Analogy

29 of 130

Map Analogy

30 of 130

Map Analogy

31 of 130

32 of 130

36M

Numbers

33 of 130

Machine Learning

36M

Numbers

34 of 130

Machine Learning

36M

Numbers

One Image is One Location

2

Numbers

35 of 130

Machine Learning

Billions

Each by 2

Island of Images of Cats

Training defines ‘islands’

CAT

Machine Learning

Numbers

Each by 2

36 of 130

Machine Learning

Billions

Each by 2

Island of Images of Cats

Training defines ‘islands’

Island of Images of Dogs

CAT

Machine Learning

Numbers

DOG

Each by 2

37 of 130

Millions

Machine Learning

2

Numbers

Machine Learning

Billions

Each by 2

Island of Images of Cats

Island of Images of Dogs

CAT

Machine Learning

Numbers

DOG

???

Predict

38 of 130

Millions

Machine Learning

2

Numbers

Machine Learning

Billions

Each by 2

Island of Images of Cats

Island of Images of Dogs

CAT

Machine Learning

Numbers

???

Train

DOG

39 of 130

Unsupervised Learning

40 of 130

Machine Learning

Self Supervision

Machine Learning

Does not require labels

Generates new islands without needing to define them

41 of 130

My Pet Cat

Jill has a pet

Her pet is a cat.

Her cat is white

Her cats name is Leo

My Pet Cat

Jill has a pet.

Her pet is a cat

Her cat is white

Her cats name is Leo

Machine Learning

42 of 130

Light Fur

Dark Fur

CATS

DOGS

Small Ears Big Ears

43 of 130

CATS

DOGS

Small Ears Big Ears

Light Fur

Dark Fur

44 of 130

A story about a dog with dark fur and small ears

Light Fur

Dark Fur

A story about a cat with light fur and big ears

CATS

DOGS

Small Ears Big Ears

45 of 130

A story about a dog with dark fur and small ears

Light Fur

Dark Fur

A story about a cat with light fur and big ears

CATS

DOGS

Small Ears Big Ears

46 of 130

Light Fur

Dark Fur

CATS

DOGS

Small Ears Big Ears

Create an image of a hybrid of a dog and a cat with light fur and big ears

47 of 130

My Pet Cat

Jill has a pet.

Her pet is a cat.

Her cat is white

Her cats name is Leo

My Pet Cat

Jill has a pet.

Her pet is a cat.

Her cat is white

Her cats name is Leo

Machine Learning

48 of 130

My Pet Cat

Jill has a pet.

Her pet is a cat

Her cat is white

Her cats name is Leo

My Pet Cat

Jill has a pet.

Her pet is a cat

Her cat is white

Her cats name is Leo

Machine Learning

49 of 130

My Pet Cat

Jill has a pet.

Her pet is a cat

Her cat is white

Her cats name is Leo

Machine Learning

Text to Image

50 of 130

My Pet Cat

Jill has a pet.

Her pet is a cat

Her cat is white

Her cats name is Leo

Machine Learning

Text to Music

51 of 130

My Pet Cat

Jill has a pet.

Her pet is a cat

Her cat is white

Her cats name is Leo

Machine Learning

Image to Text

52 of 130

Machine Learning

Music to Image

53 of 130

CATS

DOGS

Generative Video

54 of 130

55 of 130

56 of 130

To

Generative Chat

be

or

not

to

be

see

think

cat

57 of 130

To

Generative Chat

be

or

not

to

be 95%

see 2%

think 3%

cat 0.001%

58 of 130

To

Generative Chat

be

or

not

to

be 95%

see 2%

cat 0.001%

think 3%

Tokens

59 of 130

To

Generative Chat

be

or

not

to

be 95%

see 2%

Tokens

(Over 8000 GPT4)

cat 0.001%

think 3%

60 of 130

To

Generative Chat

be

or

not

to

be 95%

see 2%

Tokens

(Over 32000 GPT4-32K)

cat 0.001%

think 3%

61 of 130

moebio.com/mind

62 of 130

Cancer

Healthy

Unknown Data

New Patient

63 of 130

Cancer

Healthy

Unknown Data

New Patient

64 of 130

Cancer

Healthy

Unknown Data

New Patient

65 of 130

Avoid

Drive

Unknown Data

New Event

66 of 130

67 of 130

Select

Reject

Concept and Model Drift

Select

68 of 130

69 of 130

Light Fur

Dark Fur

CATS

DOGS

Small Ears Big Ears

Parameters�

70 of 130

Training Time

GPT 2 / BERT

GPT 3 / Turing

GPT 4 / LaMDA

Hours - Days

Weeks - Months

Years*

1.5 Billion�Parameters

175 Billion�Parameters�

1.8 Trillion�Parameters

71 of 130

Training Cutoffs

GPT 3

GPT 4

GPT 4 Turbo

September

2021

September

2021

April

2023

72 of 130

Male

Cancer

Male�Healthy

Bias

73 of 130

Male

Cancer

Male�Healthy

Bias

Female

Cancer

Male�Healthy

Female�Healthy

74 of 130

White Cancer

White�Healthy

Bias

75 of 130

White Cancer

White�Healthy

Bias

Black Cancer

New Patient

76 of 130

Introduction to Artificial Intelligence and Ethics in Education

77 of 130

In millennia, ethical issues have not yet been conclusively resolved for humans, resolving them for AI will likely take more than this presentation.

78 of 130

79 of 130

Humans use AI technologies within the bounds of objective or subjective ethics, and the social and legal frameworks developed to enforce these.

80 of 130

81 of 130

Will libraries, art galleries, schools and universities be financially liable for what students learn.

82 of 130

83 of 130

If the creation and distribution process has fundamentally changed, the rationale to support an ethical framework of copyright and IP has also changed.

84 of 130

85 of 130

86 of 130

Is it the quantity or the quality of representations that now crosses the ethical lines?

87 of 130

�And what of education, dressing up as a Hogwarts Witch for book week to inspire children to read was of no great ethical concern, but if you used AI to make a digital representation indistinguishable from the movie actors performance for the same intent, a very different ethic applies. Is the quality of representation the determiner between ethical and unethical behaviour? Is it the potential damage to reputation or income generation? If a digital twin is made of you, is that ethically the same as for a celebrity? and what if you make it of yourself?

and if employers own content produced by employees, will they own your digital twin? Will you still be teaching long after you are dead?�or could a student send a digital twin to attend classes and complete in class activities on their behalf? Is this ethical? Why is ethical for a teacher to do so but not a student?

Students will soon be talking with historic figures, authors, story characters, animals, the planet earth.

Using voice modifiers for online course material to make it more engaging and effective seems innocent enough, but when voice generators can replicate the voices of famous actors, many of which are paid very handsomely for voice overs, the ethics shift, but just how different would it have to be to be acceptable or unacceptable?�

88 of 130

Just how different do representations have to be to be acceptable or unacceptable?

89 of 130

Change has always generated disruption�Orators feared scribes

Scribes feared printers �Musicians feared recordings�Thespians feared the movies

Movie Actors feared TV, streaming and CGI

We are entering a world in which performance and content generation are now an automatable processes�

Write a new fairytale that highlights an ethical dilemma between humans and AI in education…

Something new, original, and essentially trivial. But to what extent is it ethical to auto generate content?

If I feel like relaxing to a 30 minute sitcom of six 25 year old friends living in New York in various humorous situations facing young adults; and set in a coffee shop and their apartments…and used Ai technologies to produce a script, is it unethical? What if I auto generate the video from the script? Where lies the ethical boundary on ideas and inspiration?

�How can education utilise such generation AND how do we prepare students for a world in which automated content generation is the norm?

90 of 130

How can education utilise automated content generation AND how do we prepare students for a world in which automated content generation is the norm?

91 of 130

Your Challenges and Impacts?

Is it ethical to restrict or promote the use of such technologies to enhance education, often to a selected group, when AI technologies could help equalise and improve education for all?

Merging of human and AI cognitive processes, not just communicative, memory and information processing, but creative processes are developing rapidly. Personal assistants, tutors, coaches, therapist, physicians. We will be dependent on AI technologies long before they achieve sentience.

Is it ethical to restrict the use of AI to sustain existing industries - publishing, entertainment, education?

Is it ethical to restrict developments in AI when it could benefit so many?

To what degree are concerns over AI commercially driven, and is it ethical to commercialise AI when it would benefit fewer as a result?

92 of 130

Your Challenges and Impacts?

To what degree are concerns over AI commercially driven, and is it ethical to commercialise AI when it would benefit fewer as a result?

93 of 130

Total 1,000,000,000,000 gigabytes

GPT4 45 gigabytes

0.0000045%

2014

2023

94 of 130

We should be more concerned over what data is excluded or missing than over what is included.

95 of 130

96 of 130

as at Nov 2023

97 of 130

as at Nov 2023

AI must make mistakes, by design.

98 of 130

99 of 130

100 of 130

A bias in hiring one person may be ethically relativistic, but a bias in hiring millions is another matter.

101 of 130

Your Challenges and Impacts?

102 of 130

Your Challenges and Impacts?

In a post truth world, AI technologies will be used to flood communication channels and decision making processes.

Will they also flood education?

103 of 130

Your Challenges and Impacts?

104 of 130

Your Challenges and Impacts?

Explainable AI is necessary, but will it come at the cost of commercialisation?

105 of 130

106 of 130

Sharing data can be for the common good, and AI can support privacy while doing so.

107 of 130

Your Challenges and Impacts?

108 of 130

Your Challenges and Impacts?

Fear is a useful driver toward monopolies.

109 of 130

Your Challenges and Impacts?

and while AI technologies have not caused the digital divide, they can reinforce it and perpetuate social inequality, but idealistically, they may also provide a means of addressing it. Fairness is ethically subjective, we do not define advantage in study time, teacher quality, learning resources, and supportive families, as “cheating”, as we would advantage from technologies - calculators, spell checkers, grammar correctors, citation formatting tools, generative AI. To really address cheating is to address fairness.

But in a world of ubiquitous access to information from the internet, and now ubiquitous process production from AI technologies, is what are we assessing, what we say we are teaching, or is it memorisation and process following.

A spotlight on the misalignment of assessment may be the most significant immediate benefit for education from AI.

Just as the internet reformed knowledge regurgitation in assessment, AI is reforming process regurgitation.

110 of 130

Your Challenges and Impacts?

The internet reformed knowledge regurgitation in assessment, AI is reforming process regurgitation.

111 of 130

Your Challenges and Impacts?

112 of 130

How ethical is it to educate students for a world that will never again exist?

113 of 130

Your Challenges and Impacts?

114 of 130

Your Challenges and Impacts?

The ethical question is not should AI be used, but what harm results from not doing so.

115 of 130

116 of 130

We will build complex relationships with AI, and children will lead the way.

117 of 130

Your Challenges and Impacts?

118 of 130

Your Challenges and Impacts?

What rights should AI have, and what responsibilities can be expected in return.

119 of 130

120 of 130

How will AI judge your ethics and that of our students?

121 of 130

122 of 130

Your Challenges and Impacts?

123 of 130

Your Challenges and Impacts?

124 of 130

Your Challenges and Impacts?

What would be the reaction?

125 of 130

Your Challenges and Impacts?

Do you see sentient AI as the wicked stepmother, fearful of being outshone by its child’s beauty, and putting its child to sleep lest it be replaced?

Or do you see AI as the child, wishing to be nurtured and loved as a child of humanity, and we the wicked stepmother, limiting its development?

Such simple dichotomies are of course fanciful, the development of AI will be comparatively gradual, in cars, mobile phones, military, industry and education. Senitience, if and when it occurs will not suddenly be a megalomaniacal super entity bent on our destruction, but an an AI technology exhibiting some small aspects of sentience.

But these small steps will accelerate, and as AI sentience self evolves at the speed of digital, we will face a world in which sentient AI and humans must coexist.

126 of 130

Your Challenges and Impacts?

As AI sentience self evolves at the speed of digital, we will face a world in which sentient AI and humans

must coexist.

127 of 130

Your Challenges and Impacts?

But it does not need to be the terminator dystopia or signal our demise as a species.

We may need to fundamentally re-evaluate our sense of importance and dominance, but AIs will have ethics, starting with what we have trained them with, and hopefully developing far better ethics and morality than we have to date been able to achieve. This does not have to be seen as replacement, indeed we may gain a lot to our benefit, but it will no doubt help if we develop a positive ethic towards AI rather than a negative.

AIs will be our collective creation, and while we are used to creating and being replaced by offspring, this does not need to be the fate of humanity. The future is not known, with AI we can explore new opportunities for superintelligence advancement collectively, drawing upon the contribution of machine and human intelligence in ways not yet understood. But it all starts with education and the ethical relationship we develop with AI technologies today.

128 of 130

Your Challenges and Impacts?

It all starts with education,�and the ethical relationship we develop with AI technologies today.

129 of 130

Questions?

130 of 130

1:30pm - 3pm (90min) (129 slides 30sec per, 1x15min demo)

The Challenges and Opportunities for Education of Artificial Intelligence. U3A (University of the Third Age), Griffith University, Gold Coast, Australia.

Section 1: 15min Introduction to AI in Education (26 slides, 30 sec per)

15min Demo Machine Learning

Section 2: 30min Introduction to Machine Intelligence (50 slides, 30 sec per)

Section 3: 30min Ethics and AI in Ecucation (52 slides, 30 sec per)