1) The keynote presentation discusses the need to update K-12 computing education from a focus on computational thinking to include data literacy and data agency given the rise of artificial intelligence and machine learning.
2) Examples are provided of emerging approaches to teaching AI/ML concepts to students, such as workshops exploring image recognition tools and having students design their own machine learning applications.
3) Significant changes are needed in K-12 computing education due to differences between classical programming paradigms and modern data-driven AI, including new technical concepts, problem-solving approaches, sources of problems, and ethical considerations around topics like algorithmic bias and data privacy.
Separation of Lanthanides/ Lanthanides and Actinides
K-12 Computing Education for the AI Era: From Data Literacy to Data Agency
1. K-12 Computing Education for the AI Era:
From Data Literacy to Data Agency
Matti Tedre & Henriikka Vartiainen
University of Eastern Finland
AI
July 10, 2023: ACM ITiCSE Keynote
2. Generation AI
A 6-year, 5.4M€ project
https://www.generation-ai-stn.fi/
● Univ. of Eastern Finland (CS, Edu, Law)
● Univ. of Oulu (Edu)
● Univ. of Helsinki (Law, CS, Edu)
AI
https://tinyurl.com/iticse23
3. ●
Why should everyone
know about computing?
A 1999 NRC report:
AI
● Personal rationale
(making things easier for you)
● Workforce rationale
(readiness for computing jobs)
● Educational rationale
(enabler of opportunities)
● Societal rationale
(informed citizenry)
● Technological rationale
○ Avoid uncritical adoption
○ Safe, effective, flexible use
○ Empowerment & effective
harnessing of technology
4. ●
Why should everyone
know about computing?
Computing Ethics
AI
● Ethical decisions are always
highly contextual
● Ethical questions are different in…
○ …art education
○ …health
○ …democracy
○ …physics education
● …and they require understanding
the underlying principles and
technology
6. So what should we do
about it?
A 1954 advice:
AI
● “The individual […] must be able
to visualize and express the
solution to a problem as a
sequence of logical operations.”
(p.13)
7.
8. So what should we do
about it?
AI
● 1960s: “Algorithmizing”
● 1970s: “Algorithmic thinking”
● 1980s: “Procedural thinking”
● 1990s: “Fluency with IT”
● 2000s: “Computational thinking”.
9. A 2006 “how-to”:
Computational
thinking
AI
● Primarily a K–12 education
movement
● Aims to teach learners important
skills and knowledge for 2020s
● Understand how to get computers
to do jobs for us
● See the world through a
computational lens
● Often taught using
classical/block-based
programming
11. AI
K-12 “Hour of Code” CT’s
dominant paradigm
Implicitly assumes a view of programs
as…
1. …deep chains of consecutive
steps,
2. processed serially by the
processor,
3. taking limited input data, that are
4. processed using small number of
variables
5. relying on Boolean logic and
arithmetic to establish strong
correlations between variables.*
*See B.C. Smith (2019)
12. AI
K-12 “Hour of Code” CT’s
dominant paradigm
Deterministic behavior
Stepwise execution
Discrete program states
Unambiguous transition rules for program flow
Strict syntax
Well-established notional machines
Deductive problem solving
Glass-box testing
Tracking and tracing of program states
Highly structured programs and data
Discrete data (sets of symbols)
13. AI
K-12 “Hour of Code” CT’s
problem-solving workflow
Formalize the problem statement in a specific
way (amenable to rule-based solution)
Design an algorithmic solution
Implement the solution as a stepwise program
Compile and execute the program
Test, debug, and go back as necessary
14. Challenge: Imagine you
as a K–12 teacher in
technology education
AI
Imagine having to explain, in terms of
classical programming…
● …how can your phone learn to
recognize its owner to unlock
● …how can TikTok know exactly
what you’d like to watch
● …how can image services label
what’s in the picture
● …etc.
Imagine teaching how to write that
program in Java
15. The better the service
the higher the chance it
uses some form of AI
Google’s AI-enhanced products:
● Assistant
● Cloud
● Clips
● Android
● Maps
● Search
● Home
● Gmail
● Cardboard
● Ads
● Youtube
● Translate
● Photos
● Drive
● Play
16. How well is CT doing
in 2023?
AI
● Personal rationale
(making things easier for you)
● Workforce rationale
(readiness for computing jobs)
● Educational rationale
(enabler of opportunities)
● Societal rationale
(informed citizenry)
● Technological rationale
○ Avoid uncritical adoption
○ Safe, effective, flexible use
○ Empowerment & effective
harnessing of technology
17. AI
The ability of classical
programming to help one
understand one’s digital
world is challenged by
data-driven approaches
18. AI
Calls for research on how
to teach data-driven
computing, or AI/ML, to
children are becoming
more common
21. Example 1:
NEW SET OF CONCEPTS AND
INSIGHTS TO BE LEARNED
● Gesture recognition
● Training data
○ Example (sample)
○ Name (class, label)
● Confidence
● Accuracy
● Image removed
23. Example 2:
NEW SET OF CONCEPTS AND
INSIGHTS TO BE LEARNED
● Image recognition
● Feature
● Nearest neighbor
● Training data
● Class, name (label)
● Feature space, distance
● “Cluster”
Mariescu-Istodor & Jormanainen (2019)
25. (Applied from Vartiainen et al., 2021)
1. Workshop
BASIC CONCEPTS AND
MECHANISMS OF AI
2. Workshop
DATA-DRIVEN DESIGN
3. Workshop
ETHICS AND
IMPACTS OF AI
AFFORDED TECHNOLOGIES:
GenAI Classifier
Midjourney
SOCIAL ENVIRONMENT:
Working in small groups
LEARNING TASK:
Co-design ML-based application
1. Contextualizing AI
2. Exploring ML-based
educational technologies
3. Development of
app ideas
4. Construction of
solutions
5. Sharing and reflecting
the process of design
and learning
6. Exploring the impact
and ethics of AI
26. Goals of the first workshop:
● Discuss on AI, its applications,
and its impact on everyday lives
● Introduce basic concepts of AI
● Engage children in hands-on
exploration and experimentation
with the classifier tool
Guiding questions:
● How and why ML systems are used
in everyday life?
● What kind of data are collected of
me and how is it used?
● What risks are associated with the
use of data collected about me?
First workshop:
EXPLORING THE BASIC CONCEPTS AND MECHANISMS OF ML
27. Guiding questions for becoming familiar
with the ML tools:
How do ML systems work? How can I train a
classification system?
What are the key concepts and steps in
teaching a classifier?
In which situations does the taught model
work well and where does it not? What
happens if you test the trained model on an
object that was not in the original training
data? Why does this happen?
What happens if the background changes?
Why does this happen?
First workshop:
EXPLORING THE BASIC CONCEPTS AND MECHANISMS OF ML
28. Pope et al. (forthcoming)
New educational technology
29. Goals of the second workshop:
● Facilitate ideation, design, and implementation of
children's own ML applications
● Enhance children's understandings of the core
concepts and workflows of ML
Second workshop:
DATA-DRIVEN DESIGN
30. Guiding questions for app design:
PROBLEM DEFINITION: What does the application do? What things does it need to be able to distinguish from
each other?
SELECTION OF TRAINING DATA: How many different classes does the application recognize? In what ways are
the classes sufficiently distinct from a computer's perspective? Where is the training data collected from?
TRAINING: Under what conditions is the training data presented to the computer (e.g., background)? How is the
quantity and quality of examples taken into account when selecting the training data?
TESTING THE CLASSIFIER: How well does the classifier work? How confident is the classifier in classifying
examples that were not present in the training data?
DEVELOPMENT: In what situations does the classifier fail to work? Why? How could the training data be
improved to enhance the classifier's performance?
DEFINING THE ACTIONS AND OUTPUT: What kind of things does the application do when recognizing each
class?
TESTING THE APPLICATION: Does the application do what you intended it to do? Why?
Second workshop:
DATA-DRIVEN DESIGN
31. Third workshop:
IMPACT AND ETHICS OF AI
Goals of the third workshop:
● Sharing and presenting the group's application and project
● Analyzing the benefits, risks, and impacts of ML
Guiding questions for reflection:
● TRAINING PROCESS: What was included in the training data and how was the
data classified? In which situations does the application perform well, and in which
situations does it not? How could the application be improved?
● CONSIDERATION OF BENEFITS AND HARMS: Who could benefit from this
application and how?What harm or negative impact could the application cause
and to whom?
● REFLECTION ON THE LEARNING PROCESS: What are the key knowledge and
skills that were learned during the project? What remained unlearned?
32. Third workshop:
IMPACT AND ETHICS OF AI
Guiding questions for exploring algorithmic biases:
● Describe bias: What is it like? What causes it?
● How can this bias be corrected?
33. Third workshop:
IMPACT AND ETHICS OF AI
“With ‘woman’ prompt all pictures had the same skin tone
and hair color. Everything looked kind of pretty the same.
The web has more pictures of specific kinds of women.“
“You should upload to the web more pictures of
different kinds of women. The bias can create body
appearance pressure”
34. Alluvial plot of changes in
students’ responses from the
pre-test to the post-test for the
4th graders (left) and 7th graders
(right).
Vartiainen et al. (submitted)
35. Example 3:
NEW SET OF CONCEPTS AND
INSIGHTS TO BE LEARNED
● Classifier
● Class, name (label)
● Example (sample)
● Training data
● Curation
● Training
● Input, output
● Confidence
● Actions
● Output
● Deployment
● Softness, brittleness
● Algorithmic bias
Pope et al. (2023, forthcoming)
37. Example 4:
NEW SET OF CONCEPTS AND
INSIGHTS TO BE LEARNED
● Training data
○ Web crawlers
○ Image-label pairs
● Diffusion model
● Bias
○ Misrepresentation
○ Overrepresentation
○ Underrepresentation
● Automation of creative work
● Copyright
● Ownership
● Genre hijacking
● Model collapse
39. Example 5:
NEW SET OF CONCEPTS AND
INSIGHTS TO BE LEARNED
● Automation of knowledge work
● Tracking, profiling, modeling,
predicting
● Algorithmic influencing
○ Behavior engineering
○ Targeted advertisement
○ Opinion swaying
● Bias
● Hallucination
● Risks to democracy
42. AI
Implicitly assumes a view of programs
as…
1. …deep chains of consecutive
steps,
2. processed serially by the
processor,
3. taking limited input data, that are
4. processed using small number of
variables
5. relying on Boolean logic and
arithmetic to establish strong
correlations between variables.*
*See
B.C.
Smith
(2019)
K-12 “Hour of Code” CT’s
dominant paradigm
43. AI
K-12 “Hour of Code” CT’s
dominant paradigm
Implicitly assumes a view of programs
as…
1. …deep chains of consecutive
steps,
2. processed serially by the
processor,
3. taking limited input data, that are
4. processed using small number of
variables
5. relying on Boolean logic and
arithmetic to establish strong
correlations between variables.*
*See
B.C.
Smith
(2019)
One alternative view on
K-12 ML education’s
emerging paradigm
In many cases assumes certain types of
neural networks, which involve…
1. …short chains of stacked layers
of networks of computations,
2. processed parallel on GPUs,
3. taking massive amounts of data
4. processed through very large
number of variables (parameters)
5. that are weakly correlated with
each other.*
44. AI
Problem-solving workflow
Rule-driven
Formalize the problem
Design an algorithmic solution
Implement the solution in a stepwise program
Compile and execute the program
Test, debug, go back as necessary
Data-driven
Describe the problem and data needed for
automation
Collect data from the intended context. Filter and
clean data. Label data.
Train a model from the available data
Evaluate and use the model
Test, re-curate data, find a fit between data,
architecture, hyperparameters, go back as n.
45. AI
Making programs in the
K–12 classroom
“Classical programming”
Strict, structured syntax
Structured data, strict semantics
Symbol processing
Stepwise execution
Unambiguous transition rules for program flow
Making apps with GenAI TM
No-code
Flexible with data, loose semantics
Glorified curve fitting
Passing data through a neural network
No loops, branches (in FF ANNs)
46. AI
Testing and debugging
learners’ programs
“Classical programming”
Black / glass box cross checking of outputs vs.
code
Tracking and tracing program states and code
Great program visualization tools
Often an idea of correct program
Making ML-based apps
Always black-boxed
Trial & error; Experimenting with data,
parameters, and hyperparameters.
XAI is underdeveloped
Probably approximately correct
Higher or lower confidence
47. AI
The philosophy of
programming
Deductive reasoning
Reductionism
Determinism
Type of notional machine:
Stepwise, deterministic, discrete flow of program
through states (as contents of memory
locations).
Inductive reasoning
Emergence
(Some) non-deterministic behavior
Example notional machine:
Parallel (possibly nondeterministic) passing of
data through a network
48. AI
New sources of problems
Softness
Brittleness
Opacity
Data hunger
Spoofability
Shallowness
49. AI
New ethical questions /
perspectives to old ones
Algorithmic bias, reinforcing prejudices
Sources of data, web scraping, copyrights,
ownership
Privacy, surveillance, tracking
Profiling, modeling, predicting
Behavior engineering, swaying opinion, etc.
etc. etc.
50. ●
Are we making
progress towards
these?
AI
● Personal rationale
(making things easier for you)
● Workforce rationale
(readiness for computing jobs)
● Educational rationale
(enabler of opportunities)
● Societal rationale
(informed citizenry)
● Technological rationale
○ Avoid uncritical adoption
○ Safe, effective, flexible use
○ Empowerment & effective
harnessing of technology
51. ●
From literacy to
agency
AI
Data agency:
● People’s volition and capacity for
informed actions that make a
difference in their digital world.
● Emphasizes people’s ability to not
only understand data, but also to
○ actively control and
manipulate information
flows and
○ to use them wisely and
ethically.
54. More reading
● Bommasani, R. et al. (2021) On the
opportunities and risks of foundation
models. https://arxiv.org/abs/2108.07258
● Martins, R. M. and Gresse Von
Wangenheim, C. (2022). Findings on
teaching machine learning in high
school: A ten - year systematic
literature review. Informatics in
Education.
● Schulte, C., Fleischer, Y., Höper, L.,
Biehler, R., Frischemeier, D., Hüsing, S.,
and Podworny, S. (2022). Exploring the
data-driven world: teaching AI and ML
from a data-centric perspective. In AI,
data science, and young people.
● Heintz, F. and Roos, T. (2021). Elements
of AI - teaching the basics of AI to
everyone in Sweden. In Proceedings of
the 13th International Conference on
Education and New Learning
Technologies (EDULEARN21).
● Mariescu-Istodor, R. and Jormanainen, I.
(2019). Machine learning for high
school students. In Proceedings of the
19th Koli Calling ’19, New York, NY, USA.
● Lindner, A., Seegerer, S., and Romeike, R.
(2019). Unplugged activities in the
context of AI. In Pozdniakov, S. N. and
Dagiene ̇, V., Informatics in Schools. New
Ideas in School Informatics, 123–135.
55. More reading
● Vartiainen, H., Pellas, L., Kahila, J.,
Valtonen, T., and Tedre, M. (2022).
Pre-service teachers’ insights on data
agency. New Media & Society, 1–20.
● Vartiainen, H., Toivonen, T., Jormanainen,
I., Kahila, J., Tedre, M., and Valtonen, T.
(2021). Machine learning for middle
schoolers: Learning through
data-driven design. International Journal
of Child-Computer Interaction, 29:100281.
● Tedre, M., Denning, P. J., and Toivonen, T.
(2021). CT 2.0. In 21st Koli Calling
International Conference on Computing
Education Research, Koli Calling ’21.
● Tedre, M., Toivonen, T., Kahila, J.,
Vartiainen, H., Valtonen, T., Jormanainen,
I., and Pears, A. (2021). Teaching
machine learning in K–12 classroom:
Pedagogical and technological
trajectories for artificial intelligence
education. IEEE Access,
9:110558–110572.
● H. Vartiainen, M. Tedre, Using artificial
intelligence in craft education: Crafting
with text-to-image generative models.
Digital Creativity, 2023.