Blackbox of AI

Question 1

What is the primary focus of the paper ‘Unpacking the “Black Box” of AI in Education’?

Answer 1

To clarify what AI is and its potential to advance or hamper educational opportunities

The paper discusses methods, applications, limitations, and risks of AI in education.

Question 2

What are the two schools of AI frequently used in education?

Answer 2

Machine learning and rule-based AI

These schools represent different approaches to implementing AI in educational contexts.

Question 3

Define ‘supervised learning’ in the context of machine learning.

Answer 3

A method where machines learn from a historical dataset with known outputs to predict future outcomes

It involves using labeled data to train models.

Question 4

What is the purpose of ‘unsupervised learning’?

Answer 4

To perform statistical pattern recognition without access to ground-truth labels

Commonly used for clustering similar data points.

Question 5

What is ‘reinforcement learning’?

Answer 5

A machine learning paradigm that uses rewards to encourage desired behaviors based on input states

It is often applied in intelligent tutoring systems.

Question 6

What are the two main philosophies of machine learning?

Answer 6

Frequentist and Bayesian

These philosophies influence how inferences and predictions are made from data.

Question 7

What is a key characteristic of Bayesian machine learning models?

Answer 7

They incorporate pre-existing beliefs alongside training data

This approach can help improve predictions, especially in sparse datasets.

Question 8

What is the significance of deep learning in AI?

Answer 8

It has become the dominant approach due to its ability to learn complex relationships through neural networks

Deep learning involves stacking neural networks to enhance learning capacity.

Question 9

What are some common architectures used in deep learning?

Answer 9

• Recurrent Neural Networks (RNNs)
• Convolutional Neural Networks (CNNs)
• Transformers

Each architecture has its own strengths and applications, such as image processing or natural language processing.

Question 10

True or False: The ‘I’ in AI systems is highly sophisticated.

Answer 10

False

Many AI systems still perform poorly on tasks that humans find intuitive.

Question 11

What risks are associated with AI in education?

Answer 11

• Failures to generalize
• Inability to identify causal relationships
• Potential for perpetuating unfair applications

These limitations can hinder the effective use of AI in educational contexts.

Question 12

Fill in the blank: AI refers to a collection of methods, capabilities, and _______.

Answer 12

limitations

Understanding these limitations is crucial for the effective application of AI.

Question 13

What do the authors hope to achieve by unpacking the ‘Black Box’ of AI?

Answer 13

To make terms and concepts accessible for all stakeholders in education

This aims to empower educators and researchers to engage with AI development.

Question 14

What is a common application of clustering in education?

Answer 14

To develop a typology of students based on various characteristics

This helps in designing targeted support for students.

Question 15

What is the role of historical datasets in supervised learning?

Answer 15

They provide inputs and outputs that help train the model

The model learns how features relate to target attributes.

Question 16

True or False: The terms and concepts of AI are universally understood among educators.

Answer 16

False

Many educators may find the rapidly advancing field of AI inaccessible without proper training.

Question 17

What does the term ‘machine learning’ refer to?

Answer 17

A subset of AI focused on algorithms that learn from data to make predictions or decisions

It encompasses various methodologies, including supervised and unsupervised learning.

Question 18

What are the three major architectures in deep learning mentioned?

Answer 18

RNNs, CNNs, Transformers

RNNs are suited for time-series data, CNNs for image processing, and Transformers are used in natural language processing tasks.

Question 19

What is the purpose of the GPT-3 language model?

Answer 19

To predict the next word in a corpus of text given a sequence of preceding words

GPT-3 is trained on over 570 gigabytes of text from the internet.

Question 20

What is transfer learning?

Answer 20

A technique that allows a model to pre-train using data from a related task and then fine-tune on a smaller dataset

This is useful when large datasets are not available for a specific task.

Question 21

Name two recent hardware accelerations that have impacted deep learning.

Answer 21

GPUs, TPUs

GPUs and TPUs enable more time-efficient computation for deep learning tasks.

Question 22

True or False: Rule-based AI systems infer rules from data.

Answer 22

False

Rule-based AI uses pre-defined logical propositions rather than inferring rules from patterns in data.

Question 23

What is a significant limitation of naive rule-based AI algorithms?

Answer 23

They may take an impractical amount of time to compute optimal solutions

Evaluating every possible combination can be infeasible for large real-world problems.

Question 24

What are intelligent tutoring systems (ITS)?

Answer 24

AI tools that adapt to students’ knowledge and skills to personalize learning

They can be based on machine learning or predefined rules.

Question 25

How do ITS determine a student's learning edge?

Answer 25

By providing problems they are most likely to answer correctly based on their history of responses ## Footnote This maximizes the effectiveness of learning interventions.

Question 26

What is the primary function of AI in assessment and feedback?

Answer 26

To infer students' knowledge states from data on digital learning platforms ## Footnote This could reduce the need for formal assessments.

Question 27

What does automated assessment of writing typically focus on?

Answer 27

Foundational attributes like spelling, vocabulary, and grammar ## Footnote More complex feedback on writing development is also necessary for growth.

Question 28

What role do AI chatbots play in coaching and counseling?

Answer 28

They assist students by answering questions and providing reminders ## Footnote Chatbots can facilitate communication and support enrollment processes.

Question 29

What is the goal of strategy-proof algorithms in school choice systems?

Answer 29

To prevent families from gaming the system for school placements ## Footnote This aims to ensure equity in school selection.

Question 30

What are early warning systems in education designed to do?

Answer 30

Predict which students are at risk of failing or dropping out ## Footnote These systems use historical data to make predictions.

Question 31

Fill in the blank: Transfer learning can help a model _______ itself using outputs from a related task.

Answer 31

pre-train

Question 32

True or False: Rule-based AI systems improve with more data.

Answer 32

False ## Footnote Rule-based systems do not necessarily make better decisions with increased scale or diversity of data.

Question 33

What type of learning do ITS often utilize to maximize student learning outcomes?

Answer 33

Machine learning ## Footnote They adapt based on predictions of student performance.

Question 34

What are early warning systems in the context of education?

Answer 34

Systems that predict student outcomes, such as exam failure or dropout rates, using historical data ## Footnote These systems often utilize regression techniques and can help reduce chronic absenteeism and course failure.

Question 35

What is a potential drawback of machine learning-based early warning systems?

Answer 35

They may lead to tracking, which limits a student’s ability to explore new topics ## Footnote Tracking can discourage students from taking advanced classes based on predictions.

Question 36

What is the 'cold start' problem in small school districts?

Answer 36

The lack of sufficient historical data to train accurate machine learning models ## Footnote Small districts may need to borrow data from other districts to improve prediction accuracy.

Question 37

What is one challenge school leaders face when using predictive models?

Answer 37

Determining the appropriate threshold for intervention based on predicted probabilities of dropout ## Footnote For example, deciding if intervention should occur at a 20% or 90% dropout probability.

Question 38

What is a major limitation of neural network approaches in machine learning?

Answer 38

Lack of transparency and interpretability ## Footnote This makes it difficult to understand which inputs influenced decisions.

Question 39

What is the difference between correlation and causation in machine learning?

Answer 39

Machine learning identifies correlations but does not necessarily explain causal relationships ## Footnote This can lead to interventions that fail to address root causes of issues.

Question 40

What is abstract reasoning?

Answer 40

The ability to ascertain fundamental rules of a task and apply them to different situations ## Footnote Unlike machines, humans excel at this type of reasoning.

Question 41

What does 'meta learning' refer to in machine learning?

Answer 41

The process by which a machine learns how to learn ## Footnote Current machine meta learning lacks the higher-order thinking present in human learning.

Question 42

What does it mean when a machine learning model 'catastrophically forgets'?

Answer 42

The model fails to retain knowledge of tasks it was trained on when exposed to new data ## Footnote This can occur when the model is not robust enough to generalize.

Question 43

What is a risk associated with adversarial inputs in machine learning?

Answer 43

They can fool models into making incorrect classifications ## Footnote For example, altering a training image slightly to cause misclassification.

Question 44

What is a significant ethical concern regarding machine learning in education?

Answer 44

The potential for bias and unfairness in predictions and interventions ## Footnote AI systems can replicate and scale biases present in training data.

Question 45

What questions should educationalists ask about AI applications in education?

Answer 45

1. What kind of AI is it? 2. Does the AI enable something difficult to achieve otherwise? 3. What are the potential risks? 4. How equitably are the benefits and risks distributed? 5. If you could change anything about this technology, what would it be? ## Footnote These questions help ensure ethical and responsible use of AI.

Question 46

True or False: Machine learning models can inherently understand the context behind data.

Answer 46

False ## Footnote They often lack a deep understanding of the tasks they perform.

Question 47

What is the main focus of conversations between AI researchers and educationalists?

Answer 47

Understanding the theory and practice of education and AI.

Question 48

Why is it important for educationalists to understand AI?

Answer 48

To critique, reject, or adapt AI for their own efforts.

Question 49

What does the article aim to provide an overview of?

Answer 49

Current AI techniques, their use in education, and key limitations and risks.

Question 50

What is the primary goal of AIED?

Answer 50

To improve the human condition.

Question 51

True or False: The authors report conflicts of interest.

Answer 51

False.

Question 52

What does the Early Warning Intervention and Monitoring System aim to do?

Answer 52

Get students on track for graduation.

Question 53

What does the term 'datafication of education' refer to?

Answer 53

The process of turning educational practices and interactions into data.

Question 54

What does the term 'algorithmic accountability' refer to?

Answer 54

The responsibility of algorithms to be transparent and fair.

Question 55

What significant release did OpenAI announce in 2019?

Answer 55

GPT-2: 1.5B Release ## Footnote OpenAI Blog