WorksheetsModule 1 – Core Concepts of AI
Total questions: 106
Worksheet time: 53mins
What best describes artificial intelligence?
A single software product
A branch of computer science focused on simulating intelligent behaviour
Any automated workflow
A deterministic decision system
Which element is commonly found in definitions of AI?
Forecasting
Intelligence
Compliance
Profit optimisation
Which characteristic most differentiates AI from traditional automation?
Speed
Accuracy
Learning and adaptation
Cost efficiency
AI systems are typically designed to be:
Random
Goal-directed
Fully transparent
Deterministic
Why has the definition of AI evolved over time?
Laws have changed
Computing costs decreased
Societal expectations of intelligence have changed
AI no longer uses data
Which output is commonly generated by AI systems?
Legal opinions
Predictions
Policies
Guarantees
Which feature of AI increases governance complexity?
Low cost
Opacity
Documentation
Standardisation
What does AI autonomy refer to?
AI operating without electricity
AI making decisions without direct human oversight
AI replacing humans entirely
AI operating legally
Why does speed and scale matter for AI governance?
It reduces system costs
It limits deployment
Errors can propagate rapidly
It improves explainability
What does probabilistic output mean?
Outputs are always random
Outputs are based on likelihood, not certainty
Outputs are unverifiable
Outputs are human-approved
Which category of AI currently exists in real-world use?
AGI
ASI
ANI
Sentient AI
AGI is best described as:
AI that performs one task well
AI with human-level general intelligence
AI that operates autonomously
AI used in healthcare
Why are AGI and ASI important to governance discussions?
They are widely deployed
They define current compliance rules
They inform future risk planning
They replace ANI
Broad AI is best described as:
Fully human-like AI
A single narrow model
Multiple AI systems working together
Non-learning automation
AI agents are an example of:
ANI
Broad AI
AGI
ASI
Why do autonomous vehicles qualify as broad AI?
They use sensors
They combine multiple AI systems
They are fully sentient
They eliminate human oversight
What is the purpose of the OECD AI classification framework?
Certify AI vendors
Replace national laws
Support common understanding and risk assessment
Rank AI performance
Which is NOT a goal of the OECD framework?
Inform AI inventories
Support sector-specific frameworks
Guarantee AI safety
Promote common understanding
Which OECD dimension relates to how AI decisions affect society?
AI model
Tasks and output
Economic context
Data input
Which dimension covers what the AI system does?
Data and input
Tasks and output
People and planet
Economic context
Why is data dependency a governance concern?
Data is expensive
Data quality affects outcomes and bias
Data is always personal
Data eliminates risk
Which megatrend most directly enables scalable AI deployment?
Blockchain
Cloud computing
Social media
Robotics
How does IoT interact with AI?
It limits data collection
It provides large volumes of real-time data
It replaces machine learning
It reduces governance needs
Why do mobile technology and social media matter for AI?
They reduce bias
They generate large behavioural datasets
They replace cloud computing
They ensure compliance
Privacy-enhancing technologies are relevant because they:
Remove the need for AI governance
Enable AI use while reducing privacy risks
Replace consent
Eliminate data use
Blockchain’s relationship to AI is best described as:
Competitive
Unrelated
Complementary in some use cases
Required
Why do autonomous weapons raise governance concerns?
Low accuracy
Limited scale
Potential for severe harm
Lack of data
Computer vision is commonly used for:
Legal reasoning
Image and pattern recognition
Policy drafting
Encryption
One benefit of AI in healthcare is:
Eliminating doctors
Faster and more accurate scan analysis
Guaranteed diagnoses
Removing liability
AI’s similarity to big data lies in:
Storage requirements
Volume, velocity, and variety of data processing
Legal frameworks
Manual review
Why can AI reduce human error?
AI is unbiased
AI automates repetitive tasks
AI is deterministic
AI replaces judgement
Why can AI also introduce bias?
AI lacks goals
AI depends on training data
AI is autonomous
AI is opaque
Recognition use cases include:
Route optimisation
Image or speech recognition
Forecasting demand
Risk scoring
Forecasting use cases include:
Facial recognition
Predicting demand or trends
Chatbots
Optimisation
Personalisation use cases aim to:
Standardise outputs
Tailor experiences to individuals
Remove humans
Reduce data use
Goal-driven optimisation focuses on:
Detecting fraud
Maximising outcomes under constraints
Interpreting text
Ensuring fairness
Recommendation systems are an example of:
Deterministic automation
Goal-driven optimisation
Static rules
Manual decision-making
Why is human interaction important in AI?
AI cannot operate alone
Humans shape outcomes and impacts
Humans approve every output
Humans reduce accuracy
Which challenge is central to AI governance?
Eliminating innovation
Balancing innovation and risk management
Avoiding all automation
Reducing computation
Why is understanding AI characteristics critical?
To improve marketing
To manage inherent risks
To maximise profit
To replace compliance
Which characteristic increases misuse risk?
Transparency
Autonomy
Documentation
Testing
What makes AI governance different from traditional IT governance?
Lower costs
Probabilistic outputs and learning behaviour
Shorter lifecycles
Static behaviour
AI governance must consider:
Only technology
Only data
Technical and societal impacts
Only legal compliance
Why is opacity problematic?
It slows systems
It limits explainability and accountability
It increases cost
It prevents deployment
What role does learning play in AI risk?
It removes risk
It can change behaviour post-deployment
It ensures fairness
It simplifies monitoring
Which statement is true about AI and humans?
AI replaces all human judgement
AI outcomes are shaped by human choices
AI eliminates governance
AI is value-neutral
Which AI characteristic most challenges predictability?
Speed
Probabilistic behaviour
Data storage
Hardware
Why must AI governance be lifecycle-based?
AI is expensive
AI behaviour can change over time
AI is always autonomous
AI is fully transparent
Which best describes AI’s societal impact?
Fixed and predictable
Independent of humans
Influenced by deployment and use
Limited to technology teams
What is the key takeaway of Module 1?
AI is only a technical issue
AI requires governance due to unique characteristics and risks
AI eliminates human decision-making
AI governance replaces innovation
What best defines an AI model?
A dataset used to train algorithms
A program that applies algorithms to data to make predictions or decisions
A computing infrastructure
A user interface
What does the term algorithm refer to?
A trained AI system
A set of instructions or rules to solve a problem
A data repository
A computing environment
Which term describes the full operational environment including models, data, and infrastructure?
Model
Algorithm
Dataset
System
Machine learning differs from traditional programming because it:
Uses fixed rules
Learns patterns from data
Requires no data
Is deterministic
Which relationship between AI categories is correct?
AI⊂ML⊂DL⊂GenAI
ML⊂AI⊂DL⊂GenAI
GenAI⊂DL⊂ML⊂AI
DL⊂GenAI⊂ML⊂AI
What is a defining characteristic of deep learning?
Rule-based inference
Multi-layer neural networks
Small datasets
Deterministic outputs
One advantage of deep learning over traditional ML is:
Lower data requirements
Manual feature extraction
Automatic feature learning
A key limitation of deep learning models is that they:
Cannot process images
Require large amounts of data and compute
Do not scale
Are rule-based
Generative AI systems are designed to:
Optimise numerical values
Classify existing data
Generate new content
Apply business rules
Which is a common ethical risk of generative AI?
Overfitting
Deterministic bias
Misinformation
Feature scaling
Agentic AI systems are characterised by:
Static outputs
Autonomous decision-making and action
Human-in-the-loop control only
Rule-based logic
Which learning approach uses labelled data?
Unsupervised learning
Reinforcement learning
Supervised learning
Agentic learning
A challenge of supervised learning is:
Low accuracy
Lack of scalability
Need for labelled data
No structure
Unsupervised learning is best suited for:
Regression tasks
Customer segmentation
Policy enforcement
Classification with labels
Reinforcement learning differs from supervised learning because it:
Uses labelled data
Learns through rewards and penalties
Is deterministic
Requires no environment
Which algorithm is commonly used for numeric prediction?
Logistic regression
Linear regression
Decision trees
CNNs
Logistic regression is primarily used for:
Continuous prediction
Binary classification
Image recognition
Text generation
Random forests improve performance by:
Using a single tree
Combining multiple decision trees
Eliminating bias
Reducing training data
Neural networks are especially useful for:
Simple rules
Complex pattern recognition
Data storage
Compliance checks
Transformer models are important because they:
Process data sequentially only
Capture contextual relationships efficiently
Require labelled datasets only
Are deterministic
Multimodal models differ from language models because they:
Process only text
Handle multiple data types
Are smaller
Eliminate bias
Which governance concern is heightened by multimodal models?
Performance tuning
Privacy risk
Cost reduction
Deployment speed
Retrieval-augmented generation improves GenAI by:
Reducing model size
Incorporating external information
Eliminating hallucinations completely
Replacing training
Proprietary models typically:
Are fully transparent
Are controlled by vendors
Require no governance
Cannot be deployed
Open-source models raise governance concerns because:
They are illegal
Accountability may be unclear
They do not scale
They cannot be secured
Large language models differ from small ones mainly in:
Existence
Parameter count and resources
Output type
Legal status
Small language models are often preferred when:
Broad versatility is required
Resources are constrained
Multimodality is required
Data is unlimited
Language models are best suited for:
Image recognition
Text-based tasks
Sensor fusion
Video generation
Why might organisations combine multiple model types?
To avoid governance
To handle complex tasks
To reduce costs only
To eliminate humans
Expert systems differ from ML models because they:
Learn autonomously
Use rule-based inference
Require large datasets
Are probabilistic
The inference engine in an expert system:
Stores data
Applies rules to reach conclusions
Collects sensor inputs
Trains models
Why are expert systems more explainable?
They are smaller
They are rule-based
They are open source
They are generative
Which model type raises the greatest opacity concerns?
Linear regression
Decision trees
Deep neural networks
Expert systems
Why do large models increase bias risk?
They are deterministic
They rely on massive datasets
They use rules
They are transparent
Which learning method involves exploration vs exploitation?
Supervised
Unsupervised
Reinforcement
Semi-supervised
Why is reinforcement learning risky in high-impact domains?
It lacks labels
It may learn harmful behaviours
It cannot scale
It is deterministic
Which architecture is best for image recognition?
Linear regression
CNN
Logistic regression
RNN
Graph neural networks are best suited for:
Text generation
Relationship-based data
Image processing
Regression
Why does agentic AI raise governance concerns?
Low accuracy
Autonomous actions
Limited data
Deterministic logic
Which is a key governance risk of proprietary GenAI?
Low innovation
Limited transparency
Poor performance
No data
Multimodal models raise additional privacy risks because they:
Are smaller
Combine multiple data types
Are open source
Are deterministic
Why should governance professionals understand ML basics?
To build models
To audit source code
To assess risk and engage technical teams
To replace engineers
Key takeaway of this module is that:
All models are equivalent
Model types determine capabilities and risks
Governance is purely technical
Larger models are always better
Which training method is most appropriate when outcomes are unknown and patterns must be discovered?
Supervised learning
Reinforcement learning
Unsupervised learning
Deterministic programming
Semi-supervised learning is valuable because it:
Eliminates bias
Requires only labelled data
Reduces the need for extensive manual labelling
Produces deterministic results
Which model type is most likely to require the greatest computational resources?
Linear regression
Decision trees
Large language models
Expert systems
Why are foundation models important in modern AI systems?
They replace governance
They enable reuse across many tasks
They eliminate training data
They are deterministic
Which AI architecture is best suited for sequential data such as text or speech?
CNN
RNN
Linear regression
Decision trees
From a governance perspective, why does model size matter?
It affects branding
It influences risk, cost, and bias exposure
It guarantees accuracy
It eliminates oversight
Which statement best reflects responsible AI model selection?
Always choose the most advanced model
Select models aligned to purpose, risk, and constraints
Prefer open source in all cases
Avoid multimodal models
What best defines an AI model?
A dataset used to train algorithms
A program that applies algorithms to data to make predictions or decisions
A computing infrastructure
A user interface
What does the term “algorithm” refer to?
A trained AI system
A set of instructions or rules to solve a problem
A data repository
A computing environment
Which term refers to the full operational environment of AI?
Model
Algorithm
Dataset
System
Machine learning is best described as:
Explicitly programmed decision-making
Learning patterns from data without explicit programming
Manual feature engineering
Deterministic automation
Which category relationship is correct?
AI ⊂ ML ⊂ DL ⊂ GenAI
ML ⊂ AI ⊂ DL ⊂ GenAI
GenAI ⊂ DL ⊂ ML ⊂ AI
DL ⊂ GenAI ⊂ ML ⊂ AI
What distinguishes deep learning from traditional ML?
Deterministic outputs
Multi-layered neural networks
Rule-based logic
Manual feature engineering
