Artificial Intelligence AI Ethics and Bias 3 — Questions and Answers
Question 1: What is 'hallucination' in large language models?
- The model generating images during text tasks
- Generating confident but factually incorrect or fabricated information (Correct answer)
- A training instability that causes random outputs
- Repeating the same token endlessly
Correct answer: Generating confident but factually incorrect or fabricated information
LLM hallucination occurs when the model produces plausible-sounding but false or invented information with unwarranted confidence.
Question 2: What is a 'model card' in responsible AI?
- A credit card used to purchase cloud GPU time
- A document disclosing a model's intended uses, limitations, performance, and ethical considerations (Correct answer)
- A diagram showing neural network architecture
- An API key for accessing a model
Correct answer: A document disclosing a model's intended uses, limitations, performance, and ethical considerations
Model cards, proposed by Google, are documentation accompanying AI models to communicate their capabilities, limitations, and ethical considerations.
Question 3: What is RLHF (Reinforcement Learning from Human Feedback) used for?
- Training robots to walk using human demonstration
- Aligning LLM outputs with human preferences using reward signals from human raters (Correct answer)
- Evaluating RL agents in human-designed environments
- Fine-tuning vision models on human-labeled data
Correct answer: Aligning LLM outputs with human preferences using reward signals from human raters
RLHF trains a reward model from human preference comparisons, then uses RL to fine-tune the LLM to produce higher-rated outputs.
Question 4: What is 'consent' in AI data collection?
- A model agreeing to a license agreement
- Individuals knowingly and voluntarily agreeing to have their data collected and used (Correct answer)
- The process of labeling data before training
- Approving model outputs before deployment
Correct answer: Individuals knowingly and voluntarily agreeing to have their data collected and used
Consent requires that people are informed about data collection and freely agree to it before their data is used to train AI models.
Question 5: What is 'AI safety' research broadly focused on?
- Improving hardware security for AI chips
- Preventing AI systems from causing unintended or catastrophic harm (Correct answer)
- Encrypting AI model weights
- Ensuring AI works correctly on safe inputs only
Correct answer: Preventing AI systems from causing unintended or catastrophic harm
AI safety research works to ensure that AI systems behave reliably, predictably, and without causing unintended harm as they become more capable.
Question 6: What does the NIST AI Risk Management Framework (AI RMF) provide to US organizations?
- A mandatory AI licensing program
- A voluntary framework for identifying, assessing, and managing risks in AI systems (Correct answer)
- Federal regulations for AI certification
- A standard AI training curriculum for employees
Correct answer: A voluntary framework for identifying, assessing, and managing risks in AI systems
The NIST AI RMF provides a voluntary, flexible framework to help organizations manage AI risks across the AI lifecycle.
Artificial Intelligence AI Ethics and Bias 3
What is 'hallucination' in large language models?