Aegis-GPUTrustworthy AI
5% of the NCA-GENM exam. Read each objective's key points, open “What NVIDIA says” to see the source, then practise.
Experimentation · Core Machine Learning and AI Knowledge · Multimodal Data · Software Development · Data Analysis and Visualization · Performance Optimization · Trustworthy AI
7.1 Ethical principles of trustworthy AI
NVIDIA's definition and guiding principles.
Key points
Trustworthy AI is a way of building AI, not a single feature. It puts the safety of people and openness about the model first.
What NVIDIA says (1)
“Trustworthy AI is an approach to AI development that prioritizes safety and transparency for the people who interact with it.”
Ethical principles are the values a team commits to when building AI. NVIDIA lists privacy, transparency and nondiscrimination among them.
What NVIDIA says (1)
“We work to maintain our guiding principles of privacy, transparency, nondiscrimin”
Key terms: Trustworthy AI
7.2 Privacy and consent
Consent for personal data and federated learning.
Key points
Consent means a person agreed to a specific use of their data. Privacy is about protecting personal data even when it is used.
What NVIDIA says (2)
“Developers of AI models that rely on data such as a person’s image, voice, artistic work or health records”
“should evaluate whether individuals have provided appropriate consent for their personal information to be used in this way.”
Federated learning trains a shared model by sending model updates, not raw data. It lets diverse data help the model while keeping privacy.
What NVIDIA says (1)
“Federated learning is a way to develop and validate AI models from diverse data sources while mitigating the risk of compromising data security or privacy, as the data never leaves individual sites.”
Key terms: Personal identifiable information Consent Federated learning
7.3 NVIDIA technology for trust
NeMo Guardrails and confidential computing.
Key points
Guardrails are programmable rules around an AI app. Topical guardrails keep it on subject. Safety guardrails limit language and sources. DALI means the NVIDIA Data Loading Library.
What NVIDIA says (2)
“Topical guardrails ensure that chatbots stick to specific subjects.”
“is an open-source Python package for adding programmable guardrails to LLM-based applications.”
Data in use is data being processed in memory. Confidential computing protects it with hardware-based trusted execution environments (TEEs).
What NVIDIA says (1)
“Confidential computing secures AI workloads by encrypting application code, models, and data in use—not just at rest or in transit.”
Key terms: NeMo Guardrails Confidential computing
7.4 Minimizing bias
Where bias comes from and how to reduce it.
Key points
Bias is a systematic unfairness in model outputs. Narrow data teaches a narrow view of the world.
What NVIDIA says (1)
“AI models are trained by humans, often using data that is limited by size, scope and diversity.”
Synthetic data is generated data that can balance groups that are under-represented. Checking outputs for discriminatory patterns finds bias early.
What NVIDIA says (2)
“trustworthy AI developers mitigate potential unwanted bias by looking for clues and patterns that suggest an algorithm is discriminatory”
“Synthetic datasets offer one solution to reduce unwanted bias in training data”
Key terms: Unwanted bias Synthetic data