Free AAIR Practice Quiz

What is the MOST critical rationale for integrating value creation into AI design and oversight?

1. Deploying solutions quickly to enable innovation

   Innovation is a key aspect of the AI revolution; however, innovation and quick deployment without alignment to the business does not create value.

2. Ensuring AI solutions address relevant business problems

   Embedding value creation ensures AI outcomes add business value to the enterprise in alignment with enterprise objectives. One way to demonstrate this alignment is to ensure AI provides practical solutions to relevant problems, which can help users, customers, and leadership see the value in employing the technology.

3. Reducing development and operational expenditure

   Cost reduction is a typical objective of technology programs, but it does not necessarily demonstrate alignment with organizational objectives.

4. Improving stakeholder satisfaction and return on investment (ROI)

   AI solutions can deliver on improved satisfaction and return on investment (ROI), but these may also be achieved without the use of AI. It is important to determine if AI solutions are the most appropriate means of solving the business problem through alignment with business goals and objectives.

Which of the following would BEST address ethical considerations when evaluating the use of a high-risk AI solution?

1. Data privacy impact assessment (DPIA)

   A data privacy impact assessment (DPIA) looks at how data privacy is ensured and/or affected by its use throughout an enterprise. While data privacy could have an impact on the ethical use of AI, it is not the best method to assess the ethical impact of AI use.

2. Fundamental rights impact assessment (FRIA)

   A fundamental rights impact assessment (FRIA) is mandated by many jurisdictions to identify and mitigate risk to human rights posed by high-risk AI systems. This is the best option to evaluate ethical compliance and minimize potential harm to humans.

3. Business impact analysis (BIA)

   A business impact analysis (BIA) evaluates the impact that the loss of a system has on overall business operations and resiliency and does not necessarily look at ethical implications.

4. AI risk assessment

   An AI risk assessment should consider ethical and human rights risk; however, a FRIA would be the best tool to assess these areas regarding the use of AI.

Which of the following would BEST address discrepancies in formatting when reviewing data collected from various sources to be used in AI training?

1. Data augmentation

   Increasing or augmenting the dataset alone does not resolve inherent data quality problems, such as inconsistent formatting, and may worsen model bias or errors.

2. Data cleansing

   Data cleansing is the process of cleaning, correcting, and standardizing data to ensure it is consistent across data dimensions. In this instance, the issue is with the format of the data, so data normalization would be more appropriate.

3. Data normalization

   Data normalization applies standardized formatting to data elements that will be used in AI solutions (e.g., ensuring all dates are entered as DD/MM/YYYY) and often involves organizing the data in a database. This minimizes noises and enables better data integrity.

4. Data minimization

   Data minimization ensures only necessary data is collected and used for a specific purpose. It does not address issues of formatting.

Which of the following is the MOST significant advantage of applying network segmentation and encryption within AI infrastructure?

1. Enhances the speed of AI algorithm computations

   Network segmentation does not affect the effectiveness of the AI’s processing power.

2. Decreases the volume of storage needed for large AI data volumes

   Decreased storage requirements are not an outcome of network segmentation and encryption; they are designed to enhance data security rather than optimize storage capacity.

3. Facilitates the distribution of AI models across multiple cloud environments

   Network segmentation does not enable better distribution of AI models in cloud environments.

4. Protects data while it is processed within the AI solution

   Network segmentation helps to protect AI data during transmission, minimizing the risk of unauthorized access and preserving data confidentiality and integrity.

Which of the following represents the MOST significant risk related to a prompt injection attack on a generative AI (GenAI) model?

1. Service disruption

   Although service disruption is a valid concern, it is less directly linked to prompt injection attacks.

2. Biased decision making

   Prompt injection attacks enable adversaries to alter generative AI (GenAI) outputs, which could lead to biased decisions or other behaviors, such as revealing proprietary or sensitive information.

3. Model drift

   Model drift is the degradation of model performance over time. While a prompt injection can affect model outcomes, they are not the primary cause of model drift.

4. Training data leakage

   Training data leakage would result from a lack of security surrounding the storage location of AI datasets. Prompt injections are more closely related to the AI model itself, rather than the training data.