NCA-GENL Question 193
Select 3A company is developing a large language model (LLM) to assist with medical diagnosis. The team wants to ensure the system is ethical, energy-efficient, and reliable. Which steps should the team prioritize to meet these goals?
- A
Conduct bias testing on the model using diverse datasets to ensure fair and ethical predictions.
- B
Use energy-efficient hardware such as NVIDIA GPUs optimized for AI workloads.
- C
Focus solely on maximizing the model's accuracy, without considering energy or ethical constraints.
- D
Implement explainability mechanisms to allow users to understand the reasoning behind predictions.
- E
Ignore the environmental impact during the training phase as it is a one-time cost.
Show answer and explanation
Correct answers: A, B, D
Explanation
To create ethical, energy-conscious, and reliable AI systems, it is essential to focus on bias testing, energy-efficient practices, and explainability. These steps ensure that the AI system is fair, sustainable, and transparent, aligning with the principles of responsible AI development. Ignoring these aspects or focusing solely on accuracy can compromise the integrity of the system.
- A. Correct.
Bias testing ensures that the model provides fair and ethical outcomes, especially in sensitive domains like medical diagnosis. It is a critical step for ethical AI.
- B. Correct.
Using energy-efficient hardware, such as NVIDIA GPUs designed for AI workloads, helps reduce the environmental impact during the training and inference phases, aligning with energy-conscious goals.
- C. Incorrect.
While accuracy is important, focusing solely on it without considering ethical and energy constraints can lead to unfair or unsustainable AI systems. This approach contradicts the goals of ethical and energy-efficient AI.
- D. Correct.
Explainability mechanisms allow users to trust and understand the model's decisions, which increases reliability and aligns with ethical AI principles.
- E. Incorrect.
Ignoring the environmental impact, even during training, contradicts the goal of creating energy-conscious AI systems. Training is often resource-intensive and has a significant environmental footprint.