AI-900 Question 52
Select 3A company is deploying an AI-powered chatbot to handle customer inquiries. During testing, it is observed that the chatbot occasionally provides incorrect responses or makes biased recommendations. To ensure reliability and safety in the AI solution, which actions should the company take?
- A
Implement continuous monitoring to detect and address incorrect or biased behaviors.
- B
Ensure the dataset used for training is diverse and representative of the target audience.
- C
Disable logging and auditing to enhance system performance.
- D
Establish a mechanism for human oversight to intervene when the chatbot behaves unexpectedly.
- E
Prioritize the chatbot's performance speed over accuracy to improve customer satisfaction.
Show answer and explanation
Correct answers: A, B, D
Explanation
To ensure reliability and safety in an AI solution, actions like continuous monitoring, using diverse training data, and establishing human oversight mechanisms are essential. These steps help mitigate bias, improve accuracy, and allow real-time interventions when the AI behaves unexpectedly. Disabling logging and prioritizing speed over accuracy can compromise both reliability and safety.
- A. Correct.
Continuous monitoring is a critical step in identifying and addressing issues such as bias or incorrect responses in real-time, ensuring the chatbot remains reliable and safe.
- B. Correct.
Using a diverse and representative dataset reduces the risk of bias in the AI model and increases reliability by ensuring fairness in its predictions and responses.
- C. Incorrect.
Disabling logging and auditing compromises the ability to track and troubleshoot errors, making the solution less reliable and unsafe.
- D. Correct.
Human oversight allows intervention during unexpected behavior, ensuring safety and reliability in scenarios where the AI might fail.
- E. Incorrect.
While performance speed is important, prioritizing it over accuracy can lead to unreliable and unsafe behavior, especially in a chatbot providing incorrect or biased responses.