AI-900 Question 53
Select 3A company is deploying an AI-powered chatbot to handle customer inquiries. During testing, the chatbot occasionally provides incorrect information that could mislead customers. Which actions should the company take to ensure the chatbot is reliable and safe to use?
- A
Implement a mechanism to monitor and log the chatbot's responses for review.
- B
Restrict the chatbot to predefined responses and avoid using machine learning models.
- C
Test the chatbot extensively with diverse datasets to identify and address potential errors.
- D
Allow the chatbot to operate independently without human oversight to improve efficiency.
- E
Enable an escalation feature for users to contact a human agent if needed.
Show answer and explanation
Correct answers: A, C, E
Explanation
To ensure reliability and safety in an AI solution, it is critical to implement monitoring mechanisms, conduct rigorous testing with diverse data, and provide fallback options such as human intervention. These measures help mitigate risks, improve accuracy, and maintain user trust, which are essential considerations for deploying AI solutions responsibly.
- A. Correct.
Monitoring and logging responses help identify incorrect or harmful outputs and provide data for improving the model, which is a key part of ensuring reliability and safety.
- B. Incorrect.
Restricting the chatbot to predefined responses limits its functionality and flexibility, which is not always practical in complex scenarios. This option does not address the issue of ensuring reliability and safety effectively.
- C. Correct.
Extensive testing with diverse datasets helps uncover biases, errors, and edge cases, ensuring the chatbot performs reliably across different scenarios.
- D. Incorrect.
Allowing the chatbot to operate without human oversight could lead to unchecked errors and unreliable behavior, which compromises safety and reliability.
- E. Correct.
Enabling an escalation feature ensures users can access human intervention when the chatbot fails or provides incorrect information, enhancing reliability and user trust.