Exam AI-900 topic 1 question 161 discussion

B. Take a training sample that is representative of the population in the United Kingdom. To minimize bias in the system, it's important that your training data is representative of the population you're modeling. This helps ensure that the model's predictions are valid for the full range of drivers in the United Kingdom. While option A (Remove information about protected characteristics from the data before sampling) could help in some cases to reduce direct discrimination, it might not be sufficient to minimize all types of biases, as some of these characteristics might be indirectly encoded in the remaining features. Option C (Create a training dataset that uses data from global insurers) may introduce more bias since driving conditions, laws, and demographics vary greatly by country. Option D (Take a completely random training sample) could still introduce bias if the original data pool is not representative of the population you're interested in.

I think correct answer is B. To minimize bias in the system, it is important to ensure that the training sample used is representative of the population in the United Kingdom. By including a diverse range of data that accurately reflects the population, you can reduce bias that may arise from underrepresentation or overrepresentation of certain groups. This approach helps to ensure fairness and avoids discriminatory practices in the prediction of insurance prices. Removing information about protected characteristics (option A) may help in some cases, but it is not sufficient on its own to address bias. Options C and D do not specifically address the need for representative sampling. (Chat GPT)

A. Remove information about protected characteristics from the data before sampling. To minimize bias in a system for predicting insurance prices, it is important to remove information about protected characteristics (e.g., race, gender, ethnicity) from the data before sampling. This helps prevent the model from learning and reinforcing biases based on these characteristics, which is essential for fairness in pricing and avoiding discrimination. Options B, C, and D do not directly address the issue of mitigating bias in the data and model.

By bing bot - This is a multiple-choice question. The correct answer is B. Take a training sample that is representative of the population in the United Kingdom. This option ensures that the system learns from a diverse and balanced data set that reflects the target population. Removing information about protected characteristics (A) may not eliminate bias, as there may be other variables that are correlated with them. Using data from global insurers © may introduce irrelevant or misleading factors that are not applicable to the UK market. Taking a completely random training sample (D) may result in an unrepresentative or skewed data set that does not capture the variation in the population.

its exactly like the comment from redsoxfred

B. Take a training sample that is representative of the population in the United Kingdom.

maybe the uns is C because it says to 'reduce' bias?

I vote for B. As suggested by other person, countries have different issues/behaviours etc

I think is B. Global insurers can introduce different prices based on other countries issues/behaviors/security/environment, even with values normalized.

I think its B also

i think its B (training data is representative of the population of UK)

should be B

do we know what the correct answer is? I am inclined to think B would be the best choice

Hmmm, I would have thought B, maybe D