AI for You by Kapoor Shalini;Mehta Sameep;

Author:Kapoor, Shalini;Mehta, Sameep;
Language: eng
Format: epub, pdf
Publisher: Bloomsbury Publishing India Pvt. Ltd.

One important aspect to note is there is no ‘correct’ way to clean data for AI. For example, what one application considers a data bias may be a prerequisite for another application. This is a departure from traditional data quality where missing data is considered unwanted property, irrespective of the upstream Business Intelligence task. Therefore, in some cases, we should look at quality assessment for AI as the generation of more metadata. If bias is detected in the data, then, instead of auto-cleaning, we can add the details of a data bias as metadata and different users can decide to clean the data based on their requirements.

4. Policy: Data provides a great opportunity and leverage for enterprises. However, with this opportunity comes the big responsibility of using the data in a compliant, safe, and governed fashion. In fact, it might be better to not use the data at all than to use it in a non-compliant fashion. Therefore, it is non-negotiable that organizations have strong policy enforcement. Let us look at some of the common types of policies.

Access Control: This is largely self-explanatory. Only authorized users should be able to use the data. Apart from access, the policy should also note the kind of access—read, write, append, etc. This is simple to understand but tough to operationalize. Everyone in an organization would request access to all the data. Therefore, a proper process needs to be there to vet requests and only clear valid ones. Please note that the organization should not put an artificial upper limit on the number of requests; instead, all valid requests should be approved. Moreover, a complete audit trail of the access requests should be managed.

Data Use Policy: Simply controlling access to data may not be enough because the same data could be used for a variety of applications and some of those applications may be in a grey area and not strictly compliant. Therefore, the data access request should be accompanied by the intended use of the data. The approver should carefully look at the intended use before clearing the request. Let’s say that the use of face recognition technology to mark employee attendance is a valid use of data. However, using the same technology to monitor if employees are attentive or not during office hours will be a clear breach of privacy. Please remember that the data is only as good as the application it is used for. The same data which opens multiple revenue streams for a company can also turn into a PR nightmare.

Fine-grained Policy: So far, we have looked at the policies which are enforced at a data-set level. Now, consider a data set which has one sensitive column, such as date of birth. How should we define the policies for such a data set? It should ideally be restrictive because it has one sensitive column. To handle this situation, we have fine-grained policies which are enforced on a subset of columns. One should be able to define a policy to say

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.