Читать книгу Smarter Data Science - Cole Stryker - Страница 28

The most advanced algorithms cannot overcome a lack of data. Organizations that seek to prosper from AI by acting upon its revelations must have access to sufficient and relevant data. But even if an organization possesses the data it requires, the organization does not automatically become data-driven. A data-driven organization must be able to place trust in the data that goes into an AI model, as well as trust the concluding data from the AI model. The organization then needs to act on that data rather than on intuition, prior experience, or longstanding business policies.

Practitioners often communicate something like the following sentiment:

[O]rganizations don't have the historical data required for the algorithms to extract patterns for robust predictions. For example, they'll bring us in to build a predictive maintenance solution for them, and then we'll find out that there are very few, if any, recorded failures. They expect AI to predict when there will be a failure, even though there are no examples to learn from.

From “Reshaping Business with Artificial Intelligence: Closing the Gap Between Ambition and Action” by Sam Ransbotham, David Kiron, Philipp Gerbert, and Martin Reeves, September 06, 2017 ( sloanreview.mit.edu/projects/reshaping-business-with-artificial-intelligence )

Even if an organization has a defined problem that could be solved by applying machine learning or deep learning algorithms, an absence of data can result in a negative experience if a model cannot be adequately trained. AI works through hidden neural layers without applying deterministic rules. Special attention needs to be paid as to how to trace the decision-making process in order to provide fairness and transparency with organizational and legal policies.

An issue arises as to how to know when it is appropriate to be data-driven. For many organizations, loose terms such as a system of record are qualitative signals that the data should be safe to use. In the absence of being able to apply a singular rule to grade data, other approaches must be considered. The primary interrogatives constitute a reasonable starting point to help gain insight for controlling all risk-based decisions associated with being a data-driven organization.