AI became the unofficial word of 2023 and the craze is likely to continue into 2024 as new creative applications and uses of AI emerge across industries and sectors.
But before organizations invest too many resources into foundational AI models, leadership should ensure that the organization has a firm grasp of foundations of AI – data.
The effectiveness of AI models hinges on the meticulous management of the data used in their development, testing, and deployment. Recognizing this, I propose that, in 2024, organizations adopt the time-tested CRUD (create, read, update, delete) principles, not only for individual data but also for the AI systems themselves.
1. Create for what purpose?
Organizations must thoughtfully select their data sets and clearly define their AI models. The purpose of model and the relevancy of the data are essential to consider at the "create" stage. Select the data set too quickly, and your organization may inadvertently be making decisions based on data that’s most available, not necessarily most relevant. These early steps in designing the model can help the organization identify availability bias earlier in the analytics process. Additionally, intentionality in the create stage allow organizations opportunities to consider the data sources, data lineage, and permissible uses for the data sets.
2. Read by who and when?
In any year, organizations should consider who should have access read the data that informs the modes and who should be involved in interpreting the results and findings. Establishing a firm "read" foundation helps an organization respect data privacy and security as well as establish human-in-the-loop guardrails that can prevent automation bias.
3. When should we update?
The “update” principle is not only about updating records but about organizations continuously adapting their AI models. As with any system, AI models will drift over time and require re-training and updating to prevent skewed or outdated results. Organizations should have a strong foundational understanding of data drift and best practices to continuously monitor and improve the accuracy of AI models.
4. When is it time to delete?
As AI matures in 2024, it is important to remember that models have a shelf life. "Delete" comes into play when the question asked is no longer relevant. Successful organizations will embrace purposeful data deletion. This involves regularly reassessing stored data, eliminating redundant data and respecting user preferences for data removal. This will contribute to a leaner, more secure data ecosystem.
In this era of heightened data consciousness, organizations can weave CRUD principles into their AI and data management strategies are better poised for success. The familiar acronym can be used as a visionary tool for organizations to make sure their strategies align with ethical data practices and responsible innovation.
