As we’ve discussed in previous APEX of Innovation posts, it’s critical that the industry address data ethics challenges if we hope to maximize the technology’s full potential. 

A recent InformationWeek article outlines four principles that should guide the development of any ethical AI strategy, including:

Understand Goals, Objectives, and Risks 

In the early days of AI, when the technology was first included in Gartner’s Hype Cycle, companies raced to demonstrate their AI capabilities. These investments failed to deliver much value. According to InformationWeek, these early AI strategies were essentially an afterthought on top of existing analytics projects because companies lacked a clear understanding of the problem they wanted AI to address. 

The first step in developing an ethical AI strategy is understanding all goals, objectives, and risks and creating a decentralized approach to AI. 

Do No Harm 

It’s important to take preemptive measures to ensure AI solutions do no harm. One strategy for addressing this is to institute a framework that prevents any negative impact on algorithm predictions. The InformationWeek article offers customer surveys as an example. Without suitable modifications to the algorithm to incorporate linguistic nuances, the organization would fail to understand customer sentiment returned via a non-English language—or whichever language the AI algorithm is trained in. 

Develop Underlying Data that is All-Encompassing 

A holistic, transparent, and traceable data set is critical for successfully implementing AI. Companies must develop a framework for the data standards used and ingested by their AI models. For example, slang, abbreviations, and code words can easily confuse a highly technical algorithm. Organizations must incorporate these human nuances and other relevant variables to ensure a holistic data set. 

Avoid a Black-Box Approach to Algorithm Development 

Full transparency is essential for ethical AI. To ensure this transparency, companies must understand how each node or gate in the algorithm makes its conclusions and interprets results. This requires a robust technical framework that can review the underlying code to explain both model and algorithm behaviors. Take a look at the InformationWeek piece for more on how to achieve this. While there are certainly some ethical pitfalls to navigate, AI can deliver numerous advantages, provided the right ethical strategy is in place. You can read more on just a few of these via this previous APEX of Innovation article.