There’s an increasing need for brands to oversee and understand their artificial intelligence (AI) processes as they’re used more and more across organizations. Ethical AI, or explainable AI, provides the foundation for doing just that. Though AI oversight might immediately make one think of a chief data officer or chief technology officer, in actuality, all members of an organization play a key role in ethical AI. This blog post focuses specifically on the role of the chief marketing officer (CMO) in ethical AI implementation.
For ethical AI thought starters across the rest of the C-suite, check out our Embracing Ethical AI ebook.
There are a few specific reasons that ethical AI is important for CMOs:
For more detail, read our blog post about why ethical AI needs to be on every brand’s roadmap.
The first step to uncovering and addressing bias is discovering where it may exist. Conduct a thorough audit to see how machine learning (ML) and artificial intelligence are being used to make decisions across the marketing organization. Some of the common areas to look out for are audience targeting, segmentation, modeling, loyalty and promotions, content, and creative. At this stage, it’s also important to understand what data is being fed into any ML and AI processes and identify where that data is coming from.
A trickier assessment is the potential for bias in the media buying process. No matter what controls brands put in place, there are potentially biased algorithms that will ultimately place and deliver our ads. Though you may not be able to immediately act on this issue, it’s important to be aware of it and start having conversations with media partners to gameplan how to eliminate potential bias in existing and future marketing efforts.
Once you’ve identified potential sources of bias, it’s time to see whether bias actually exists in those places. One effective way to do this is by comparing your data set to the general US population to see if there are any outliers in terms of distribution across different characteristics.
The goal isn’t to have your audience distribution mirror the US population – in fact, there are very few meaningful audience segments that would align closely with the US population. The goal is to understand where there is deviation so that we have the knowledge and the transparency to see bias in the data and make changes that eliminate it.
For example, zip code targeting is a very common practice that, on its surface, seems harmless and effective. It can help predict income, presence of children in the household, etc. However, zip code can also be a proxy for ethnicity, which is a bias that you’d want to address in your data and algorithms. This evaluation step is critical for unearthing similarly troublesome trends.
Assessments and evaluations arm you with the information you need to start making changes. When you can see where bias exists and what that bias is, you can make decisions to eliminate it.
This means building models differently and feeding new information into your machine learning and AI processes to account for the biases you’ve uncovered. It also means developing an ongoing plan to continually evaluate your data and AI – both new and existing – for bias. Regular checks are critical to ensure that the changes you’ve implemented are having the intended outcomes and not unintentionally creating new biases.
AI is a powerful tool in marketing, but it needs to be understood, transparent, and free of harmful bias. Ethical AI is not just the right thing to do – it also makes business sense to deliver engaging experiences to the right customers and prospects. With the right blend of human intelligence and AI/machine learning, brands can deliver better and more sustainable results than a person or AI could deliver on its own.