The Reality of Data Bias and How to Make the Most of It
Artificial Intelligence (AI) is chock full of bias. We see it everywhere from bots to recommendation engines. We find bias in website analytics and CRM data. Even AI platforms including Google and Amazon, have been challenged with data bias. It’s practically inescapable.
Marketers see AI as the future of consumer engagement for brands. Yet, it’s easy to understand why a brand would be hesitant to risk building consumer-facing AI when consumers are increasingly cognizant of the dangers of data bias. Given that the data that drives AI can also be large and complex, it’s understandable to have data analysis paralysis.
The reality is – all data is biased. I’m here to tell you that it’s ok. Not all biased data is bad, and it doesn’t need to be a dead end. It’s what marketers and brands do with that data that matters. Here are five ways marketers can work with brands to navigate the choppy waters of data bias (in no particular order).
Compare your data against comparative data sets
Remember, all data is biased. It’s often easy to identify obvious examples of bias in your data or AI algorithms, but unfortunately, most data bias is not so obvious. Ask yourself if the data reflects a diverse consumer base: is there a gender bias? Are the data touch points equally distributed?
With the primary goal of identifying gaps and anomalies, data comparison is necessary to understand the level of bias in your data. The phrase, “bad data in, bad data out,” applies.
The gaps created by bias can also serve as opportunities to gather additional data that will enrich the user experience. At the agency where I work, we have a Business Intelligence team dedicated to separating “gaps” from “whitespace” and identifying where future opportunities exist from data.
If your data isn’t diverse, your brand strategy team better be
Data rarely comes from a diverse origin. Analytics are historically biased by bots. Survey responses contain their own bias and are more frequently answered by those who have a negative opinion of the brand.
If AI is powered by data and that data is powered by humans, it doesn’t take a genius to recognize that the humans powering that data, or at least filtering and augmenting it, needs to be diverse. This is where agencies are uniquely positioned at the intersection of diversity and culture. Think of it like a cultural think tank where data is run through the gauntlet of diverse perspectives.
Sometimes you have to throw away your data and start over
If your initial data set isn’t satisfactory enough to train your AI platform, you may have to completely start over because either the bias is insurmountable or it includes partiality towards segments that aren’t aligned with the brand’s KPI’s.
Throwing away data is easier said than done. Big data needs to be robust as well as strategic. Yet, many brands find themselves feeling precious about their data and want to force a square peg in a round hole. But this is the wrong approach and it’s the data scientist’s job to determine the viability of data. Data scientists need to work side-by-side with brand strategists to understand the data patterns where data bias exists. It’s this collaboration that is critical to identifying bias early in the AI development process.
Don’t try to be all things for all people
AI is capable of doing some amazing things. Unfortunately, it can’t do everything all the time. At least not yet, anyway. AI platforms succeed when they are used to address a specific set of needs. Sephora’s Color IQ and Equinox’s Digital Coach focus on a short list of consumer needs in order to leverage AI that is focused and reliable.
Sticking to a short list of specific features reduces bias both in the data and the AI algorithms. The reason for this is two-fold. First, the scope of the dataset is limited to specific subjects. The data can be more easily evaluated during initial design to ensure bias is reduced to a necessary minimum. Second, AI algorithms can be kept small and precise. Reducing the number of, “if this, then that,” scenarios reduces the introduction of programmatic bias.
More importantly, a focused feature-set generates a feedback loop where the data tracked through user interaction will result in a more reliable consumer experience while keeping user-generated bias at bay.
AI needs a personality that can push against bias
Very few of us find data engaging. No matter how robust and diverse your data is, AI can’t create relationships with users without a personality and the best AI platforms have one that is relatable and entertaining. A personality that seems like it has a story to tell if you ask it the right question.
That personality can also be a powerful tool for fighting bias. It requires the alchemy of creative storytelling and technology to turn biased data into inclusivity. It can respond to questionable consumer inputs in a way that is provocative and entertaining.
AI will continue to become an essential part of the brand experience. No amount of AI training will take away bias. AI can’t, on its own, stay relevant to the culture. It needs a cultural ambassador and mentor. As marketers, this is our duty to play that role for brands.