Artificial intelligence (AI) is used to make life-altering decisions every day, from who gets a job and receives access to a loan to early disease detection and diagnosis.

“Technology, and the people who create it and apply it, will play a key role in shaping what's to come,” Simone Ross, technology curator for TED​ said, during her opening remarks of Dell Technologies World TED Salon talk. “So how do we define that future as opposed to letting it define us?"

Never mind the benefits of AI in simplifying tasks and improving efficiencies, when it comes to fair and equitable policy decision-making, Mainak Mazumdar, chief data and research officer at Nielsen, said AI has not lived up to its promise. “AI is becoming a gatekeeper to the economy,” Mainak explained. “It is reinforcing and accelerating our bias at speed and scale with societal implications.”

So it begs the question: Is AI failing us? Are developers designing algorithms to deliver biased and wrong decisions?

According to Mazumdar, it's not the algorithm, but the biased data that is responsible for inequitable policy and decision-making AI – and it demands an “urgent reset.”

“Instead of algorithms, we need to focus on the data,” he said. “We're spending time and money to scale AI at the expense of designing and collecting high quality and contextual data.”

To stop perpetuating the human and societal data bias that we already have, Mazumdar prescribed a three-pillar realignment that focuses on data infrastructure, data quality, and data literacy. 

Algorithms are only as good as their data. For example, if a word has been misspelled in training data, the algorithm will perpetuate the error. That is to say, machine learning (ML) algorithms are driven by the data they are fed, and outcomes rendered are only as unbiased as the data upon which they are based.

And with it being a census year, that type of error is analogous to the possibility of undercounting minority groups. The once-a-decade tradition of counting every person living in the U.S. is the foundation for many social and economic policy decisions. However, with the pandemic, and the politics of the citizenship question, Mazumdar expects “significant under counting of minorities,” which will in turn “introduce bias and erode the quality of our data infrastructure.”

“When minorities are undercounted, AI models supporting public transportation, housing, and health care insurance are likely to overlook the communities that require these services the most,” Mazumdar said.  

Mazumdar went on to explain that most AI systems rely on data that has been previously collected for other purposes because it's "convenient and cheap." However, “data quality is a discipline that requires commitment, real commitment,” he added. “This attention to the definition, data collection, and measurement of the bias is not only under appreciated in the world of speed, skill, and convenience, it is often ignored.” 

If bias in algorithms mirrors the real world, then the first step to improving the results is to make that database – in this case, the census data – representative of age, gender, ethnicity, and race of the population. 

“Investing in this data, quality, and accuracy is essential to making AI possible, not for only few and privileged, but for everyone in the society,” Mazumdar said. “Our once in a lifetime opportunity to reduce human bias in AI starts with the data. Instead of racing to build new algorithms, my mission is to build a better data structure that makes ethical AI possible.”