Responsible Artificial Intelligence Is Still Out Of Reach For Many Organizations

There’s strong support for analytics and data science and the capabilities it offers organizations. However, the people charged with developing analytics and artificial intelligence feel resistance from business executives in getting fully on board with data-driven practices. In addition, efforts to ensure fairness in AI are lagging.

That’s the word from a recent study of 277 data managers and scientists out of SAS, which finds that overall, more than two-thirds were satisfied with the outcomes from their analytical projects. At the same time, 42% say data science results are not used by business decision makers, making it one of the main barriers faced.

A lack of support from above is cited as the leading challenge to getting data analytics initiatives off the ground, the survey shows. Data quality issues ranked second, followed by lack of adoption of the results by decision-makers. It’s interesting that “explaining data science to others” is also seen as a challenge, suggesting that a big part of managers’ jobs needs to be evangelizing and educating their business counterparts on the benefits data analytics can deliver to their organizations, and how to do it right.

Here are the leading challenges to becoming more data-driven in today’s environments:

• Company politics/lack of management or financial support 46%
• Dirty data 43%
• Results not used by business decision makers 42%
• Explaining data science to others 35%
• Lack of data science talent in organization 34%

“Managers are generally more satisfied with the company’s use of analytics compared to individuals; however, individuals seem more satisfied with the outcome of analytics projects,” the study’s authors state. “That difference echoes the possible difference between satisfaction with their own projects’ outcomes versus how they’re deployed, and that data science as a whole is more than siloed, individual efforts.

The study also tackled the question of delivery of ethical and unbiased AI. A substantial segment of companies in the survey, 43% do not conduct specific reviews of their analytical processes with respect to bias and discrimination. And only 26% of respondents indicated that unfair bias is used as a measure of model success in their organization. The top two roadblocks were a lack of communication between those who collect the data and those who analyze it, and difficulty in collecting data about groups that may be unfairly targeted.

The study’s authors recommend working “to make data science a team sport” in organizations, and to make an effort to provide data managers and scientists a more active role in the organization. They also urge managers and executives to get more proactive about responsible AI. “Get started with your own project and find ways to add in the means to detect and measure bias,” they advise. “Document your work and present it to management. Sometimes a working example of success is what’s needed to get started – there’s no reason it can’t be your work that sparks the beginning.”