New Survey of Data Science Pros Finds that AI Explainability is their Top Concern
In late October 2020, venture capital firm Wing conducted a survey, “Chief Data Scientist Survey,” of 320 of the senior-most data scientists at both global corporations and venture-backed startups, in advance of its annual Wing Data Science Summit.
Among the survey’s findings:
- Explainability is data scientists’ #1 challenge with respect to models currently, at 46 percent.
- Data labeling (29 percent), model deployment (28 percent) and data quality checks (25 percent) round out the top 4.
- A Fortune 10 data science leader said that business users tell his team, “You’re doing this magic. I don’t trust it yet. That’s why I like to use the rules we have used forever, even if your models show they are performing at only 30-40 percent.”
- When asked what technology developments will be most important for data science and machine learning teams over the next 12 months, again Explainability was the #1 choice.
According to the report, “Model explainability was highlighted by participants as the biggest challenge with models by a wide margin. Among the issues raised was trust across many parties—stakeholders in the organization, regulators, and end users. Participants also raised the issue of data bias. While there is a strong desire to take action, there are concerns as to whether current strategies to deal with data bias are effective and fair. Explainability is an area in need of not only continued scientific research and socialization of best practices but also commercial solutions to help companies implement the state of the art.”
“Lack of explainability leads to so many problems in data science – from a lack of trust both internally and externally to compliance issues and even unfair bias in models,” said Will Uppington, co-founder and CEO of Truera, a Wing portfolio company that provides the industry’s first Model Intelligence platform. “This new data shows that explainability is a key challenge that data scientists are seeking a solution for – and it’s Truera’s entire focus.”
Truera’s technology builds on six years of AI Explainability research performed at Carnegie Mellon University (CMU). This work was led by Co-founders Anupam Datta, President & Chief Scientist, Truera and Professor, CMU; and Shayak Sen, CTO, Truera.
Sign up for the free insideBIGDATA newsletter.
Read More …