NTIS Chief Data Scientist: Public-private partnership authority can help agencies with explainable AI
Written by Dave Nyczepir
Agencies looking to better use data for explainable artificial intelligence should leverage the National Technical Information Service’s “unique” partnering authority, its chief data scientist said.
Speaking at an ATARC webinar on Thursday, Chakib Chraibi said Congress had the foresight to allow the National Technical Information Service (NTIS) to partner with top tech companies, academic institutions and nonprofits outside of federal acquisition regulations to address national data challenges and accelerate AI-based capabilities within all agencies.
NTIS is part of the Department of Commerce and works with agencies to determine how they collect, label and classify, and use data to improve these processes and ensure the reliability of machine learning models.
The agency created a Data Skills Task Force a few years ago and mapped out skills for roles like data engineer, data scientist, and data analyst.
The government wants the United States to lead the world in what Chraibi called the “first technology of the 21st century,” going so far as to issue an AI bill of rights on Tuesday affirming its commitment to democratic values, from development to deployment. But the field is changing rapidly — machine learning models become more explainable in weeks, not years — which means agencies need to improve their access to quality data.
“The problem is that a lot of federal agencies are still struggling with legacy data architectures that serve hundreds of applications in a siloed, vertically-oriented approach,” Chraibi said.
Agencies also need mitigation policies in place starting with prototyping to deal with AI risks such as bias as they arise, Chraibi said.
There are a number of AI frameworks promoting responsible AI that agencies can choose from, but they cannot be implemented without metrics that quantify the explainability of the model.
“At NTIS, we have a very agile framework that we work with,” Chraibi said. “And we work very closely with the agency because they are the experts.”
Another area where agencies need to improve is ensuring that their staff have the required data engineering and architecture skills to modernize their infrastructure. Machine learning skills, while important, come later, Chraibi said.
“We are trying to find out what skills are within the Ministry of Commerce and what [employees] need to become a data analyst to scale from the inside, as well as identify what we need from the outside,” Chraibi said.