President-elect Donald Trump's plans for federal cybersecurity aren't clear, but some contractors are gearing up for a spike in business over the next few years.
One company that builds an artificial intelligence-based forecasting product called Eureqa for federal customers, including the Air Force, is segueing into the cybersecurity market. Nextgov spoke with Michael Schmidt, founder of Nutonian, which is currently piloting Eureqa at a few federal agencies, and David Rubal, chief technology officer of analytics at re-seller DLT Solutions, about trends in federal demand for artificial intelligence.
Nextgov: What kinds of use cases are federal customers looking for with artificial intelligence?
David Rubal: Cybersecurity is definitely top of mind. It affects the forensic aspect of data management ... and also making it prosecution quality. There's a lot of other use cases leveraging the internet of things as data generators. Applications like smart cities, being able to offer new citizen services and being able to predict the patterns of people in cities. Everything from trash pickup to road repair.
» Get the best federal technology news and ideas delivered right to your inbox. Sign up here.
The White House's recent report on the state of AI basically set the stage for the relevant use of algorithmic [technology]. From a cyber prevention perspective, ensure that as threats are known and processed. The ability to match behavioral analysis with unstructured data ... with structured data like what systems people access, and what they're doing while they have access to the system. That all feeds into the the very quickly detect the presence of the possibility, or the potential, for cyber risk for the potential for cyber risk.
NG: Trump's cyber plan is relatively vague. Is there any uncertainty about business opportunities in the next few years?
DR: The Cybersecurity Information Act of 2015 ... was the first attempt at getting governance around the fact that a broad cybersecurity plan is needed. Time will tell based on the new administration’s acceptance of that act and the progression, but there’s a couple other factors happening ... IT modernization is an issue that’s affecting agencies of many levels.
Many agencies find themselves in the very basics of IT modernization, looking at infrastructure built 20-30 years ago …[introducing] a couple billion internet of things devices into their environment creates a real stress point. Cybersecurity is wound into that as one of the co-dependencies. It’s not a choice; agencies will continue to spend money to ensure the IT environment will continue to get modernized.
We’re looking at new ways of being able to help the government in areas like data and analytics [and] invest[ing] in strategic technology while they are modernizing, being able to address pressing needs like cybersecurity ... Last year, we built a cybersecurity practice to specifically address that growing need.
NG: What does Nutonian's cybersecurity technology actually do?
Michael Schmidt: We've had tremendous interest for specific types of security projects, things around detecting insider threats ... about finding new forms of attacks. Identifying and helping people pinpoint things they should quarantine and look at, things called "honeypots."
A use case for this might be password-stuffing, where bad actors can buy huge lists of logins and passwords to try to gain entry into someone’s system ... there'll be a large spike [in activity]. [Customers can] put those connections into a sort of quarantine honey pot, where they're tracking every single possible detail. It's actually hard to detect those ...i f you look at all the information at once and try to model it at once. If you're trying to look for anomalies ... Eureqa could build very granular models.
NG: How does this technology qualify as "artificial intelligence"?
MS: [We're] helping businesses predict security threats. [We also] work closely with the U.S. Air Force, primarily around how to develop, design, new advanced materials—how do they get certain performances out of their engines and their parts, and new alloys.
If they want to design a new material, their engine, a turbine blade, there are a thousand things that can go wrong. How do they treat the metal? How quickly do they heat and cool things? It's a lot of sensor data. It's way too much to process manually.
The subset we focus on is anything with forecasting: time series datasets. That's where data science is extremely difficult today. Part of the challenge is the nature of that data: it's very noisy.