DHS maps its goals and next actions for safe AI

The U.S. government has posted an AI roadmap for the Department of Homeland Security, outlining algorithm initiatives and how they will protect Americans.

DHS officials say theirs is the most detailed such document of any federal agency. That may be, but what is certain is that the document hits most of the must-dos for any organization hoping to create ongoing innovation and responsible use.

There’s even a couple surprises in the 24-page document. It bears reading because Homeland Security will be a heavy buyer and builder of AI, including biometric recognition. One of the examples provided of how the Department currently uses AI is the use of biometrics for touchless air passenger processing by the Transportation Security Administration, as with TSA’s CAT2 scanners.

The trick will be getting a sustained majority of Congress willing to fund and prod DHS leaders.

DHS’ roadmap closely follows the White House’s executive order 14110, directing the bureaucracy to develop trustworthy, safe and secure AI. The order also demands that administrators use it responsibly for the benefit of the nation.

The roadmap is broken into many nested goals and intentions, starting with so-called lines of effort: leverage AI responsibly within DHS, promote AI safety and security nationwide and lead in the AI field through cohesive partnerships.

Many and maybe the most points made under those headings are just commonsense – and that’s not a criticism. Steps needed to create reliable and trusted AI are still not universally known much less implemented in any level of government or among private enterprises and researchers.

The not-exactly-simple messages can’t be over-communicated.

Officials have pledged, for instance, to create a DHS AI sandbox and to do it with the government’s privacy, civil rights and civil liberties offices, and the White House’s general counsel. They want it completed this year.

Along the same lines, DHS leaders want their Science & Technology Directorate to create a federated AI testbed that will give the government independent assessments of systems. This one has a five-year deadline.

DHS also is calling for safe, secure and trusted AI use through “robust governance” and oversight. If the goal is to avoid people feeling subjected to AI, transparency and accountability are needed. But as with all organizations involved in AI have signed on to this idea, they have to go beyond good intentions.

A 2024 goal to improve governance and oversight is forthcoming guidance on AI security from the Cybersecurity and Infrastructure Security Agency. The guidance will be composed with help from NIST, along with best practices and guidance on red-team cybersecurity tests for AI systems.

One of the surprises in the document is the goal to eradicate child sexual abuse material. That’s not a core mission for DHS to most people until they consider the society-wide, daily economic costs of abuse that continues for each victim’s complete life.

The roadmap doesn’t state that AI will be used, but that’s an eventuality. There’s no realistic way to address this situation without it.

Article Topics

AI  |  biometrics  |  DHS  |  facial recognition  |  TSA  |  U.S. Government