US to develop AI to identify authors of anonymous letters

The U.S. intelligence community has launched the HIATUS program to develop artificial intelligence for identifying the authors of anonymous letters. The Register reports.

The Office of the Director of National Intelligence (ODNI) said the program aims to create software that can perform “linguistic fingerprinting”.

“People and machines generate a huge amount of messages on the internet. The text contains linguistic features that can reveal the author’s identity,” said the Intelligence Advanced Research Projects Activity (IARPA).

The agency believes that a well-designed model could help them determine the writer’s stylistic consistency across different samples. Such an algorithm could also alter language templates to render a letter anonymous, they added.

The HIATUS AI is expected to support multiple languages.

“We have every chance to achieve the goals, provide the intelligence community with the much-needed capabilities […], using the latest advances in computational linguistics and deep learning,” said Timothy McKinnon, head of the HIATUS program.

HIATUS developers plan to approach the aims from an adversarial AI perspective: authorship attribution and text anonymization. Within the project, researchers will create two experimental groups that will pit against each other.

“Attribution systems are evaluated by their ability to match elements of the same author across large corpora. At the same time, privacy systems are evaluated by their ability to prevent attribution systems,” says an IARPA statement.

The agency said it also plans to develop explainability standards for HIATUS.

According to an ODNI spokesperson, if successful the program will help counter foreign influence, identify counterintelligence risks and protect authors.

The experiment is planned to be completed within 42 months.

As reported in June 2022, Pentagon AI chief Craig Martell announced intentions to speed up the department’s technological modernization.

In April, the U.S. military requested $29.8 million to develop AI infrastructure.

In August 2021, the Pentagon tested a decision-making system, capable of “seeing days ahead”.

Subscribe to ForkLog news on Telegram: ForkLog AI — all the news from the world of AI!