US Intelligence Community Launches Research Into AI That Can Conceal – Or Expose – Authorship
The research and development arm of the Office of the Director of National Intelligence (ODNI) has launched a program seeking to develop new artificial intelligence technologies capable of attributing authorship of a text, or concealing it to protect the author’s privacy.
The Human Interpretable Attribution of Text Using Underlying Structure (HIATUS) program is the Intelligence Community’s latest research effort to advance human language technology, with the program led by the Intelligence Advanced Research Projects Activity (IARPA), the ODNI’s research and development arm. The ODNI hopes that HIATUS will be able to develop technologies capable of countering foreign malign influence activities, identify counterintelligence risks, and assist in safeguarding authors who could be endangered should their writings be connected to them.
The HIATUS program aims to create technologies capable of three goals through identifying an author’s “linguistic fingerprint”:
- Perform multilingual authorship attribution by identifying stylistic features — such as word choice, sentence phrasing, organization of information — that help determine who authored a given text.
- Protect the author’s privacy by modifying linguistic patterns that indicate the author’s identity.
- Implement explainable AI techniques that provide novice users an understanding, trust, and verification as to why a particular text is attributable to a specific author or why a particular revision will preserve an author’s privacy.
According to the ODNI, over 20 academic institutions, nonprofits and businesses are already participating in HIATUS, with research contracts awarded to lead organizations including Charles River Analytics, Leidos, Raytheon BBN, SRI International, the University of Pennsylvania, and the University of Southern California through a competitive Broad Agency Announcement.
“Each of the selected performers brings a unique, novel, and compelling approach to the HIATUS challenge,” said program manager Dr. Tim McKinnon. “We have a strong chance of meeting our goals, delivering much-needed capabilities to the Intelligence Community, and substantially expanding our understanding of variation in human language using the latest advances in computational linguistics and deep learning.”
Lawrence Livermore National Labs, Pacific Northwest National Labs and the University of Maryland Applied Research Laboratory for Intelligence and Security will serve as the test and evaluation team for HIATUS technologies.
Should HIATUS result in workable technologies in the future, the members of the intelligence community will decide on if, or how, they will be implemented into systems ready for operational use.