The intelligence community prepares to develop new artificial intelligence systems that can identify who, or what, authorizes any particular document — and on the other hand, advanced systems that regulate authors’ privacy protections.
“This effort, we think, could turn into a game to track down campaigns to kill disinformation, and things like anti-trafficking and other violent activities going on in online blogs, and elsewhere using text,” Drs. Timothy McKinnon told Nextgov recently released. interview.
McKinnon is the program manager for Intelligence Advanced Research Projects Activity leading the project, which is considered HIATUS — or, a transliteration of human-language text input — a program.
IARPA will likely issue a comprehensive agency announcement next week to request HIATUS research proposals. McKinnon provided an early look at the project to be unveiled, highlighting recent research and the development of IC in human language technology.
Read:New Autonomous Black Hawk Testing
“We expect the program to take 42 months after it is launched,” he assured.
The challenges that IARPA intends to address with HIATUS are incredibly complex.
“In a small context, just imagine if you have 100 different people, and then ask them to explain something simple – like how you open a door – in two sentences or one sentence, you will probably get 100. Different answers, right?” Said McKinnon. “And, you know, each type has its own idiosyncrasies as a writer who may have been used by written systems.”
Lots of multilingual text produced by anonymous writers — both human and machine — on a daily basis. As the program manager noted, such items often contain portions of language that can be used directly to identify who designed the information, or to protect the authors’ identity if the interpretation could endanger certain types.
“Frankly, what we do is reflect the elements of style. Therefore, these are things like wording and syntax that can identify who wrote a particular text. Think of it as similar to your written fingerprints, right? What are the features that make your writing unique? Technology will therefore be able to identify that fingerprint using Intelligence against the chorus of other texts, and then align it from a single author, ”he explained. “In terms of privacy, what technology can do is find ways to change the text so that it does not look like human text.”
Read: How the AI-Enabled Worker Will Deliver Better Government in 2022
The program is structured in a way that puts the two components in the competition to promote development on both sides. With HIATUS, officials actually accept ownership of ownership and privacy as a learning problem for the opposing machine as development and testing involves competition between the two parties.
To date, there are only three phases of coping strategies that the IARPA team aims to address.
With traditional handmade methods, human experts can analyze the text and seek out the details or attributes of a particular author. Another category includes machine learning and algorithmic methods such as retrieval or Bayesian models, but McKinnon said those do not fit well into different text types.
The third set of strategies is “very new,” he notes. It incorporates emotional language models, which are complex systems that represent human language.
“The problem with these models is that even though they are very fast — and very efficient — we do not really understand what is going on between them. They are very complex, ”said McKinnon. “And then HIATUS wants to do, among other things, find some of the reasons behind the behavior of those models, so that you can, in fact, know, when we do an annotation or create a patent, we know. to really understand why the system behaves so well, and to be able to make sure that it does not pick up where it left off and that it does the right thing. ”
At the launch of the BAA, proponents will have the opportunity to highlight their own research and development in the field and suggest how IARPA should move forward to meet its overall goals.
“We are looking to develop dynamic programs in all domains and genres — and, there will be foreign languages involved in the process as it progresses,” said McKinnon.
As a key research and development hub for IC, IARPA conducts experimental businesses and does not play a major role in fully extracting or deploying creative technologies. Once completed, the tools are sent to agencies to work based on their specific needs.
About 70% of its completed research efforts shift to other government partners.
For that reason and others, however, officials do not speculate as much about the use cases that can flourish, in what they have produced.
However, McKinnon noted that this work could have far-reaching implications for combating human trafficking, or for understanding and stopping campaigns of increasingly violent online influence.
Keep visiting us
Visit here for more.
