Monday, December 19, 2005

Further thoughts on TIA 2: FISA Boogaloo

The President "was so desperate to kill The New York Times’ eavesdropping story, he summoned the paper’s editor and publisher to the Oval Office." Other people are dealing with the legal aspects of this case and they are, like, real lawyers 'n' stuff. So I'll stick to technical ideas.

TIA is an interesting program from a technological perspective. I've only offered a possible explanation on how a system could safely operate outside the confines of the FISA statute. Stalky was mentioning earlier that analyzing only the patterns of communications wouldn't be nearly as useful as analyzing the content too. And he's right.

In previous examples I used telephone calls as the example of what this theoretical TIA system could do. But in the Information Age, the NSA would not limit itself to telephone calls. A more total TIA system would collect and analyze not only traffic patterns of emails, chat rooms, web traffic, and phone calls, ideally it would also parse the content.

For telephone conversations, off the shelf voice recognition software is already very good. I'm sure what the NSA uses is much, much more advanced. So let's turn back to the bullet list of TIA goodies and look at a few:

Collaboration and sharing over TCP/IP networks across agency boundaries

This one is not really TIA specific, but does address a big problem with government systems. Integration. I would imagine the "former" IAO was tasked with integrating inter and intra-agency communications in order to share intelligence more effectively. That's a pretty big task on its own.

Large, distributed repositories with dynamic schemas that can be changed interactively by users

I'm guessing this refers to creating a semantic web of information where analysts can define the taxonomy to suit a particular situation. Pure speculation though.

Foreign language machine translation and speech recognition

The NSA probably has the recognition part pretty well covered, but machine translation is still pretty tricky. Keyword recognition is probably more important than contextual analysis.

Biometric signatures of humans

Facial recognition? Voice recognition? :shrug:

Real time learning, pattern matching and anomalous pattern detection

Machine learning. Check the link. Wikipedia can explain it better than any short description I can offer.

Entity extraction from natural language text

This also refers to the semantic web and more importantly, contextual analysis. If I'm not mistaken, this is pretty cutting edge stuff, e.g. cool. Who knew government work could be fun?

Human network analysis and behavior model building engines

This was touched on previously.

Event prediction and capability development model building engines

There was a spectacular PR failure by the administration regarding a real world implementation of this idea. I'll search for the link later. It was amusing.

Structured argumentation and evidential reasoning

More AI goodies.

Story telling, change detection, and truth maintenance

Sounds AI-ish, maybe 1984ish. This one will require some research.

Business rules sub-systems for access control and process management

That sounds more like government work...

Biologically inspired algorithms for agent control

Genetic algorithms for AIs -- data mining stuff.

Other aids for human cognition and human reasoning

Sounds like more AI training/research.

If my predictions prove correct, I'll probably cover these topics one by one, in the future. Perhaps Stalky will join me, as my knowledge is shallow in some of these areas, and he'll be able to fill in alot of the gaps.


Post a Comment

<< Home