AI to Monitor Foreign Press for Threats

jefu writes "According to the New York Times, the US Department of Homeland Security is funding AI tools to monitor the foreign press in order to detect threats to the United States. While the article says there are restrictions on doing this kind of monitoring within the US, there are no restrictions on media outside the US. (No hint is given as to how this would apply to syndicated articles written in the US and published abroad.) This is as yet experimental."

Comments

Re:Open Source Intelligence

[jefu]

I'm the original poster and I'm not sure I see anything wrong with doing this since the information is public. I found the article interesting (if a bit skimpy). A number of questions came to mind, but I only included the one on syndication since the article specifically said that the US press would not be monitored and that foreign press would. If things are syndicated, how do you tell the difference?

I didn't want to load the post with questions but I wondered in particular what natural language understanding techniques could provide that Google News (and other news aggregators) does not already provide - certainly the trending information available seems particularly interesting. I also wondered if the language understanding is really all that strong for other languages (including specifically Arabic since the lack of Arabic speakers has come up as newsworth a number of times) - especially in US universities. A deeper problem might be to distinguish between anti-american attitudes and anti-(american-president-and-administration)-attit udes. Finally, I found the mention that specific journalists attitudes might be tracked a bit troublesome - would this be used then to deny visas to journalists who consistently manifest anti-american(-president-and-administration)-attit udes.

The holy grail of OSINT

[Kadin2048]

I think it's a little more subtle than that.

There is a theory, which I have heard articulated from time to time (although I don't know if there's a name for it or not) which says that right before a major event there is a lot of "chatter"; subtle yet distinct signs that something is about to happen, but which are too minor on their own to generate any attention. Once the event happens, in hindsight you can look back and recognize them. It's sort of a reverse butterfly-effect; the assumption is that no matter how good at being secretive you are, you will make some signs in the course of executing your plot, and some of those signs will percolate and be reported in papers somewhere. So you just need to know what to look for.

So basically what you might do, is take a big pattern-matching AI system, and "teach" it using the media records preceding other big terrorist events. 9/11, London, Madrid, etc. You have it comb through all the world media before those events, and see if you can find patterns, the little things that in retrospect might have alerted you that something was up. Then, thus primed with information and hopefully some patterns, you set it loose on the real-time news feeds.

In theory -- if the theory holds water, anyway -- the system might then be capable of giving you a warning of something big heading your way, picking up on stuff that a person might not recognize.

Anyway, I'm not sure if that's the theory that this particular system is going to try and use, but it's one idea that I've heard described; sort of as the 'holy grail' of machine-derived OSINT. More likely, you'd end up with a system that just gives you statistical summaries of the number of anti-US editorials in various countries or something. Useful for the State Department perhaps, but I'm not sure for preventing the next 9/11.