[FoRK] How to spot a spook; also, minor quibble for JAR

J. Andrew Rogers andrew at ceruleansystems.com
Fri Jun 25 22:39:17 PDT 2010


On Jun 25, 2010, at 3:07 PM, Jeff Bone wrote:
> One rare and minor quibble for JAR:
> 
>> Second, we have neither the mathematics nor computer science to actually implement warfare like this in a general way.
> 
> The rate of progress in that field is absolutely mind-blowing.  RT analytics and visualization software for combat vehicles and battlespace management.  Realtime AR distilling multiple data streams from all kinds of sources, doing target identification and visualization and fire control and even active defensive armor countermeasures like you were discussing.  It is absolutely astounding;  it's, like, a peek into what most folks think is a decade out.  


Yeah but the algorithms typically used are not remotely scalable enough. We're still in the toy zone. Algorithms required for real-time analytics, data fusion, augmented reality, etc are several orders of magnitude shy of current modest requirements, never mind what everyone would like to do. A couple of these problems are solved and not widely distributed but others (e.g. feature registration) border on being AI complete in terms of solving them. 

Most of the military systems like you mention above are based on pretty vanilla capabilities outside of some narrow algorithm areas like feature discrimination. DARPA is still giving people money to try to solve these problems.

A lot of these applications hit the algorithm scaling wall a long time ago. To use augmented reality and mirror worlds as an example, the US military hit the theoretical limit with off-the-shelf computer science in the 1990s -- which they still mostly use. We can throw hardware at them and make them prettier, but that only buys you so much. 


> Of course, that's the tactical angle.  There's also incredible progress on the network mapping end of things, and on the game-playing / scenario analysis / prediction / threat assessment and identification parts of it.  The "back office" of future warfare, if you will.  A couple of startups in that space worth watching that I'm aware of;  you track this stuff pretty well so you're probably aware of even more of them.


Okay, this part is getting much better very fast. There is also a very large capability gap between what is public and what isn't, especially on the backend. The real trick is being able to do the required operations at extreme scales; hundreds of thousands of CPUs if you want to make it interesting and you can't use distributed hash tables. 

A lot of companies are nominally working on these types of problems but as far as I know only three organizations have the tech to make it scale.


> I'm sure you're aware of all this;  your cautionary note clearly relates to the obvious extrapolations of the use of this stuff as it gets more mature and better integrated.  And sure, we're nowhere near that level of maturity yet.  But do you doubt we will be, and quickly?  


Not a doubt in my mind.


> The effects of what's already out there are already starting to be felt.  If we were a decade further along, this would be a major factor.  Given cycle times of change in military and the constant "fight the last war" problem, we should be anticipating all this and starting to restructure and retool around it *now.*  I don't see any evidence of that happening in anything but half-measures, but perhaps I'm missing it.


You overestimate how long the required technology has existed. Remember, these types of systems have to scale to a ludicrous number of processors to be useful in real-world environments and they have a limited ability leverage existing software toolsets. Give it a year or two. The military are huge fans but the *good* tech is very young.


More information about the FoRK mailing list