Comment by bdg | Hacker News Reader

Comment by bdg | Hacker News Reader

Parent Full thread

As a human analogue to the drones, it is not unrealistic to imagine that phones, shovels, or other objects have been mistaken for guns by an observer in a tense situation (for example: a SWAT raid). Granted, one is clear visual cues that are confused by a human mind in the heat of the moment, while the other is bad quality readings. We're talking about two different implementations of the error.

This drone project could go one of two ways (ending the project is highly unlikely). The first way is upgrading the video quality, but this brings a number of issues with it such as latency (quite important in combat as any on-line gamer will tell you). The second is to provide AI to the drones... which frightens me when we can't get a voice recognition phone system to work. Google cars are just driving now after years of research (not flying, evading, identifying IEDs, identifying shovels, and engaging targets strategically on any given battle zone)... I don't expect the drones to fair better at this task because they're funded by the gov.

The only way the drone program would be pulled for a while is if the back doored encryption protocols were routinely exploited. We won't see an end to drones, or any end to proliferation at home or at war, and I'm skeptical we'll see drones get "better" any time soon.

The crucial difference is that drones are used in contexts where they have a very high false positive rate. It doesn't matter how accurate your missiles are if you are operating in an area where it is impossible to accurately identify targets. Considering the total lack of oversight and regulation on these secretive programs, it is unsurprising that they have the latitude to cause large amounts of civilian damage without inviting any limits on their activity.

Do drones have a higher false positive rate than human-piloted air support? Numbers are going to be hard to come by -- the military doesn't want to reveal anything. Qualitatively, the Collateral Murder video that wikileaks posted a while back (in which a US helicopter gunned down a pair of Reuters reporters because they were carrying objects that could have been weapons) sounds very similar to what's described here.

The data on both are pretty much impossible to find. Having looked at a number of civilian think tank studies into the issue, even if you come up with a statistic like 3 militants killed per civilian (as reported in some studies) it is very hard to say how that compares to other methods, or if alternatives even exist. Most of the comparisons in support of drone ops are to the wholesale target:CIVCAS for the entire ground forces, which of course is inflated by actual engagements with enemies.

I think we should look at this in terms of whether or not we should even be operating with outright force in the hinterlands of Yemen or AfPak, where there are very high concentrations of civilians. "Kinetic operations" with high explosives are, in my opinion, highly inappropriate for dealing with insurgents hidden among the population. Our intelligence isn't good enough, and is often gleaned from huge leaps of SIGINT logic, leading to entire families being killed over nameless metadata correlations to suspected terrorist nodes.

How does upgrading the video affect latency?

Drones already have long latency times, their "flight" which is the bit that requires potentially quick response times is already "AI" controlled.

People don't move so quickly you can't handle latency in terms of targeting. Vehicles can be identified and the weapon system can handle tracking them - latency doesn't come into it.

Of course they will improve. It is like saying 4 years ago that smartphones won't improve - they are already small, have limited battery, limited throughput on mobile internet etc. Drones will improve like any modern technology that's being used.

Reply on news.ycombinator.com