Hacker News Reader

Comment by srl | Hacker News Reader

The real reason, of course, is that war is just so much cooler when people are dying. Sources: G. Carlin.

Honestly, though, I fully expect AI to advance sufficiently in the next 20 years that a 10s delay is sufficiently small, and the AI can handle anything likely to occur in under 20s on its own.

View on HN

>I fully expect AI to advance sufficiently in the next 20 years

Yes, people have been expecting AI to become human-like within 20 years for the past 6 decades.

http://en.wikipedia.org/wiki/History_of_artificial_intellige...

("Researchers expressed an intense optimism in private and in print, predicting that a fully intelligent machine would be built in less than 20 years")

Did you read the rest of his post? He didn't predict general intelligence within the next 20 years. Rather, he predicted that the special purpose weapons AI would be sufficient to handle the sort of latencies that would happen in some hypothetical space combat.

Perhaps I'm reading your posts wrong, but it seems a fairly conservative prediction to me.

We're talking about replacing people in combat vehicles with AI. Considering the need for tactics, strategy and improvisation in a conflict that can take place light-years away, that won't work with anything less than general-purpose AI.

Where is anyone talking about conflicts light-years away? I think we may be discussing different things. The article itself concentrates on intra-system conflicts, primarily because it is trying to deal with realistic space combat using current technology. FTL drives are not part of that paradigm.

Yes, obviously if something is light-years away you would need to develop general purpose AI to make those decisions or send a command ship with a human crew to the location, but the post you are replying to was discussing 10s to 20s delays associated with local space combat.

> The article itself concentrates on intra-system conflicts

Neptune is more than 4 light-hours away, and the Oort cloud (discussed as a possible source of projectiles) a full light-year further. So I don't see how anything short of general purpose AI can be tele-operated for intra-system conflicts.

Again, why does it require general purpose AI?

The two examples you cherry-picked are the furthest objects in our solar system. You've literally picked the outlying examples and made them the rule.

The same "problem" could be solved by sending a bunch of what are essentially missiles-carrying-missiles with algorithms that boil down to orders to shoot at any objects that are hot or fast moving or match certain radar signatures. This technology has been in place since the first heat-seeking missile... 1956! There are plenty of other solutions that do not require a general AI (such as sending command ships out).

The point is that the 10-20 seconds latency that was mentioned is not even close to covering the next planet, much less the entire solar system. Yes, you can send off a bunch of heat-seeking missiles, but, as I mentioned in my original post way up this thread, these missiles are routinely defeated by counter-measures even when seconds away from the target. So, in the absence of strong AI, it's vastly more efficient to actually have a few humans on board the ships.

So, in the absence of strong AI, it's vastly more efficient to actually have a few humans on board the ships.

Have you done the math on that? Humans (and more importantly their life-support systems) are hugely, hugely expensive in terms of the carrying capacity of a space ship.

There's also the rather large cost in reduced maneuverability when you want your on board humans to be able to survive.

How does a human-operated ship mitigate anti-missile counter-measures better than an AI operated ship, assuming the same munitions? You have yet to give a single concrete example. If anything, a human-operated vessel would have a worse reaction time than a machine-operated vessel (or AI-operated if you want to be romantic). This is simply due to the physics of operating a biological construct to mechanically manipulate things instead of an electronic construct to manipulate much smaller things on a much smaller level.

And to sum up, you've used these as synonyms in the argument:

strong AI

general-purpose AI

human-like AI

These are completely different things. What, exactly, are you referring to?

Huh, why do you require human-like AI to control a bunch of high-velocity objects, calculate intercepts, and release other high-velocity objects on those intercepts?

The only human-level decision that needs to be made is simply confirming the object is a target and should be attacked.

Most people haven't served in the military, but have you ever played a war game? Deciding where to move the weapons platforms and which objects are targets for what is the very essence of fighting a successful battle.

Have you ever played against the AI in Starcraft or other countless video games with AI?

And? If you are 10 light seconds away from a battle your round trip control delay is 20s. Also, if weapons are zipping around at 2,000 km/s then 10 light seconds is only 25 minutes flight time. We certainly won't have 2,000 km/s projectiles in 20 years but in 100 we might.

Reply on news.ycombinator.com