[Kevin M]

Originally Posted by sperry

I'm confused to how having people driving supply trucks in a convoy is somehow better than having robots drive them. Let's say I've got a convoy: 2 HMVs, 4 supply trucks, 1 Bradley. They get attacked. The HMVs and Bradley return fire to try to protect the convoy, the trucks scatter to take cover. If the trucks are shot up, we lose the supplies and the drivers.

The Bradley is wishful thinking. Anything with heavier weeaponry than an M2 .50 cal is usually on one end or the other of the supply train. But the 2 HMMVs and 4 trucks is reasonable enough. So let's compare what happens to aconvential force supply convoy, a robot-driven convoy with human soldier defense accompaniment, and an all-AI convoy.

Originally Posted by sperry

Now let's take the same convoy, except now the trucks are piloted by robots. They have essentially the same defensive capabilities, 'cept now, if they're destroyed we're just out some additional hardware. It's been shown that soldiers are far more valuable (both in ability and in cost to train) than hardware. IIRC each soldier costs something like $1.5M to train and equip. That's a hell of a lot more expensive than $300,000 in servos, sensors and computers.

It doesn't take near that much to get a soldier to the battlefield. Numbers like that come from taking huge portions of the DoD budget and dividing by the number of soldiers within it. In reality, taking me from Day 1 to the end of my enlistment cost the Army far less than $1.5 million. $500k, tops. But the fiscal argument isn't really the reason there won't be any independent AI on the battlefield .

Originally Posted by sperry

That's *why* we use robots for recon and other high-risk operations, the hardware is far cheaper than the people. If people were so cheap, we'd build suicide cruise missles, and wouldn't bother with drones and the like. But it's more economical to get a job done w/ a 50% success rate at 1/3 the cost than to get a 90% success rate w/ the loss of soldiers. It's got nothing to do with the sanctity of human life, it's just a matter of affording the fight.

Recon and supply are two completely different missions. What you're really missing in these statements is simple: re-supply is absolutely crucial to waging war successfully. In a war, by which I mean an extended conflict lasting years, such as WWI, WWII, Korea, Vietnam, etc., logistics are paramount to victory. You need bullets, gas, food, and clothes, or you can't fight at all. So, while 50% success at 30% of the cost saves money, it's no way to fight a war. Not only would our fighting units be insufficiently equipped to destroy the enemy, probably 2/3 of what didn't get to them would fall into enemy hands. More importantly, those supply convoys with no human defenders are going to be attacked far more often than they otherwise would be, and more successfully. Why bother fighting or even resisting the tanks, bradleys, and infantry, when all you have to do is stop them from getting fuel and ammo? Further, it is impossible to expect to create AI that, even if it could strategize and defend itself against armed combatants, would have serious issues distinguishing the enemy from non-combatants. Especially considering the human soldiers it's expected to emulate often have difficulty with it in modern warfare.

Originally Posted by sperry

As far as the "nothing extremely difficult" when it comes to autonomous vehicles... it's actually *extremely* difficult. What we've achieved is essentially a robot that can follow a map. In the 50 years or so we've been working on it, all we've come up with is "when you get to point X, turn left". Autonomous aircraft have the advantage of being able to employ canned routines (like a pre-programed anti-aircraft avoidance manuver), which make them look smart, but if they were placed up against a capable human opponent, they'd get whupped... ever play a video game?

The real reason that the Air Force is so interested in autonomous/remote piloted aircraft (aside from the aformentioned economic savings) is because we've just about reached the limits of human anatomy w/ regards to g-force. A remote/robot pilot can pull 20+ G's w/o breaking a sweat, while human pilots have about a limit of 9, and only for brief manuvers. An aircraft w/o a pilot can make up for it's lack of smarts by simply being so rediculously manuverable it's impossible to shoot down with current technology.

Unmanned aircraft and unmanned ground vehicles have much different circumstances affecting their ability to succed. You just pointed out why unmanned recon and attack aircraft are feasible- all that needs to be done is to create an aircraft that is agile enough and fast enough. If we make a plane that has a higher top speed than enemy manned fighter craft, all it has to do is pull 1, maybe 2 high-G maneuvers its opponent can't match, then boogie out of range. Creating such an aircraft that could avoid being destroyed even 90% of the time is probably well within the parameters of current technology. It would probably not be as successful at attacking targets, but it would be 100% successful in destroying enemy targets with zero human casualties, and probably at economic savings as well, at least in terms of operational costs, though perhaps not in developmental cost. Ground units do not have this advantage. If one combat group cannot kill the enemy combat group it comes into contact with, it usually dies.

While DARPA is demonstrating that AI drivers are potentially feasible, you'd still need human soldiers to defend the convoy at a minimum. So, why bother spending billions developing and deployingrobot drivers when the rifleman who has to be there anyway can do it better with minimal training and cost from the military? There is a parallel to this- The last generation of Soviet tanks had mechanical loading systems for their main guns, so they had crews of 3 instead of 4 like the M1 and the T62. Problem is, developing and building the simple system to move shells from one place to another 3 feet apart cost more than training another tank crewman for each tank, and is far less reliable. It also can't fire a rifle, or replace the driver or gunner of another tank if necessary the way the 4th crewman the US uses can. Also f it breaks, there's nobody there to load, and you can't hardly do it anyway because the broken stuff is in the way. I've seen the system General Dynamics developed to put in the M1 before the project was fully approved, and the Army was wise to decline to use it.

Reliability probably wouldn't be an issue with AI driver systems- the computer systems currently in use for communications and navigation in most combat line-unit vehicles are pretty bulletprof, both literally and figuratively. But it still can't kill the enemy.

Originally Posted by sperry

Finally, back to the Darpa project, there are some really good things that can come from this contest, outside of the military application. Besides just the experience with complex problem solving AI, I see this stuff ending up in the space program for remote rovers where the transmission delay is so long they have to drive themselves. Then the next step is AI for non-military vehicles... MattR and qksubi are gonna be outta work when robots are driving trucks and trains. And eventually I look forward to being able to get into my car for a long trip and just kicking back as the onboard computer drives me to Vegas at 180mph. I enjoy driving, but not for a long distance or in a straight line. The only problem is that we're probably another 50 years away from that sort of stuff... hell back in the 60's people assumed we'd have flying autonomous cars by 2000... when it comes to AI, people tend to drastically underestimate how difficult it actually is.

Absolutely. Darpa kicks ass, even though I'm on that unemployment list too if such technology came to fruition. It just has very limited potential application for the military, unlike most new technologies of the last 200 years. Yay for civilization, we're finally making new ideas into good things that aren't very useful for killing each other.