AI how advanced

[Xero010]

Wouldnt the real problem with AI mecha pilots be a lack of emotion and therefore the lack of the true will to live? Humans evolved to have emotions for this reason. So the real question is "how the hutch do we get an AI to be afraid?" An alternative would be to use a quantum computer based AI that can respond to fast paced situations during combat, running the necessary calculations in a sixth of second. On another note, has anyone considered installing a VI system into their mech that tells the pilot about approaching enemies as well as damage and helps with navigation, like Cortana from Halo

[ferrelferret]

[Joshua A.C. Newman]

Xero, in general, with real-life AI systems, getting stimulus-response is the easy part. Cockroaches know how to run away and we have robots as smart as they are. Hutch, I built one out of $10 of Radio Shack parts. The hard part is abstracting meanings from things, generating symbols that gel with other symbols and a limbic environment into a coherent sense of self, other, objects, and ground. Like, what's this?


or this?


Part of what's interesting about the study of artificial intelligence to me is that we make these proposals, like, "It's something that can have a conversation with you." And then we discover that, even though something can have a conversation with you, it still doesn't feel like a person; and furthermore, such a test turns out to give false negatives; humans fail the Turing Test all the time, even when properly administered. We used to say, "It's something that can use tools." But it turns out lots of animals use tools, and some of them are clearly not intelligent, though elephants and all apes do it at least some. It's a very subtle thing, intelligence. "Will to live" is clearly not an element of it, though. Mantises, as Chuang Tzu said, will bravely face a cart wheel. No lack of "will" there. Walking through a field of tall grass results in a shower of grasshoppers flying away. Likewise, no lack of will to live.

So, if you could make a robot that could feel suffering and avoid it, feel satisfaction and seek it, would you have an intelligence? Not if it can't tell the difference between a window and a painting of what lies beyond it. Not if it can't tell that you've changed your mind about something. Not if it can't paraphrase a series of tangentially related events into a coherent story.

A VI like Mila is an interface, not a true intelligence. It's a grammar parser and producer, not a heuristic, abstract, and emotional agent.

[ferrelferret]

[Xero010]

[Joshua A.C. Newman]

[Soren]

More to the point, the stuff we want AI for - higher-order abstract thinking and sorting processes - are very likely to be simulatable without consciousness. We don't want AI to be sapient and autonomous, because sapience and self-determination interferes with their designed purpose - serving us.

The one design spec that might plausibly produce 'genuine' AI - is spamming. The combination of faking out filtering mechanisms that depend on human perception + sentence structure, grammar, and responsiveness - that might be enough to produce, if not AI, then the first really good human-behavior simulations.

Of course, from the viewpoint of pure human chauvinism, we have a tendency to move the goalposts and say that if a computer can do it, it's not 'True AI'. I imagine we'll hit a point where very complex programming running on high-speed hardware can simulate consciousness with enough verisimilitude to fake us out in person unless we're specifically looking for it - expect the sexual, legal and child safety panics, and screaming that robots don't have souls and will drag us down to hell, to start around then.

[Joshua A.C. Newman]

Curiously, I came across this just last night: .

[Xero010]

I checked out a video of how the vehicles work, but not sure what to make of it. The behaviours i saw certainly due show what seems to be some level of intelligence. Maybe if they had solar panels then the actual need to find light would be more important.

[Foxfire]

There is a way to get an effect similar to what you are looking for without going the full blown AI route.

Assume that your group has made some breakthrough in pattern recognition that allows for very advanced autopilots. They have frames that are capable of following basic commands(attack this target, move here) while employing basic self evasive routines. Assuming that the frame autopilot can handle 80-90% of the basic functions necessary for piloting the frame, you now have a strong force multiplier. 1 human could control 5-10 frames instead of just a single one.

You end up with a force of mostly drones that require a human for higher level guidance. If you want a better idea of what that might look like, look at riggers in 4th ed Shadowrun.

[Joshua A.C. Newman]

If we're gonna do that, we can have 1000 hobby drones and IEDs. It just doesn't make sense in this setting.

[Foxfire]

We already have predator drones, and cruise missles. Are you saying that no one in MFZ has thought of using a predator drone full of C4 as a weapon? Of course they have, thus there must be something else in place that prevents this from being a good tactic. Today, it might have something to do with the fact that predator drones cost 4 million dollars a pop. If your drones are expensive enough, then using them as one shot weapons is not cost effective. Even today, Tomahawk cruise missles cost almost a million dollars a shot. Thus, cruise missles have a point where using them is not cost effective. If the enemy has defenses capable of shooting down cruise missles, they are even less cost effective.

Any effective drone on the battlefield will need:
1. Enough electronic defenses to avoid getting jammed or hacked most of the time.
2. Good enough AI to survive short periods of getting jammed or lost communication.
3. Enough sensors the give the remote pilot a good situational awareness
This gets expensive really fast.

Cheap hobby drones simply are not going to have the level of reliability necessary for use on the battlefield. Once the jamming and electronic counter measures start up, most hobby drones are going to be reduced to paperweights. Any drone that is reliable enough to be used on an active battlefield will be expensive enough that using it as a disposable weapon is a losing strategy the majority of the time.

But it is not entirely crazy to imagine a group in MFZ that for some reason prefers to use drones over regular frames. Maybe they place a much higher value on human life, and thus they don't care that battle drones are more expensive or less effective than an equivalent frame. Maybe they have a methods of communication that are immune to most normal jamming or advance AI that could fill in if the drone did get jammed. If a group did have a technological edge in the drone race, they would guard it zealously, lest other groups find ways to counter or exploit it. "Hey, that's a Wang Tech series 3 drone, it has known issues with tracking targets through thermal noise, launching flares".

As long as at the end of the day, a drone is slightly less effective and/or more expensive that putting an actual person in a frame on the field, it does not automatically follow that you end up with "1000 hobby drones and IEDs". That doesn't make it an unthinkable concept either.

[Soren]

This is a setting with highly portable laser weaponry. I invite you to consider what laser point-defense would do to aircraft. (Hint: Phalanx, thirty years ago, could track and bring down artillery shells in flight during testing, and practical anti-artillery systems are coming into use now. Aircraft, even supersonic aircraft, are a lot slower.)

I'm sure the explosions would be very pretty, though.

Various cost/benefit balances exist, of course - it's a developing galaxy, after all - and in some places, I'm sure aerial drone weaponry makes sense - because the defenses aren't present, because you can afford a thousand drones for two or three hits, or because you don't care very much about killing bystanders by accident. Pick your favorite.

[Ryujin]

[Soren]

[Foxfire]

[Soren]

I'm just going to leave this here: http://what-if.xkcd.com/5/

[gusindor]

Here's a thought; drones as 1-shot rockets. How much intelligence does it take to crash into something?

[Foxfire]

[The Hydromancer]