"Why are you still alive?"

ThinksMarkedly wrote:

ThinksMarkedly wrote:And as I said in my previous post in this sub-thread, who's to say that there wasn't an actual portion of randomness in the Alliance target selection? The Alliance does have access to the same software, if not better, and if they could predict their own targets using the same or better data set, they should assume the RHN could too. But you can't predict random.
[cut]
TNG episode "Remember Me," the single weirdest logical conclusion, ever: "if there's nothing wrong with me, maybe there's something wrong with the Universe."

Another Star Trek reference comes to mind: Ferengi Rule of Acquisition 75:

Once in a while, declare peace. It confuses the hell out of your enemies.

The best way to make sure you can't predicted is to be unpredictable. It's not too difficult to insert noise into your signal so those expert systems can't be accurately used.

Nope. That won't foil an expert system. That is why it is called an expert. These computer programs are no different than any other. Upon completion, test data is entered into it that should resolve to a specific conclusion. Then rogue data is inserted to see how it is affected. A professional expert will find a pattern if one is there, if the processing power - and the time - is willing and the creak don't rise. I have an inside track on the design. Some use a top down approach (trying to analyze a pattern using all data), some use a bottom up approach (taking a couple data points and trying to resolve it), but they all default to the opposite of what the default is if it catches no joy. So a pattern would be found in some of the systems Honor hit.

The problem with this powerful approach is processing power and time. Very good chess programs are on the market today that can beat grand masters. But the problem comes when the player deviates from the book openings. Most chess programs are too reliant upon professionals not opening up with dumb moves (following recommended book openings). So the programmer, to save processing time, memory and decision trees omit the attention to these decision trees. The program can still search those patterns, but they must be given loads of extra time. Which is prohibitive to playing a good practical game. Who wants to wait an hour for your opponent to move. Honorverse computers should be speed demons chewing up decision trees like bubble gum. So the expert system isn't handicapped from the beginning.

The expert system would simply find the pattern that exists between the corresponding systems. The expert system will indicate any pattern it found and list the data points that present the pattern. It will indicate which points lay outside the pattern it found.

Thanks for providing the name of the episodes.

ThinksMarkedly wrote:And as I said in my previous post in this sub-thread, who's to say that there wasn't an actual portion of randomness in the Alliance target selection? The Alliance does have access to the same software, if not better, and if they could predict their own targets using the same or better data set, they should assume the RHN could too. But you can't predict random.

It wouldn't be entirely random. But if you've got enough forces to hit 5 systems; but your initial target criteria gives you, say, 12 you might certainly randomly pick 5 from that 12 to attack. (Rather than, say, tightening your criteria until you were down to 5. Or letting the planners simply pick 5 off the short list using whatever ad hoc criteria they wanted).

Even that small amount of randomness would make it harder for an analysis or expert system to identify the true criteria being used -- because both would be at high risk of overfitting the data. Trying to figure out what extra (non-existent) criteria Manticore used to narrow down to the final 5.

That overfitting could easily lead to excluding systems that actually were on Manticore's next potential target list.



The problem from the expert system end is that while they're great at picking out correlations in vast amounts of background data; they're pretty poor at working out which correlations are likely to be predictive and which were just coincidence.
The best way to figure that out is to have even more data to use to validate the predictive model.

(Hence the stories of one occasionally hears of them learning entirely the wrong correlation -- like the one of the neural net that was supposed to be learning to identify camouflaged tanks but instead learned to identify sunny days. Because, oops, the folk running the trial didn't noticed that all the 'no tank' pictures in the training data set also happened to be overcast. That was a screw-up on whoever designed that training set data; but when you're trying to model enemy behavior you don't have a way to get more examples until after the enemy has attacked you some more)

And while the databases feeding the expert system presumably have at least tens of thousands of pieces of potentially relevant data about each system and planet in the Republic, with only 5 "picks" to initially work from an expert system is going to find tons of correlations -- but nearly all will be irrelevant coincidences. Even if you hold back picks from the initial model creation to use to attempt to verify it you're still working with just 5 - so you could make a model with just 3 and attempt to verify it with the other 2; but that's not enough pass/fail criteria to weed out most of the coincidences.

And as you pointed out, thanks to randomness, those 5 that were hit may not be the only 5 that matched every criteria Manticore used to pick targets. They may not even be the 5 that best fit the criteria.


Now getting 4 more hits in Cutworm II is going to help refine the model. But it's also an escalation. Manticore picked fewer, tougher, targets for II which means they changed their targeting criteria - which means it's not a pure verification of the original model.



I haven't double checked that it's correct, but the wiki claims that by the 2nd war the Republic of Haven only had 75 odd inhabited systems. Since the analysts' best guess was 10 primary and 15 secondary picks that means they basically said over 1/3 (25/70) of Republic's other systems might be Honor's next target (and even at that they still missed one of the systems that Cutworm II actually hit)

cthia wrote:Nope. That won't foil an expert system. That is why it is called an expert. These computer programs are no different than any other. Upon completion, test data is entered into it that should resolve to a specific conclusion.

If the data is randomly selected based on inputs from a chaotic system there is no pattern to analyze, regardless how much number crunching you do or what software you use.

Joat42 wrote:

cthia wrote:Nope. That won't foil an expert system. That is why it is called an expert. These computer programs are no different than any other. Upon completion, test data is entered into it that should resolve to a specific conclusion.

If the data is randomly selected based on inputs from a chaotic system there is no pattern to analyze, regardless how much number crunching you do or what software you use.

But it wasn't completely random. There was a method to Honor's madness. We need the textev of Honor's thinking. As I remember it, she gave some details on the reasoning behind her choice of targets.

But, if it had been completely random, then it really would have been left up to pure intuition.

Tarot cards or an actual tea leaf reading would have been even better in that case. Or the old-fashioned eenie meenie miney mo. The report actually mentioned the criteria that the intuition was based on. The pattern.

Joat42 wrote:

cthia wrote:Nope. That won't foil an expert system. That is why it is called an expert. These computer programs are no different than any other. Upon completion, test data is entered into it that should resolve to a specific conclusion.

If the data is randomly selected based on inputs from a chaotic system there is no pattern to analyze, regardless how much number crunching you do or what software you use.

cthia wrote:But it wasn't completely random. There was a method to Honor's madness. We need the textev of Honor's thinking. As I remember it, she gave some details on the reasoning behind her choice of targets.

But, if it had been completely random, then it really would have been left up to pure intuition.

Tarot cards or an actual tea leaf reading would have been even better in that case. Or the old-fashioned eenie meenie miney mo. The report actually mentioned the criteria that the intuition was based on. The pattern.

As Jonathan mentioned, if the operational parameters Honor specified fit X system which are then attacked in a random order, there is no pattern to discern. The only thing you could possibly deduce is a guesstimate of the operational parameters which only narrows down the number of systems that may be attacked - not the order.

In the end it just comes down to pure luck, a factor we shouldn't overlook. In this case the "luck" is derived from the authors need to add some suspense and dramatic action here and there in the story.

Joat42 wrote:

cthia wrote:Nope. That won't foil an expert system. That is why it is called an expert. These computer programs are no different than any other. Upon completion, test data is entered into it that should resolve to a specific conclusion.

If the data is randomly selected based on inputs from a chaotic system there is no pattern to analyze, regardless how much number crunching you do or what software you use.

cthia wrote:But it wasn't completely random. There was a method to Honor's madness. We need the textev of Honor's thinking. As I remember it, she gave some details on the reasoning behind her choice of targets.

But, if it had been completely random, then it really would have been left up to pure intuition.

Tarot cards or an actual tea leaf reading would have been even better in that case. Or the old-fashioned eenie meenie miney mo. The report actually mentioned the criteria that the intuition was based on. The pattern.

joat42 wrote:As Jonathan mentioned, if the operational parameters Honor specified fit X system which are then attacked in a random order, there is no pattern to discern. The only thing you could possibly deduce is a guesstimate of the operational parameters which only narrows down the number of systems that may be attacked - not the order.

In the end it just comes down to pure luck, a factor we shouldn't overlook. In this case the "luck" is derived from the authors need to add some suspense and dramatic action here and there in the story.

True as well. But they didn't need to guess the order, as I believe (Theisman's subordinate?) pointed out. They just needn't to guess one of the systems that would be hit and pull back units from the front line and let Harrington come to them. Possibly forfeiting the systems she hit before she got busted, but as was said, they would make up for it by hurting her more than they got hurt collectively.

Especially, as was said, if they could mousetrap Eighth Fleet, which almost happened. Or, possibly kill Honor.

It is the old screen play in football. Allow the offensive line to fold giving the defense a false sense of certainty. Then boom.

.

cthia wrote:True as well. But they didn't need to guess the order, as I believe Theisman's subordinate pointed out. They just needn't to guess one of the systems that would be hit and pull back units from the front line and let Harrington come to them. Possibly forfeiting the systems she hit before she got busted, but as was said, they would make up for it by hurting her more than they got hurt collectively.

Especially, as was said, if they could have mousetrapped Eighth Fleet, which almost happened. Or, possibly killed Honor. It is the old screen play in football. Allow the offensive line to fold giving the defense a false certainty. Then boom.

It's actually pretty easy to predict one target Eighth Fleet would need to come to at some point: Haven.

Forfeiting systems counts as an Alliance victory, because she comes in, destroys their industry, causes outrage by the populace and costs for the federal Havenite government. The senators for each of those systems will be screaming to get pickets in their system, so the RHN would be under pressure to divide and thus be destroyed in detail. And all with no losses on the Alliance side.

Therefore, forfeiting is not an option. It damages the republic while the Alliance grows stronger. Remember the Python Lump? This is also at a time where the IAN SD(P) are nowhere to be seen, clearly getting a refit, but the RHN doesn't know where or why, so they have to tread extremely carefully.

They had to guess better, for the next target or two.

They were actually very smart at it. They clearly couldn't guess them all, so they rolled the dice and covered a few systems with sufficient force to mouse-trap her, leaving the rest uncovered. They got lucky at Solon, which combined with their skill resulted in a Havenite victory.

ThinksMarkedly wrote:

cthia wrote:True as well. But they didn't need to guess the order, as I believe Theisman's subordinate pointed out. They just needn't to guess one of the systems that would be hit and pull back units from the front line and let Harrington come to them. Possibly forfeiting the systems she hit before she got busted, but as was said, they would make up for it by hurting her more than they got hurt collectively.

Especially, as was said, if they could have mousetrapped Eighth Fleet, which almost happened. Or, possibly killed Honor. It is the old screen play in football. Allow the offensive line to fold giving the defense a false certainty. Then boom.

It's actually pretty easy to predict one target Eighth Fleet would need to come to at some point: Haven.

Forfeiting systems counts as an Alliance victory, because she comes in, destroys their industry, causes outrage by the populace and costs for the federal Havenite government. The senators for each of those systems will be screaming to get pickets in their system, so the RHN would be under pressure to divide and thus be destroyed in detail. And all with no losses on the Alliance side.

Therefore, forfeiting is not an option. It damages the republic while the Alliance grows stronger. Remember the Python Lump? This is also at a time where the IAN SD(P) are nowhere to be seen, clearly getting a refit, but the RHN doesn't know where or why, so they have to tread extremely carefully.

They had to guess better, for the next target or two.

They were actually very smart at it. They clearly couldn't guess them all, so they rolled the dice and covered a few systems with sufficient force to mouse-trap her, leaving the rest uncovered. They got lucky at Solon, which combined with their skill resulted in a Havenite victory.

Oh, I agree that they didn't want to forfeit. But textev says that is essentially the plan they adopted. They had no choice as the textev stated.

However, being defeated in detail is counting chickens before they hatch, don't you think? Because the prize for mousetrapping Honor could have been the entire Eighth Fleet. The RMN's first string. The possible payoff was worth the risk. Plus, they had to make the Manties rethink their strategy. And there seemed to be no better option available.

The RMN would have been screwed if they lost Eighth Fleet and its commander.

cthia wrote:Nope. That won't foil an expert system. That is why it is called an expert. These computer programs are no different than any other. Upon completion, test data is entered into it that should resolve to a specific conclusion.

Joat42 wrote:If the data is randomly selected based on inputs from a chaotic system there is no pattern to analyze, regardless how much number crunching you do or what software you use.

cthia wrote:But it wasn't completely random. There was a method to Honor's madness. We need the textev of Honor's thinking. As I remember it, she gave some details on the reasoning behind her choice of targets.

But, if it had been completely random, then it really would have been left up to pure intuition.

Tarot cards or an actual tea leaf reading would have been even better in that case. Or the old-fashioned eenie meenie miney mo. The report actually mentioned the criteria that the intuition was based on. The pattern.

It is not going to help you, because it is basically what the "tea-leaf readers" thought it was; in fact the text about the tea-leaf guesses gives a better description of her plan. At All Costs, chapter 15:

"It's good to see all of you gathered in one place at last," she said, after a moment. "And, as Commodore Brigham commented as we docked with Imperator, it's about time. Eighth Fleet is officially activated as of twelve hundred hours, Zulu, today."
No one actually moved, but it was as if an invisible stir had run around the compartment.
"We can anticipate the arrival of the remaining units of our initial order of battle over the next three weeks," she continued levelly. "We're all aware of how tightly the Navy is stretched at the moment, so we won't dwell on that just now. I met with Admiral Caparelli immediately before my departure for Trevor's Star, however, and he emphasized to me once again the importance of beginning active operations as quickly as possible.
"Commodore Brigham, Captain Jaruwalski, and I have given considerable thought to the most appropriate initial targets for our attention. This isn't simply a military operation. Or, rather, it's a military operation with a political dimension of which we must be well aware. Specifically, we want the Havenites to divert forces to provide rear security against our raids. That means balancing vulnerability of target against economic and industrial value, but it also requires us to think about which target systems are most likely to generate political pressure to divert enemy strike forces to defensive employment.
"I'm confident we can find such targets, but accomplishing our objective is almost certainly going to require us to operate widely dispersed attack forces, at least in our initial operations. That means we're going to be relying very heavily on the judgment and ability of our junior flag officers—more heavily than we'd originally anticipated.

cthia wrote:Oh, I agree that they didn't want to forfeit. But textev says that is essentially the plan they adopted. They had no choice as the textev stated.

However, being defeated in detail is counting chickens before they hatch, don't you think? Because the prize for mousetrapping Honor could have been the entire Eighth Fleet. The RMN's first string. The possible payoff was worth the risk. Plus, they had to make the Manties rethink their strategy. And there seemed to be no better option available.

The RMN would have been screwed if they lost Eighth Fleet and its commander.

Yes the RMN would have been screwed if they'd lost 8th fleet.
OTOH Haven was lucky that they caught as little of 8th fleet as they did.

Honor hit Solon with 2 Invictus-class SD(P)s, 5 BC(P)s, 6 CAs, 8 CLs, 3 DDs, plus detached LACs. And it cost her about half of that to shoot herself out of a trap involving 18 Republic SD(P)s with screen, plus the Moriarty platforms controlling all the system's missile defense pods.

But if she'd had the rest of her then current Eighth fleet along she likely could have simply reversed course and blasted her way back out through Giscard's squadron of SD(P)s (Bogey Four) all the while fending off the fire from the flanking Bogies 2 and 3; and never come into effective range of the system defense pods.


Alice Truman, for the Lorn attack, had 8th fleet's other 4 SD(P)s [at least one apparently being another Invictus; and 2 known to be Medusas; not sure about the 4th], another CLAC squadron, at least 8 BCs (though older, non-pod, ones), around 6 CAs (including all 4 of their known Sag-Cs; while Honor kept the Star Knights and older Saganamis). So if Alice's detachment of 8th fleet had been with Honor that would have likely tripled the firepower and defenses the RMN had at Solon. Adding four more Keyhole II equipped ships, plus doubling the number of LACs, and throwing in the Sag-Cs, probably could have stood up quite well against the deeply stacked salvos Haven had to resort to to crack Honor's existing defenses. And as it was Giscard's wallers nearly shot themselves dry against far weaker defenses.

Still -- 18 on 4 SD(P)s is still enough of a force imbalance that I believe it would have been a strategic defeat, in that Honor still would have been driven from the field, but she'd have taken far lighter losses, and inflicted more damage, on her way out.