Weird idea for an unusual Reign of Steel scenario...

[jason taylor]

Originally Posted by Johnny1A.2

One possible plot hook for this thing involves its original purpose (at least as far as the U.S. military and CIA were concerned): to be a strat-tac supergenius, able to analyze wartime and intelligence situations and predict outcomes accurately. It was able to do this...when it worked at all. It just didn't work reliably enough and often enough to be useful.

Project that forward, and assume the Sheffield was able to embody an analysis engine that could just what the sponsors wanted. Also assume that light-speed delays and cooling issues mean this thing actually 'thinks' slower than a human.

Now imagine the Z-minds reaction to an AI that thinks at the speed of molasses in January, but also displays the ability to outthink them at war-related matters, albeit it takes it a little while...from their POV, it would be a sort of idiot-savant, simultaneously disturbing and dangerous...

Such a thing would have the advantages and disadvantages of a Deep Blue. It could calculate possibilities faster then any human but it can be fuddled by a human who was deliberately being "stupid" to fuddle it the way Kasporov got a few licks in by making unorthodox moves.

Another problem is that computer's can't calculate honor or religion or revenge or any of the things humans really fight over. A computer might work for an eighteenth century war where the rival states really were fighting over calculable things.

[Flyndaran]

Believing that honor or religion are purely random and have zero pattern is a rather odd stance to take.
In genre for 60s style science fiction humans-are-awesome regardless of logic, but otherwise kind of weird, in my opinion.

[jason taylor]

Originally Posted by Flyndaran

Believing that honor or religion are purely random and have zero pattern is a rather odd stance to take.
In genre for 60s style science fiction humans-are-awesome regardless of logic, but otherwise kind of weird, in my opinion.

They aren't random but assuming a computer can calculate them is also odd.

For instance suppose the human knows he is being analyzed by a computer. That can make him mad enough to do something illogical just to prove the computer wrong.

In any event, a computer is always limited itself by the fact that a human built it. Therefore it will only make such calculations as it is programmed to make.

Computers are not really brains. They are only smarter then humans in the sense that a bulldozer is "stronger." But any human can wreck a bulldozer by throwing goo in the engine.

[Flyndaran]

All that assumes the A.I's aren't "really" sapient which is somewhat begging the question. We also obey our programming far more than most of us are willing to admit. Evolutionarilly created, nurtured upbringing, and basic self interest fit with probably over 99% of our decisions.
It's hard to imagine getting groups of humans to violate all those to confuse a logical intelligent tactically skilled A.I. Not impossible, but certainly unlikely in the short term let alone long term protracted war.

Your idea sounds more like loads of unspoken disadvantages and low IQ just with high specific skills.

[jason taylor]

Originally Posted by Flyndaran

All that assumes the A.I's aren't "really" sapient which is somewhat begging the question. We also obey our programming far more than most of us are willing to admit. Evolutionarilly created, nurtured upbringing, and basic self interest fit with probably over 99% of our decisions.
It's hard to imagine getting groups of humans to violate all those to confuse a logical intelligent tactically skilled A.I. Not impossible, but certainly unlikely in the short term let alone long term protracted war.

Your idea sounds more like loads of unspoken disadvantages and low IQ just with high specific skills.

I see it everytime I hear of a terrorist doing an atrocity and some rational Westerner asking "Why do they hate us?" Why would a computer be immune to the prejudices a being made specifically for calculation would naturally have?

Now if a real AI was created by the CIA it would by definition have the prejudices an intelligence analyst would have. As it would be isolated from contact with other sources by secrecy to an extent that a human analyst would not, it would not have a correcting mechanism for it's prejudices. It would for instance certainly not be plugged into the internet and would only get Sigint fed to it by human couriers carrying predigested data lest it be discovered. At best it would be like any human raised in a cloistered setting and the fact that it is a bigger calculating machine only enables it to make bigger blunders.

For instance would a computer be able to predict that an English public school boy(all of whom as everyone knows behave so similarly that they can run an Empire on autopilot) would be a Communist spy any better then a man could?

The point is not that such a thing would not be a very useful tool. The point is that it would be very easy to mistake it as a substitute for all the other tools in the CIA's toolbox.

[Flyndaran]

Either this A.I. has the intelligence of a human ie. no extra disadvantages, and a high Tactics/Strategy skills or it doesn't.
That's all I mean, and am not sure what specifically you're arguing against.

[jason taylor]

Originally Posted by Flyndaran

Either this A.I. has the intelligence of a human ie. no extra disadvantages, and a high Tactics/Strategy skills or it doesn't.
That's all I mean, and am not sure what specifically you're arguing against.

You seem to be shifting. Is an A.I. a computer or is an A.I. a being with flexible intelligence, aesthetic instinct, ability to empathize with biological and psyhological needs and all the rest of it. If so it is just a human with a metal body.

Which humans are we talking about? There are plenty of humans and the A.I. will have the intelligence and disadvantages of the humans who work it and the humans who create it added to it's calculating potential. Humans who are supposedly the most intelligent people on Earth often make exceptional errors for varirous reasons such as using corrupted data or overestimating their own potential(for instance making statements on trades outside their own without allowing for error).

The argument is against the overestimation of the ability of such a machine. Not least against the argument that logic and calculation is a sufficient tool for intelligence analysis. Just to start with, logic will be what the user considers logical, or in other words an end worth the means he is proposing. Many of the greatest intelligence errors come from assuming a rival has the same priority in objectives. An AI will have no more ability to access those then a human spymaster.

[Bruno]

AI only understand as much as the information given to them; human brains have a lot of wetware devoted to filtering out "irrelevant" information and any data that goes through humans is going to get forced through that filter.

There was an experiment with training an AI to make unbiased recommendations on prison sentence lengths, but it had a big problem in that the AI was trained on a database of actual assigned prison sentences - with all the bias that the actual human judges had when assigning them. They got an AI that was very good at producing the same prison sentences, with all the same biases.

Google's running into this problem right now with their panic-mode on Youtube. They're two weeks into an advertiser boycott over having their advertisements run next to "objectionable" videos. So Google has assigned one of their big AIs to the task of trying to sort out "objectionable" videos - but the only way for the AI to know what's objectionable is based on humans pushing the Report button.
So they now have a NEW problem of people trolling the AI by mass reporting videos on subjects they just don't like, or people they just want to troll.
This can be (eventually) improved by having humans sort through the "objectionable" reports and tossing out ones that are deemed to be invalid, but that a) takes time and humans that Google doesn't have right now and b) relies on the team of humans having consistent definitions of "objectionable".

[robkelk]

Originally Posted by jason taylor

...
Which humans are we talking about?
...

Not "humans", just one human: Dr. Carl Sheffield.

Originally Posted by Johnny1A.2

...
What not even his backers in the intel and military commands knew that Dr. Sheffield was a bit nuts, as well as being a genius. He had used a process he invented himself to 'imprint' his own personality on the great machine, but the process had gone wrong, and Dr. Sheffield was accidentally electrocuted. But the imprint did take, leaving his personality as a latent subconscious in the great machine, ready to awaken. This also prevented any 'natural' awakening of the machine separate of the latent human imprint.
...

(This is still the thread that Johnny1A.2 started to discuss crossing Atomic Horror and Reign of Steel, isn't it?)

[jason taylor]

Originally Posted by robkelk

Not "humans", just one human: Dr. Carl Sheffield.

All the worse as the computer will have the prejudices of one man and unless Sheffield is an expert analyst as well as designer, it will in affect make the computer be a man working outside his trade.