How smart were CASPAR drones?

[marauder648]

Because i'm curious.  I've seen accounts of them being sentient and aware, or of them being like HAL, or simple machines that solve problems amazingly fast, but I can't find anything definitive. 

So..how smart were they? Where they aware and sentient?

[Liam's Ghost]

The official canon position is that the Caspars were not sapient.

[Giovanni Blasini]

I will be monitoring this thread with interest, but, sadly...yeah, what Liam said.

[Liam's Ghost]

There's still a lot of room for interpretation if one wants to step outside the standard continuity though.

[Mendrugo]

Combing through the various references, we have the following information about the Caspar AI:

The ultrasophisticated computer system (the "pinnacle of artificial intelligence") could operate autonomously with only a handful of humans monitoring the system, in both passive and aggressive modes.

NETC designed the AI programs for the automated warships.  They developed the first "mental map" of the human brain - Admiral Kinru Dvarahal, one of the Hegemony's greatest naval strategists.  The mental map wasn't placed into the Caspar drones directly, but was placed into a special battle computer network (along with numerous other of the best SLDF admirals) - the Autonomous Tactical Analysis Computer (ATAC).  This was an an advanced neural network that could process millions of simulations a second and then relay the results out to all the ships connected to it through a massively redundant network of HPG-equipped command-and-control satellites.

The drones used tactics that long-dead admirals had programmed into their computers.  The AI used this tactical knowledge to identify and prioritize enemy vessels according to threat level, calculating which vessels would harm the attacking force the most through their destruction.  They prioritized heavy destroyers, cruisers, troop transports, and Pentagon DropShips.  They rarely wasted a single shot, and when they calculated that their firepower was insufficient, they switched to ramming tactics.  They were able to recognize attempts to decoy them off station, and largely ignored them.  It calculated when the crews of enemy vessels would be at their lowest ebb - having not slept for over 24 hours, and chose that moment to attack.

Most Caspar AI actions are dictated by a list of pre-determined events and reactionary orders.  However, the computers were capable of fast and intuitive thinking that was superior to the structured programming of older robotic control systems, and verged on true sapience. 

------------

So, from the looks of it - they aren't sapient.  They don't dream of electric sheep.  They can, however, run through millions of simulations in seconds and determine primary threats, vulnerabilities, and likely objectives and tactics, as well as countermeasures.  The trillions of "IF" "THEN" setups that the pattern recognition and recommended responses are matched against derive from the mental maps of Admiral Dvarahal and whoever else they stuffed into the ATAC.

[Cryhavok101]

Reading the rules for them in interstellar operations makes me think they are about as intelligent as our current video game AIs can be. They have set reactions to predefined scenarios programmed into them by the star league's greatest admirals. They follow what amounts to a flow chart in making decisions, and can only really switch which flow chart they are following between a small number of them, depending on how aggressive they are set to be.

Interstellar Operations gives a bunch of restrictions, mechanics wise, that lead me to believe that, however, fluff wise, the same book makes them seem like more:

Based on the first complete and successful “mental map” of a human brain, these computers were capable of fast and intuitive thinking that easily trumped the structured programming of older robotic control systems. Verging on true sapience, combined with the experience of one of the Hegemony’s greatest naval strategists, the AIs that resulted could out-think all but the greatest military minds of the Star League.

The underlined part seems to me to be indicating that it was close to true intelligence, but not quite there.

[glitterboy2098]

Basically they would be like the HAL9000 in 2001 and 2010. Able to imitate humanity in narrow areas pretty well, but decidedly not sapient.

[Giovanni Blasini]

Welcome to Nebula California also gives another clue to Caspar intelligence.  They mention that the AutoMechs are capable of passing a Turing test, and of some degree of machine learning, but that the Interstellar Expeditions team wasn't sure they were really sapient. They also believed that they were more advanced than Caspars in their AI.

[Maingunnery]

It can also be that CASPAR drones were really smart, but that their intelligence was highly specialized, then they would still look dumb. 

[Dayton3]

If they were really smart they would never have allowed themselves to be controlled by Stefan Amaris.

[Alexander Knight]

There's a difference between being smart and having free will.

[Natasha Kerensky]

I think a lot turns on the term "mental map".

If by "mental map", we imagine the actual brain architecture and neural signaling of SLDF admirals getting copied and simulated in electronic form, then the CASPAR drones were arguably intelligent in a human-like way, i.e., they had strong artificial intelligence (AI), also known as artificial general intelligence (AGI).  This is true scifi in the 2001/Star Wars droids/Blade Runner/Terminator/Star Trek TNG Data/Matrix/AI/I Robot/Ex Machina/Chappie vein.  Even today we really have no idea how or why cognition, sapience, or consciousness emerges in human and other animal brains, nevertheless how to recreate it artificially.

If by "mental map", we mean that the decision making processes of SLDF admirals were written down and expressed in terms of computer software, then the CASPAR drones only had weak AI, which is really just massive data processing and pattern recognition that can or cannot be hooked up to value preferences and a decision tree.  Weak AI is emerging in spades today, in everything from voice recognition software to driverless cars to grandmaster-beating chess computers.  But it is very narrow (can only do one thing), and it does not imply cognition, sapience, or consciousness (Siri understands your voice but not your train of thought).

[The term "mental map" as actually used today involves neither of these definitions, FWIW.]

Although the differences between weak and strong AI existed back in the 1980s and 1990s when the CASPARs were first written up, those differences were probably not as well appreciated back then as they are today with the emergence of weak AI systems in the real world.

Given that BattleTech is supposed to be an extension of 1980s technology, I would stick with the weakest weak AI when describing the CASPARs.  It should be little more than a lot of ship data getting reduced to some key variables running a kajillion quick simulations, the best of which is picked using a value matrix and decision tree based on how a bunch of SLDF admirals reacted under similar conditions in a bunch of war games.  Yeah, it's hooked up to fusion drives and naval-grade weapons and has tons of safeguards, but at its heart, a CASPAR drone should be little more than a muscular version of IBM's Deep Blue supercomputer when it beat Gary Kasparov back in 1996.  Deep Blue ran on specialized VLSI chips, while today's weak AI processing typically runs on gaming GPUs.

My 2 cents... YMMV.

[truetanker]

ROE


Simple: Rules Of Engagement

Every ROE has a number ( # ) assigned to each. Lowest #'s mean more weight in decisions. More weight = faster response time, which equals faster decisions. Mind you a ROE can " stump " any AI into conflict, then it " seizes " up until a resolve is worked out.

0 divided by 0 = 0, but 0 divided by .0 = conflict undefined.

In such cases, another ROE comes into effect causing it to make a decision to force the balance, if unable an auto-reboot might happen. That's why we have Internal DNS problem solvers to help out... ask a geek or better yet one of the nerdy geeks like Xtol or better yet Kit, to tell you about them.

In other words, if the # isn't low enough to warrant a resolve, continue with SOP.

Example:
ROE#21 says SILNTOPS = low thrust, .25% or less, inactive defense weapon systems, NO squack unless IDENT from SDSCOMM, ( ROE21a: if 25% fuel , ROE#21b: if 50% armor return to SDSBASE )
ROE#20 says if UFO detected, thrust IN for close-in flyby sweep, if NO detection continue ROE#21
ROE#13 says if attacked, activate defense weapon systems and goto ROE#10, Squack IDENT to SDSCOM with SitRep
ROE#10 says defend @ range, Medium, squack IDENT to SDSCOM with SitRep
ROE#04 says attack @ range, Short if TARGET is Large, squack IDENT to SDSCOM with SitRep
ROE#01 says full thrust and attack ram target, squack IDENT to SDSCOM with SitRep
ROE#00 says REPEAT ROE#01, squack IDENT to SDSCOM with SitRep

TT

[SteelRaven]

I'm going with 'the AI in your video game' smart. Well program but still little more than a automated gun-turret (that could fly)that probably need some sort of supervision from ground control in case it locked on to a civi target without a fof indicator.

[Giovanni Blasini]

It's worth noting that a Caspar type AI in the SLDF planning division predicted the fall of the Star League as far back as Jonathan Cameron'a reign, IIRC. It was certainly before Simon Cameron took power.

[Feenix74]

Thankfully for the BT Universe, the Caspar AI was not smart enough to recognize the greatest threat to the Caspar AI's existence was human-kind and launch pre-emptive strikes to remove the threat. Unlike Skynet, Colossus/Guardian/World Control, etc.

[Cryhavok101]

Thankfully for the BT Universe, the Caspar AI was not smart enough to recognize the greatest threat to the Caspar AI's existence was human-kind and launch pre-emptive strikes to remove the threat. Unlike Skynet, Colossus/Guardian/World Control, etc.

Oddly, this comment makes me want to run a 1st Star League era aToW game modeled after Tron.

[monbvol]

It is an interesting question that I've realized does have a bit more of a complicated answer.  The Caspars themselves don't actually seem that smart but their controllers do.

Do have to keep in mind that the fluff does seem to indicate that the main predictive/strategic/tactical intelligence of the whole SDS system was not aboard the M5 Caspar drones themselves.  This is evidenced by the fact that once the SLDF developed special jamming gear the Caspars were a lot easier to spoof and lure into unfavorable engagements.

Still the implications of how effective the SDS was when not jammed make me feel like it is underselling how sophisticated the AI programming really was by calling it "video game AI".

[SteelRaven]

Whenever I say Watson, people make the leap to Skynight.

[Jellico]

Thankfully for the BT Universe, the Caspar AI was not smart enough to recognize the greatest threat to the Caspar AI's existence was human-kind and launch pre-emptive strikes to remove the threat. Unlike Skynet, Colossus/Guardian/World Control, etc.

If they were really smart they would never have allowed themselves to be controlled by Stefan Amaris.

Or did they?

[Weirdo]

Or did they?

Oooo...now THAT'S a fun conspiracy theory! If we assume that:

1: The Caspars were smarter than any humans suspected, at least smart enough to 'want'.

2: They were, for the time being, relatively content in their role as system guardians(you gotta admit, it's a pretty cushy gig for a WarShip).

Then it actually seems plausible for them to play along when Amaris's mooks tried to suborn them. After all, they can listen to the news as really as anyone else, and also have access to plenty of SLDF intel through the SDS network. They know Richard has gone off the deep end and is getting more paranoid by the day(they'd probably view it in the context of a programming error leading to a feedback loop), and fear that sooner or later he's going to take a good close look at the massive battlewagons constantly flying about without any real supervision. If he were to discover just how aware they were, there's no way he'd leave them unfettered, or even intact. Similarly, they may not trust Kerensky either, as he might view them as a threat to the SLDF. On the other hand, Amaris habitually turns a blind eye to the activities of his minions, letting them get away with whatever they wish as long as they also carried out his orders. For the Caspars, that means continuing their cushy gig. If their version of morality doesn't care what Amaris does to humans, they may not be concerned by his actions on that front.

[Giovanni Blasini]

Even better: not all Caspars were suborned. Even they were split in opinion and loyalties.

[Maingunnery]

What about the Drone that rammed Pluto? Suicide? Insanity? Emo?

[Giovanni Blasini]

Maybe exactly what they said on the tin: faulty wiring and control relays.

Otherwise, maybe it talked to the AI that predicted the fall of the Star League, and decided it didn't want to stick around for that.

[Daryk]

Oddly, this comment makes me want to run a 1st Star League era aToW game modeled after Tron.

I don't think that's odd at all...

[truetanker]

Even better: not all Caspars were suborned. Even they were split in opinion and loyalties.

Sounds like this:
https://www.youtube.com/watch?v=nLS2N9mHWaw

TT

[worktroll]

Even better: not all Caspars were suborned. Even they were split in opinion and loyalties.

Someone ought to write some fanfics based on this ... oh wait!  ;D O0

(I miss Sybil & Tabiranth ... )

[Daryk]

Seconded!  Those were very interesting stories...

[pheonixstorm]

Still the implications of how effective the SDS was when not jammed make me feel like it is underselling how sophisticated the AI programming really was by calling it "video game AI".

The problem here is that Game AI CAN BE hyper intelligent. The problem most game companies run across for AI development is cost and time. AI doesn't sell a game, eye crack does, so generally the AI is made to be passable as far as how it works. The number of games where the AI uses advanced techniques are few and far between. It is possible today to get the SDS level AI up and running for games but it would take a good while to "train" it for the most part. iirc there were a couple games in the 90s that might fit this to some extent. One of which had you control a bot (or some such) that would learn as you played and if it figured out that you were constantly trying to kill it eventually it would stop listening to your input (an AI revolt if you want). There was also the AI training in the game Black & White. That game gave you a choice on how to raise your pet. Beat it, mistreat it, and force it to eat the people and it would eventually become an evil creature. Do the opposite (praise, pet, etc.) and it would become a good creature.

We can easily create artificial intelligence, what we can't yet create is artificial life (AI v AL).

[Giovanni Blasini]

Someone ought to write some fanfics based on this ... oh wait!  ;D O0

(I miss Sybil & Tabiranth ... )

I can take a hint. Only have that, Diana, and the Syberia fic to work on...

[monbvol]

The problem here is that Game AI CAN BE hyper intelligent. The problem most game companies run across for AI development is cost and time. AI doesn't sell a game, eye crack does, so generally the AI is made to be passable as far as how it works. The number of games where the AI uses advanced techniques are few and far between. It is possible today to get the SDS level AI up and running for games but it would take a good while to "train" it for the most part. iirc there were a couple games in the 90s that might fit this to some extent. One of which had you control a bot (or some such) that would learn as you played and if it figured out that you were constantly trying to kill it eventually it would stop listening to your input (an AI revolt if you want). There was also the AI training in the game Black & White. That game gave you a choice on how to raise your pet. Beat it, mistreat it, and force it to eat the people and it would eventually become an evil creature. Do the opposite (praise, pet, etc.) and it would become a good creature.

We can easily create artificial intelligence, what we can't yet create is artificial life (AI v AL).

I know we're capable of doing some pretty impressive stuff now(IBM's Watson scares me just a little) but I'm not sure we're quite to the point of what the SDS was capable of yet.

Either way you are quite right.  The Intelligence part is easy and the SDS certainly meets all the requirements for it.  Sentience/sapience is another question entirely.

I can take a hint. Only have that, Diana, and the Syberia fic to work on...

If it helps any I'll keep working on getting Liam's Ghost to watch a certain anime that I'm reasonably sure would give him some bad^TM ideas for his own stories.

[Farnsworth]

There's always the "Neural-Dimensional Computer Technology" bit on pp54 of the Star League book.
 
"Not able to think but its ability  to calculate and make rational decisions was a quantum leap beyond the capabilities of computers."

It pretty much describes 1980's imagineering of quantum computing Qbits combined with something like neuromorphic computing. SIBYL may not have been able to think like a human but as an ultra fast expert system mimicking human thought, it would likely have fooled an army of experts.   

It's kind of cool/scary that we will be closing in on this sort of technology within a lifetime.

[Sigil]

Personally, I credit NETC with developing the first autonomous artificial intelligence in the BTU based, at least initially, on the "neural map" of Admiral Kinru Dvarahal.  I always figured they "made a machine in the likeness of a human mind" and by the end it became both autonomous and sentient. 

To me, it's much the same as the "is there life on Mars" question.  It comes down to what we as people accept as "life."  In much the same way we are physically limited to seeing only a portion of the spectrum, so to are we limited by our own understanding of "life" and, in this case, what constitutes "intelligence."  We can only perceive the effects and never the cause.