Robots in the honorverse

Silverwall wrote:Much of it is an offshoot of the dumber than a bag of rocks computers that are present in the Honorverse.

With computers that limited robots would be of very little value. Especially in a naval context.

Reading RFCs other work it is clear that he has no problem with drones and the like, especially in a ground combat context. Therefore the lack of automation visibility is probably because we never really see organised ground combat though the Hexapuma's marines do use a variety of drones in the raid on the rebel base. I am sure there are plenty of industrial robots following pre-programmed tasks in industry.

Robots in classic sci-fi sense for other roles rely on what could be described as "true AI" reminiscent of Data on star trek or even that posessed by the terminator and this is explicitly something that is avoided in all RFCs writing.

It is also somthing that is perpetually 2 decades away in the real world and has been that far away since the topic started in the 60s. In fact estimates of the development timeframe of True AI range for 2 decades to centuaries to millenia to NEVER. Personally I feel true AI is hundreds of years off given how little we understand about the mind right now.

RFC (IIRC) has stated on several occasions that computers are quite sophisticated in the honorverse and that expert system AI is commonplace. What you don't have is much general AI. You'd need a TON of expert system AI to run an honorverse starship, with vessels and weapons traveling at relativistic speeds. The execution details of navigation, energy weapon employment, etc. are far beyond human crew. Humans *are* shown a bit too much in the executive decision slots at all levels of HV starships, which might be stretching things a bit... not sure you'd ever need a crew on a grazer mount to do anything other than try and fix it if it takes damage.

I think there's a good chance we'll see some general AI much sooner here in the real world than 'hundreds of years', as the latest systems today are approaching the computational power of mammalian brains. You don't need to 'understand' the mind at all to build an AI... you just need to build something sophisticated enough that it can learn. I do agree that 'programming' an explicit AI will likely never happen... more likely the march of ever smarter general purpose AI till the day when you're hard pressed to tell the difference.

Is a taught AI that can interact just like a human 'alive'? That's an existential question, not a technical one. The fundamental question here is whether or not an information storage and retrieval system can grow sophisticated enough to become self-aware. Such an AI would be very, very alien. Regardless, this is a well-tread area in SF already that the MWW doesn't want to go down in the HV.

If you haven't read them yet, Ian Banks' Culture novels have some of the best treatment of this topic I've ever seen.

The reason is fairly clear: he wanted a universe that he could model on "Age of Sail." That means he didn't want a universe where there were autonomous starships that didn't need humans aboard. The easiest way to avoid that was to ban truly autonomous AI.

There is a long way to go for truly sentient AI. As it turns out, a major step was announced just yesterday: https://arstechnica.com/science/2017/10 ... -security/ .

JohnRoth wrote:The reason is fairly clear: he wanted a universe that he could model on "Age of Sail." That means he didn't want a universe where there were autonomous starships that didn't need humans aboard. The easiest way to avoid that was to ban truly autonomous AI.

There is a long way to go for truly sentient AI. As it turns out, a major step was announced just yesterday: https://arstechnica.com/science/2017/10 ... -security/ .

Nice post, although personally I don't think man will ever obtain a truly sentient A.I. That'll be too much akin to Adam biting the apple again. We are not gods.

It must be said that the potential of current pattern recognition prowess is direly limited by current computing power, i.e., much more capable algorithms could be written in traditional algorithmic form, now, if the computing power was there. Not as efficient as they probably could be, but that is besides the point.

Any programmer worth his weight in silicon has written such algorithms and know the lay of the land. Incredible amounts of information is needed to be juggled to truly accomplish it now, in a strongly polynomial time-limited fashion that just isn't possible with current computing power.

JohnRoth wrote:The reason is fairly clear: he wanted a universe that he could model on "Age of Sail." That means he didn't want a universe where there were autonomous starships that didn't need humans aboard. The easiest way to avoid that was to ban truly autonomous AI.

There is a long way to go for truly sentient AI. As it turns out, a major step was announced just yesterday: https://arstechnica.com/science/2017/10 ... -security/ .

Very interesting article, and computer vision is one of the key areas of research that we have a LONG!! way to go in before we can make classic robots of the Terminator/Data type. Current systems are totally different from biological vision and can be confused by things that would never confuse a human. See https://www.theverge.com/2017/4/12/1527 ... telligence

There is also a possibly apocrophal story from the early days of Neural net research about the military training an early neural net to identify tanks in photos it was fed. All was going well until they fed it a new set of photos and it failed dismally because it turned out it was focussing on sky brightness not on the presence or absence of a tank. This is a mistake that would never be made even by a young child and shows that the way neural nets work in the lab is very different from how the neural net in the brain works. probably somthing to do with all those eveolutionary tweaks to brain processing our visual cortexes have evolved.

Silverwall wrote:

JohnRoth wrote:The reason is fairly clear: he wanted a universe that he could model on "Age of Sail." That means he didn't want a universe where there were autonomous starships that didn't need humans aboard. The easiest way to avoid that was to ban truly autonomous AI.

There is a long way to go for truly sentient AI. As it turns out, a major step was announced just yesterday: https://arstechnica.com/science/2017/10 ... -security/ .

Very interesting article, and computer vision is one of the key areas of research that we have a LONG!! way to go in before we can make classic robots of the Terminator/Data type. Current systems are totally different from biological vision and can be confused by things that would never confuse a human. See https://www.theverge.com/2017/4/12/1527 ... telligence

There is also a possibly apocrophal story from the early days of Neural net research about the military training an early neural net to identify tanks in photos it was fed. All was going well until they fed it a new set of photos and it failed dismally because it turned out it was focussing on sky brightness not on the presence or absence of a tank. This is a mistake that would never be made even by a young child and shows that the way neural nets work in the lab is very different from how the neural net in the brain works. probably somthing to do with all those eveolutionary tweaks to brain processing our visual cortexes have evolved.

Yep. The latest phase of "artificial intelligence" made "great strides" by ignoring the way the brain works and simply throwing lots of computing power and training data at it. Quite a few people have been pointing out that this is a dead end, but results count.

Going back to basics and getting good results with less computing power and radically less training data ought to shake up "the way it should be done."

The trouble is, visual and auditory processing in the primate brain is the best understood part of the brain. A lot of the rest is understood as "if you see these areas light up together, the subject is probably doing X." Maybe. If the researcher isn't misreading insufficient data from a dead fish. (Not a joke - someone once put a dead fish in an fMRI machine and got readings from its brain. The problems in analysis pipelines have since been corrected. Probably.)

At least this research is on the right track.

NervousEnergy wrote:
I think there's a good chance we'll see some general AI much sooner here in the real world than 'hundreds of years', as the latest systems today are approaching the computational power of mammalian brains. You don't need to 'understand' the mind at all to build an AI... you just need to build something sophisticated enough that it can learn. I do agree that 'programming' an explicit AI will likely never happen... more likely the march of ever smarter general purpose AI till the day when you're hard pressed to tell the difference.

The biggest issue isn't just raw computing power but rather parallelism and multithreading and interconnectedness. This is where the tradional chip sets suck wholesale but where Neural nets are better and even a small mammalian brain blows them out of the water.

There have been mentions of remotes and robotic farm equipment. Homes probably have Honorverse equivalents of Roomba's and robotic lawnmowers and harvesting equipment

Iirc, there are remotes used for damage control on ships. But as they are always in background exactly what this consists of is unclear.

kzt wrote:Iirc, there are remotes used for damage control on ships. But as they are always in background exactly what this consists of is unclear.

I always interpreted remotes as being more like the remote control bomb disposal robots of real life. Camera equipped servo powered macines that can go into environments not suitable for organics. However the computational power is of such systems is probably limited to the smooth operation of mobility functions and sensor enhancements but with no programatic decision making functions.

Silverwall wrote:

kzt wrote:Iirc, there are remotes used for damage control on ships. But as they are always in background exactly what this consists of is unclear.

I always interpreted remotes as being more like the remote control bomb disposal robots of real life. Camera equipped servo powered macines that can go into environments not suitable for organics. However the computational power is of such systems is probably limited to the smooth operation of mobility functions and sensor enhancements but with no programatic decision making functions.

I think RFC has deliberately put robots and AI in the background.
A lot of it would take away from the story. Right now a lot of surgery is being handled by robots.Two millenia from now, there should be a lot more of that.

We know the TAC computers have to be advanced. We read about programs that are put in...no way for human reflexes to move fast enough when trying to stop waves of missiles.

And astrographers use computers. And I'm sure there's a lot more. It would make sense that robots be used for a lot of battle repair damage as well.

I mean we can see robots as waiters in restaurants now. People are using expert sessions a couple of miles from my house to grow exotic plants...and that means at least some things close to AI if not there.

People for centuries seem to have traveled the stars while in sleep beds. Who ran the ships if not robots?

But robot marines are boring...no wisecracks, no practical jokes. In theory you could have very small crews or none at all. But there are limits. Even LACs have crews of ten. Practical for long cruises but they're generally near either a CLAC or base.
Having five instead of ten people would casualties in half, however, in battles.

But this is RFC's universe and he makes the rules.