What's Your Astrogator Good For?

[jfwking]

DOn't know about the rest of you but my highly skilled navigator/astrogator is the result of years of training and experience. He can do 4 dimension math with scraps of paper and a pencil. He can plot jumps to within a fraction of the normal error levels.

Mostly though he fleeces the passangers at the regular evening card games. Bugger can count cards and run the odds like no one else I ever met

 

[mark_rotteveel]

An amazing landing is one where the passengers are willing to get back in the plane with you next time.

A skill my party are still mastering, Andrew....

Mostly though he fleeces the passangers at the regular evening card games. Bugger can count cards and run the odds like no one else I ever met

Captain Jonah, I reckon you just hit the nail on the head, mate.

"Ladies and Gentleman, we have a winner"

 

[kaladorn]

How do you hold a computer legally liable? Is it conscious of its 'charge of care' with which the astrogator is normally placed (the care and safety of the passengers)? Does it realize *it* has its hide on the line if things go wrong?

Usually, the answer is you can't hold it liable and it has no real concept of charge of care or its own mortality. An astrogator does.

Also, a computer will (and they have) quite happily execute any valid instruction. Valid means 'recognized as part of the language of the computer and featuring valid memory or i/o space locations'. Valid does not mean 'safe for the crew or passengers'.

There once was an airbus that decided it wanted to fly North. The autopilot started it heading out over part of Europe, away from its destination. The pilots attempted to disengage the system but could not. They could not reorient the plane. The system was a voting system. Pilot had one, autopilot had two votes. Guess who won?

They frantically called the engineers and the engineers said they thought this condition might maybe be cleared by a reboot of the system. They advised having everything else shut down when this was done and doing it slowly. The TCA pointed out that the situation existed *on a plane in flight* and *how could this be done on a plane in flight?*. The engineers said "Oh...." and "Oh... My....". Then they said "You might be able to boot it up quickly if you did this. It might come up. maybe." Having no other choice, they did the reboot in flight and it worked.... but no one knew that it was even possible, but the other option was to fly until the plane ran out of fuel....

People are aboard starships as a redudant backup and to handle the circumstances that the Navicomp, etc were not setup to handle. Computers respond well to things within their programming. Outside that programming, oh my.....

But complex programs of today (and getting moreso) make it easier and easier to miss a 'boundary case'. Maybe one or two programs of any size have ever been totally vetted, and I think NASA did one that cost a stooooopid amount of money. And every mod would expose a whole new possibility of side effects, etc. So it is just possible that some odd combination of factors could cause the Navicomp or Flight Control subsystems to do something really obviously (to a human) undesireable.

This is where your human comes in. He says 'no, I don't think we'll fly into the sun thanks!'. He can quickly make a judgement of what is reasonable, unlike the computer - it just knows what is 'valid'.

 

[Hecateus]

Kaladorn,

how does this reasoning compare with the AI on Deep Space One probe? > Linky < and other similar AIs?

 

[Morte]

Here's how I envisage the role of the computer in flying, extracted from my story hour:

{pretty harmless spoilers for "The Kursis Charter" here}

Alone on the bridge, Silea wanted to scream through her forced calm as she juggled tasks and seconds. The Avarice Rewarded hove up 525 meters behind Malfeasant, 17 meters below track, 85 meters left, and closing at 22 meters per second. About 10% off the numbers she wanted. Fine. Now for the hard part.

“Final manoeuvre. Closing by phased array radar. I have radar visibility through the gas. Malfeasant is not tumbling. She has extensive hull modifications… some sort of stabilising fins. There is either extensive modification or severe damage to the engine area. Closing… Closing… Coming side on… Waist airlock looks normal. Trying for dock.”

She handed over to the software. She’d given it every advantage, now it was over to tolerances and luck. Wire frame projections sprang to life on the screen before her. She clenched everything she had as they wobbled and resumed their glide to alignment.

So the pilot made decisions about what the flight computer was to do. The T/Pilot skill check for that difficult manoeuvre was rolled earlier, as a (hidden) roll to decide whether she could accurately predict whether the computer could dock the ship. She didn't make any skill checks related to manually flying the ship. Pilot is very much an INT based skill in this case (it can also be DEX in T20).

Now if somebody were flying a basic grav vehicle through twisting canyons at high speed in gusting winds, that might be different.

As for astogators, my astrogator is the owner aboard, steward and "face man" for the party. IMTU a merchant ship that wishes to carry passengers for a fee must have a qualified astrogator. This particular noble/scout is qualified to astrogate.

 

[Andrew Boulton]

People are aboard starships as a redudant backup and to handle the circumstances that the Navicomp, etc were not setup to handle. Computers respond well to things within their programming. Outside that programming, oh my.....

True, but since the Vilani have had jump drives for *ten thousand years* you'd think they'd have the bugs figured out by now...

 

[sandman]

Originally posted by Andrew Boulton:

True, but since the Vilani have had jump drives for *ten thousand years* you'd think they'd have the bugs figured out by now...

But isn`t it Traditionnal to have the computer to have Buffer Overflow on certain pre-determined condition?

 

[kaladorn]

Originally posted by Hecateus:
Kaladorn,

how does this reasoning compare with the AI on Deep Space One probe? > Linky < and other similar AIs?

Two points:

1. Traveller doesn't do really good AI until *very* high tech levels. Reality is likely to show this progression is too slow (ie we'll have good AI at TL 11).

2. AI is interesting. Most AI is done using neural networks. I had a friend do his Phd in computer science on this stuff. I remember his spending months to 'train' his network for optical shape recognition. I also remember it had to go through N generations to get to the point where it was mostly reasonable in performance. I recall him saying that if you altered the information the net was trained with, you altered the net structure. If you altered the order of the information, that too could affect the net structure. So, a *lot* depended on the training. Somewhat like conventional programming, if you write some code, you test it, but you might miss a boundary condition, in the case of a neural network, you might fail to provide it with training input that gave it the ability to 'reasonably' solve a problem. Yes, a neural net appears to have some decision making capability and a network of neural nets even moreso. But a lot depends on training. And if confronted with something outside of its experience, it can still make mistakes, just like a human. And the other interesting thing about this is that if you futz up your training, you have to go back and start again (or at least step back to the last 'image'). You don't always seem to know if the result is going to work until after you've trained it. So you can't always see where a problem was introduced either. True intuitive leaps still seem beyond current systems. Humans can make such leaps of non-associative reasoning.

Neural nets and other AI technologies offer great potential. Given enough time, you can usually develop an expert system for a given task. It takes time, training, and often it is very difficult to modify a glitched system without retraining it (unlike structural code, which you can, with luck and design, fix more easily). But, given time and effort, you can produce good neural nets that allow flexible responses within a certain range of degrees of freedom.

However, when you pass beyond that range, bizarre behaviours can still result. Just like when you enter an untested boundary case in a structured piece of code.

This is where your human can contribute. He may even just 'check' the computer 99% of the time. But in that 1% case where the computer decides it makes sense to dive into the nearby moon, the pilot can say 'no, I don't think so'.

 

[kaladorn]

Originally posted by Andrew Boulton:
</font><blockquote>quote:</font><hr /> People are aboard starships as a redudant backup and to handle the circumstances that the Navicomp, etc were not setup to handle. Computers respond well to things within their programming. Outside that programming, oh my.....

True, but since the Vilani have had jump drives for *ten thousand years* you'd think they'd have the bugs figured out by now... </font>[/QUOTE]On the other hand, the Vilani probably don't hit every boundary condition either.

"We've never needed to test that before. It wouldn't be traditional."

Seriously, this comment would be valid if you didn't have the same problem we have as software engineers in the modern world - people *constantly* want code to do more, do it faster, do it differently, do it while juggling some plates, a chainsaw and a loaf of whole wheat, etc.

Due to constant modifications to existing programs to accomodate new hardware, new features in the interface, new interconnections to other systems, the jump drive being repaired with a new flux plenum from Frambulon IV instead of an original equipment manufacturer part from Gfrovalbin II, etc., you'll always have some untested or (I hate to say, as a software engineer) some *untestable* cases that will eventually cause some grief. In some cases, it isn't even that you don't know about a possible error condition, it is just that you have no feasible way to generate it, so you put in a catch routine that you *think* will handle the situation, but since you can't test it....

Maybe in YTU, things are nice and static and people only ask for a new release of ship software every 100 years and everyone uses OEM parts within spec and all the OEMs comply with standards exactly and all the standards are well enough written so that the software works in all cases exactly as it should, but that (to me) doesn't map to the real world. IMTU, things happen. Sometimes they are oddball events like an astrogator jumping inside 50D while his ship is under attack.... and other times they are simply avoidable things like the captain not doing the maintenance soon enough... and thus things don't work as they should.

And this is where the rapidly adaptable and multi-purpose human comes in.

Plus someone has to be able to put in new computer boards when the old ones fry....

 

[mark_rotteveel]

Maybe it's a bit like in the old Buck Rogers series. Remember how they had this just-like-a-bought-one, you bewt navigation software, but they all still got shot down, until the humans took over?

Is it sorta like that?