OOTS #1172 - The Discussion Thread

[danielxcutter]

We talked about stuff like this with the Dispel Magic, Antimagic Field and "giant elemental falls from the sky" theories.
The spell could be cast, but odds are it wouldn't work on account of the spell being interrupted by the caster being completely stoned before they can finish.

I'm sorry, giant elemental does what?

[GregTD]

{Scrubbed} I've given you my sources, you've given me personal anecdote.

Good bye.

Grey Wolf

{Scrubbed}

[chiefwaha]

You’ve usually had 15 or 16 or whatever the legal age in your country is of living - which includes risk assessment
If you’re 15 and can’t cross the road safely there’s probably something wrong

There is a significant difference between crossing the street, something that you can take as much time as you need to make a safe decision, and driving a two ton machine, where a millisecond between decisions could literally be the difference between safely avoiding a collision and the deaths of all involved.

[mjasghar]

There is a significant difference between crossing the street, something that you can take as much time as you need to make a safe decision, and driving a two ton machine, where a millisecond between decisions could literally be the difference between safely avoiding a collision and the deaths of all involved.

Strange how billions are able to do it daily

[Sir_Galliant]

I just realized something: when the hammer returns, it might shatter Durkon's statue! Then what?

[Kantaki]

I just realized something: when the hammer returns, it might shatter Durkon's statue! Then what?

The cleric needs a raise.

[Cazero]

Strange how billions are able to do it daily

Same with walking. Doesn't make it easy, or even simple.

[Dion]

Hey guys, here’s something I bet nobody has thought about: what’s going to happen when the hammer comes back?

I bet Rich is trying to figure out what happens next right now. He’s probably rolling dice to decide!

[Schroeswald]

Hey guys, here’s something I bet nobody has thought about: what’s going to happen when the hammer comes back?

I bet Rich is trying to figure out what happens next right now. He’s probably rolling dice to decide!

It depends on how Durkon rolled in aim.

[Resileaf]

I just realized something: when the hammer returns, it might shatter Durkon's statue! Then what?

It's fine, people who are turned to stone aren't alive and can't die even if their statue is shattered. At least until stone to flesh is cast. If the statue is repaired beforehand, Durkon will be fine.

[Kantaki]

Hey guys, here’s something I bet nobody has thought about: what’s going to happen when the hammer comes back?

Payday!

I bet Rich is trying to figure out what happens next right now. He’s probably rolling dice to decide!

You're thinking of Wildbow.
The Giant doesn't roll dice.
He draws lots.

[Fish]

You're comparing "things the best computer engineers in the world know" with "things the average guy in his Toyota Corolla knows". Those are in two different worlds.

You’re the one who said “humans aren’t good at it,” or words to that effect, without clarifying that what you meant was an average driver with a few hours’ training.* Not to mention that the average driver would make a worse gold standard than experts would does not validate your point.

I think what you’re trying to say is “the best system can be made better than the average driver,” but what you seemed to be asserting was a computer would be “slightly better than a human,” which I read to mean “any given human.”

*A curious definition of the relevant skill set — restricted only to driver training. Computers must not only trained to drive, but also to understand basic things like “weather” and “speed” and “identifying and reading a road sign” and “what a human looks like, and not to kill any.”

Right now, the best intelligence in your car is you. ... On the contrary, the best intelligence in a self driving car was created by someone who studied hundreds if not thousands of hours of traffic analysis. Are you saying that the latter can never be more intelligent than the former? That's absurd.

I think you’ll find that was you. Just now.

No, what I said was (and you quoted this part, so your strawman here is most puzzling) “even if computers were better [at making decisions, which was the context of my statement], we would have no way to measure it.” It would be like asking a toddler to grade his older sister’s calculus homework. It might be right ... but the toddler has no way to prove it. (I’ll set aside for now that you shifted the goalpost to “more intelligent than” from “always makes the best decision.”)

To recap, you seemed to assert that “a computer is better than humans” and “humans don’t know how make the correct decisions 100% of the time.” I was pointing out that the first assertion is something, by definition, that cannot be known, because we cannot measure what you pretend to possess knowledge of.

So, no, a computer can't know more than the human who created it. However, a computer can know more than the average human...

If you had made this claim clearly, instead of the over-broad implication from before, I would have had no issue with it.

But you have changed “humans” to “the average human” and “make the right decision 100% of the time” to “know more.” I think you can see those aren’t the same claims. If this new version is your actual thesis, I agree — with the caveat that knowing is a vague word that would have to be defined. A computer can store and recall billions of pages of poetry, but we’re still trying to teach it how to “know” what any of it means.

[GreatWyrmGold]

Spoiler: Car Talk

Show

...most of which has been spent discussing the realities and promises of AI driving (again: mea culpa, mea maxima culpa)

At worst, you set down the first part of the powder trail. It's not your culpa any more than it's mine or any other individual's. Arguably, my verbosity makes me more culpable than you.

Why would your car pay to sit in a lot when it can instead get paid to drive people around rideshare-style?

We might need to review the definition of "your car".

Yes, that is a problem when designing a decision system from a machine learning algorithm: you have to feed the computer a gold standard to learn from. If humans can only make the right decision 66% of the time, which you can read as “the gold standard designers couldn’t agree on what was Right in 34% of scenarios,” the computer won’t be any better than humans at it — and even if the computer was better, we have no way to measure it.

Actually, we do. We can tell good human driving from bad human driving by whether or not we crash, and (in the likely event that we don't) if we get to our destination on time.
...I didn't realize that this needed to be said, but we're not teaching cars to be better than human drivers by teaching them to copy human drivers. We probably started by programming basic rules, then used machine learning and controlled test tracks to get them good enough we were comfortable putting them on uncontrolled roadways, and built up our database from there.

Personally, I'm having difficulty believing Boston didn't have taxis before Uber. Or that taxis in Boston didn't congregate near stadiums when they were about to finish the game. I've never been there, but once you've seen taxis congregating at airport exits, you know that "it's an easy place to pick up fares" is not a new invention.

Grey Wolf

Yeah, but ye olde taxis aren't part of a new wave of technology that you know you don't grasp the limits of and are hence prone to blaming bad things on.

That said, a driverless vehicle would be susceptible to other kinds of things that a human would never fall for — like catching it at a stop light and standing in front of it, so its safety systems prevent it from driving while your buddies loot it.

I'm kinda fuzzy what stops this from happening now. Have one buddy run in front of the truck (or drive a car there, if you're worried about the driver's ethics) and have another guy with a hammer and a knife smash the driver's window and threaten him with a pointy thing while the rest of the gang takes what they can carry/stuff into available vehicles. I'd imagine it's something that could be carried over to autotrucks, like a subtle button to hit that calls the police or locks on the cargo bit.

Sure, but you don’t have to build a new network. I believe you’ll find that nationwide networks of gas stations for truckers already exist, minus the automation. If the business model supports it, the gas station owners will add the automation.

Or heck, they can just hire a few gas station attendants to fill in the missing links. Sure, it'll add some extra employees to stations outside New Jersey and Oregon (tying this back with a tangent from earlier in the thread), but it'll be cheaper for the shipping companies to pay the stations to do so (directly or via higher fees) than for them to pay all their truck drivers. Minimum-wage gas-pumpers are cheaper than truck drivers, and you probably wouldn't need as many of them.

You're comparing "things the best computer engineers in the world know" with "things the average guy in his Toyota Corolla knows".

More precisely, "things every computer engineer working on cars has been able to teach every car they've worked on". Okay, Google and Tesla might not share data, but either company alone has enough brilliant engineers and raw data to make an excellent driver, and the quality of all their autopilots goes up with anything one car learns. That alone guarantees that they'll be better than drivers with only one lifetime of driving maybe a few hours a day, let alone teenagers who just got their first car.

Strange how billions are able to do it daily

First off, a quick Google search says barely over one billion people drive regularly. Second off, almost 3,300 people die due to failing to drive properly, and something like 53-131,000 more are injured. Daily. Sure, the odds are something like 10,000 to one (per day), if we only count injuries and deaths (as opposed to property damage), but those are still odds we can decrease. Odds we should aim to decrease by any means necessary, because those statistics translate to tens of millions of tragedies (of varying scale) per annum.

I think what you’re trying to say is “the best system can be made better than the average driver,” but what you seemed to be asserting was a computer would be “slightly better than a human,” which I read to mean “any given human.”

Why? He didn't say that a computer would be better than all humans.

*A curious definition of the relevant skill set — restricted only to driver training. Computers must not only trained to drive, but also to understand basic things like “weather” and “speed” and “identifying and reading a road sign” and “what a human looks like, and not to kill any.”

I'd put that under "driver training". And also under "things that computers are good at". In reverse order:
1. You know that camera software that figures out where faces are and tries to get them in focus despite the operator's worst efforts? It's not that big of a leap to apply that same logic to finding whole people.
2. You know those captchas which have you identifying road signs or horses or whatever? What do you think those are for? Combine that with basic image-to-text software designed to handle human handwriting, and a computer can handle signs designed to be clearly legible.
3. It's called a speedometer, Fish.
4. Weather.com. (More seriously, sensors to figure out "Hey, the traction's not so good right now" are present in manually-driven cars.)

No, what I said was (and you quoted this part, so your strawman here is most puzzling) “even if computers were better [at making decisions, which was the context of my statement], we would have no way to measure it.” It would be like asking a toddler to grade his older sister’s calculus homework. It might be right ... but the toddler has no way to prove it. (I’ll set aside for now that you shifted the goalpost to “more intelligent than” from “always makes the best decision.”)

Again, we know the computer is doing good if it doesn't crash. Sure, a car doesn't crash often, but when it has a large enough sample set, it can tell what responses are more likely to end in negative outcomes.
Also, your argument relies on saying that since the toddler can't figure out how good of a driver her big sister is, the toddler should be driving.

But you have changed “humans” to “the average human” and “make the right decision 100% of the time” to “know more.”

He never said that computers would always make the right decision; if he did, he wouldn't have compared them to humans. And what kind of fool assumes "humans" to always mean the best experts? After all, most humans are about average, and half are worse.

Spoiler: Ardor of the Tick

Show

If they don’t know you’re watching it isn’t interference.

And it's so in-character for the honorbound demigod of an honorbound race to secretly observe proceedings which he knows would be unfairly altered if he openly observed them.

Though it'd be a more democratic system if the clan members could observe the voting of their chiefs.

No one really knows how the game is played, the art of the trade, how the sausage gets made. They just assume that it happens, but no one else is in the room where it happens.

ETA: Hel...is clearly not an infallible source of information about what the other gods are doing.

In other news, Belkar is not a very nice person.
Seriously, if Hel knew half of what she claims to know, or if half of the absolutes she claimed were close to true, this arc would be over (and/or wouldn't happen because the world would be hilariously dysfunctional if things were that absolute). Hyperbole is one of her defining character traits. Why do people keep accepting her absolute statements as literal?

The cleric needs a raise.

I'll say, he's been working his butt off this arc!

[Fyraltari]

Anyone dwarf can just walk into the voting chamber, so they probably can. However, remember that they are not the dwarven parliament. These meetings are usually a waste of everyone’s time.

Grey Wolf

Not that I disagree with your general point, but I don't believe that the bodyguards and heavily-armored dwarves in th eMiddle chamber would let just anybody in.

[KorvinStarmast]

Same one 3.5 uses: If you can't catch it or have moved since you threw it, it falls to the ground in the square it was thrown from.

That fits the edition most closely aligned with the comic, thank you for digging that up.

You're not just paying someone to drive the truck
{snip}
I don't see much difference. Which means if your proposal was economically viable, we'd be seeing a lot more goods going by train

Amen on your first point, but on your last: a lot of stuff does go by train.

Maybe, then again, this whole book has been propelled by one principle: panel 3.

Yes, yes it has. (Except for Haley's talking the mechanic into a better price and Haley bluffing Crystal into not fighting Haley and walking (eventually) into the lava pit.

isnt there technically some wiggle room in his wording? http://www.giantitp.com/comics/oots1016.html

Yes.

[Mightymosy]

You are grossly overestimating the capabilities of single-system control. You've just described a far more complex version of the global flight management system. There are about 100,000 flights per day across the entire world. Los Angeles alone, it seems, has four times that many cars per day in a single one of its roads. The global flight system is run by some of the best, most reliable computer systems we've ever built, and is still heavily dependent on human beings doing the heavy lifting. And unlike cars, it can avoid vehicle intersection in three dimensions, with far greater margins of error you find in any road.

The idea you could control all cars in a city with a single computer, with little enough delay in communication to make safe decisions (where decisions need to be made and applied in fractions of a second) is at this point sci-fi.

ETA: To illustrate the issue, what exactly do you think this computer would do if a child in a bike swerved into traffic? How would the computer 1) identify the problem, 2) calculate the rerouting for every car in the vicinity, avoiding them crashing into other already-calculated paths and 3) transmit the corrections to all vehicles that might be involved in the approximately second and a half it takes for that kid to be run over?

AI cars deal with this by each carrying a computer of their own, processing it's own visual data in real time, and driving defensively when it is called for it, and aggressively when it must. Centralizing the problem would require a computer that would dwarf google's entire operation, and it'd still be incapable of dealing with the sheer raw data from billions of vehicles all needing second-to-second decisions.

Grey Wolf

Well, it doesn't have to be a single computer - that part was over-simplification. The important part is that the cars are controlled centrally, and not individually.

That being said, what you describe is something like finding the perfect solution for a traffic net in some sort of mathematical perfection.

Which I think is not necessary.

I was not looking for creating perfect routes for every car. We don't have that now, do we? And still traffic works.

Traffic works even though a lot of people use Google Maps ONLY to find their destination. I know some people who wouldn't be able to navigate to the nearest city without help from a computer, that's how well-educated humans have become....
Such people participate in traffic, and it still WORKS. Lousy, sometimes, but it does. You don't think computers could do at least as well?

You don't NEED to solve the ENTIRE TRAFFIC at once, with one desktop PC.
I imagine an actual dystem would divide the roadmap into small sectors, with one computer (or one program, or one virtual pc or what have yoiu -one calculating entity is whatI mean) each controlling one such sector. When a car leaves a sector and enters the adjacent one, the computer that is responsible for the next sector takes over (just as flight control Germany takes over Austrian planes when they fly over Bodensee or how ever that works exactly).

Imagine, just as an idea, that each road (crossing to crossing) and each crossing or each roundabout traffic itself is controlled individually by one computer (computing power of your choice, as per current hardware capabilities).

Also add that you have lines on each road so that cars can follow them (google "line follower robots" for an example of what I am trying to get at).

With that scenario, you DO NOT get OPTIMISED traffic, in the sense that each way is PERFECTLY calculated for minimum traffic time of all cars.

But WE DO NOT HAVE THIS NOW EITHER.

This scenario would mean that people would probably get similar travel times as right now (for all people who use Google already). Maybe even LESS, because less traffic due to people who need to drive back and forth because they don't know how to find their destination.



So, with that scenario I described, do you not think current tech could handle it?

Because I think it can.




What does the computer do when a child runs in front of one of the cars?
Well, what does a human do?
BRAKE
Only the car quite probably has better sensors, faster reaction time AND STRICTLY adhered to a tempo limit that allowed the child to survive.

[Fyraltari]

Yes, yes it has. (Except for Haley's talking the mechanic into a better price and Haley bluffing Crystal into not fighting Haley and walking (eventually) into the lava pit.

These had nothing to do with rule-lawyering, though.

[KorvinStarmast]

These had nothing to do with rule-lawyering, though.

Yes, that was kinda my point. It was the exception to the rules lawyering thing as a theme, used to tie up an old plot thread/conflict.

[Fyraltari]

Yes, that was kinda my point. It was the exception to the rules lawyering thing as a theme, used to tie up an old plot thread/conflict.

Ah, I see. But rule-lawyering isn't a theme, per se, I'd say "adherence to the rules and what it means" is what is being explored here, with rume-lawyering the plot device to allow it in the whole Godsmoot arc.

In a way, I could argue that Bozzok died because he ignored a rule (or rather a tradition): not giving golem autonomy, without considering why it was in use.

EDIT: and Crystal died because of how she always let somebody lead her, rather than thinking for herself.

[GreatWyrmGold]

Yes, that was kinda my point. It was the exception to the rules lawyering thing as a theme, used to tie up an old plot thread/conflict.

Disregarding whether or not rules-lawyering per se is a theme, the fact that one plot point is tied up without it doesn't negate it being a theme. See, most works of fiction longer than short stories have multiple themes, and they're not all given equal weight in all scenes.
Besides, themes aren't like physical or mathematical proofs; you don't disprove it by offering one counter-example. They're more like statistics; if enough parts of the work support X theme and few if any detract from it, it's a theme.

[Grey_Wolf_c]

I imagine an actual dystem would divide the roadmap into small sectors, with one computer (or one program, or one virtual pc or what have yoiu -one calculating entity is whatI mean) each controlling one such sector. When a car leaves a sector and enters the adjacent one, the computer that is responsible for the next sector takes over (just as flight control Germany takes over Austrian planes when they fly over Bodensee or how ever that works exactly).

What you imagine and what you described are not the same thing. That is not centralised at all. In fact, its so decentralized that, taken to its logical conclusion, it involves just allowing each car to make its own decisions, which is the actual solution self-driving cars are going for. You are literally suggesting that a central computer gives them the route, and the car takes care of the second-to-second decisions. That is what current Ai cars do: they uses google maps to know what turns to take, and the rest is decided by the car.

Also, as I said, the entire global flight control system is dwarfed by the number of vehicles in single cities. The flight control system also depends in large numbers of overworked humans to do the handling over and the like. It is by no means "automated" in the way you think it is. I suggest watching the wendover video on the topic if you want to know more, but if your automated car solution is "like the air traffic control system, but for cars", then you are mistaken about how automated it is.

Imagine, just as an idea, that each road (crossing to crossing) and each crossing or each roundabout traffic itself is controlled individually by one computer (computing power of your choice, as per current hardware capabilities).

I can imagine that. And I can imagine it not working, since there is a blurry line between such crossings that'd be a nightmare to coordinate the handovers. Besides, what exactly would this roundabout controller do? Ask the car where it is going (so it knows what exit to route them to? The car already knows that bit) and then steer the car? But you say that the sensors are in the car - the car is the one in charge of breaking if a child steps into traffic. So the roundabout isn't driving the car either. So it's doing literally nothing that the car isn't capable of doing already, except, it seems, tell the car "you can enter the roundabout now" - something the car can tell for themselves because it can detect that there is a gap in the traffic on their left.

And of course, autonomous cars have the advantage of being able to deal with the interim time when some cars are not AI controlled, whereas the roundabout can't deal with them since it can't predict what they are doing.

Grey Wolf

[Gallowglass]

Imagine, just as an idea, that each road (crossing to crossing) and each crossing or each roundabout traffic itself is controlled individually by one computer (computing power of your choice, as per current hardware capabilities).

....

So, with that scenario I described, do you not think current tech could handle it?

Because I think it can.

So what you have is, when that crossing computer goes down and the redundant backup goes down, all traffic that routes through that crossing is either uncontrolled or stops.

Have you ever been at a blinking red light and watched how terribly all the humans try to take turns the way they are supposed to? Except now the cars will jsut stop because they won't have the ability to do anything until that crossing computer is fixed.

Whereas if each car controls itself autonomously, if one computer goes down, that's a single car that can be programmed to fail gracefully by gliding to a stop. And all the other cars around it responding to its failure with their own avoidance systems.

What does the computer do when a child runs in front of one of the cars?
Well, what does a human do?
BRAKE!

Well, hang on now... how many points is this child worth exactly?

[Kantaki]

Well, hang on now... how many points is this child worth exactly?

Reminds me off that joke about the Mercedes Star and its purpose.

[Squire Doodad]

Well, hang on now... how many points is this child worth exactly?

2 and a half goats.

[Mightymosy]

What you imagine and what you described are not the same thing. That is not centralised at all. In fact, its so decentralized that, taken to its logical conclusion, it involves just allowing each car to make its own decisions, which is the actual solution self-driving cars are going for. You are literally suggesting that a central computer gives them the route, and the car takes care of the second-to-second decisions. That is what current Ai cars do: they uses google maps to know what turns to take, and the rest is decided by the car.

Also, as I said, the entire global flight control system is dwarfed by the number of vehicles in single cities. The flight control system also depends in large numbers of overworked humans to do the handling over and the like. It is by no means "automated" in the way you think it is. I suggest watching the wendover video on the topic if you want to know more, but if your automated car solution is "like the air traffic control system, but for cars", then you are mistaken about how automated it is.


I can imagine that. And I can imagine it not working, since there is a blurry line between such crossings that'd be a nightmare to coordinate the handovers. Besides, what exactly would this roundabout controller do? Ask the car where it is going (so it knows what exit to route them to? The car already knows that bit) and then steer the car? But you say that the sensors are in the car - the car is the one in charge of breaking if a child steps into traffic. So the roundabout isn't driving the car either. So it's doing literally nothing that the car isn't capable of doing already, except, it seems, tell the car "you can enter the roundabout now" - something the car can tell for themselves because it can detect that there is a gap in the traffic on their left.

And of course, autonomous cars have the advantage of being able to deal with the interim time when some cars are not AI controlled, whereas the roundabout can't deal with them since it can't predict what they are doing.

Grey Wolf

Well, no. What I said is that computers control areas of multiple cars at once, just not every car in the world, or even in a city.

Why you say that the logical conclusion is that each car would then make decisions on its own is something I don't understand.
Since my point is that ONE decision making computer for all cars in a small area seems superior to each car making its own decisions, I'd rather have this point discussed than delving into the rest of it.

My example with flight control is to exemplify the idea of making calculations simpler by splitting them into parts. Doesn't have to work like flight control

[Grey_Wolf_c]

Well, no. What I said is that computers control areas of multiple cars at once, just not every car in the world, or even in a city.

No, that's what you said to start with. When I pointed out an issue, you switched to saying it was the car making the decisions, not the controller.

Why you say that the logical conclusion is that each car would then make decisions on its own is something I don't understand.

I say that because when I asked you about who stopped the car from running over the child, you told me it was the car, not the roundabout controller:

What does the computer do when a child runs in front of one of the cars?
Well, what does a human do?
BRAKE
Only the car quite probably has better sensors, faster reaction time AND STRICTLY adhered to a tempo limit that allowed the child to survive.

you have lines on each road so that cars can follow them

---

Since my point is that ONE decision making computer for all cars in a small area seems superior to each car making its own decisions, I'd rather have this point discussed than delving into the rest of it.

Your "point" immediately admitted that it was NOT superior the moment real life moment-to-moment decisions of driving a car popped up. I'm just showing you those scenarios via examples. If all the controller does is pick a route, and the car then makes all the decisions, we call that the satnav. It's not a central decision making at all.

My example with flight control is to exemplify the idea of making calculations simpler by splitting them into parts. Doesn't have to work like flight control

And my counterexample is that your "flight control" example doesn't actually exist even where you think it does. My point, indeed, from the very start is that what you are suggesting isn't practical at all, and that in the end you are describing the current AI system, where the car gets a route from the satnav, and then a computer in the car uses its sensors to make its way there, avoiding other cars , pedestrians, etc.

Grey Wolf

[Rogar Demonblud]

Although with far better sensors than the flight control system uses. Most of those radars are older than the majority of the Playgrounders, to the point that when NYC had to use a Navy Aegis destroyer (1970s tech) during the massive blackout a few years ago, it was a substantial upgrade.

[Grey_Wolf_c]

Although with far better sensors than the flight control system uses. Most of those radars are older than the majority of the Playgrounders, to the point that when NYC had to use a Navy Aegis destroyer (1970s tech) during the massive blackout a few years ago, it was a substantial upgrade.

...

Which honestly, it's another issue for the "centralize the system". I'd imagine that the cost of setting up the sensors everywhere would be substantial, and most locations would simply never upgrade them, so 50 years after they were first deployed, they'd look like air traffic sensors do today.

Grey Wolf