Reasonable assumptions about THS

[sir_pudding]

One thing about total surveillance is that it's inevitably a very high noise-to-signal ratio. Even with NAIs combing over every piece of data you'll need to set your tolerances pretty high if you don't want to be constantly flagged with false alarms. This means that criminals don't need to hide from surveillance as much as they just need to fool the NAIs that are analyzing the data. If you can appear innocuous, you'll vanish in the noise.

[vicky_molokh]

Originally Posted by Jürgen Hubert

I don't think this is a handwave. Computational problems that were glitchy and error-prone two decades ago have become trivial today, and even though Moore's Law has broken down in the setting, 90 years of computational and software advances are still a lot. And modern software does have plenty of error-correcting elements, or else nothing would work today. How much better are they going to be in the future when there is self-aware software able to correct its own programming when something doesn't work?

Given that the standard procedure recommended by IT support fellas when encountering an error is 'reboot and try again, not sure what that was' (compare to simple calculators, which basically either worked or didn't work at all), it seems like the number of unfixed errors is rising. At first this would seem to imply a decreasing errors:productivity ratio, but the problem is that an error or two is often enough to 'taint' a big chunk of a program, sometimes to the point of crashing the whole OS. Sure, builds like the GNU/Linux seem more robust, but even they aren't perfect.

[Xplo]

Originally Posted by Jürgen Hubert

modern software does have plenty of error-correcting elements, or else nothing would work today.

That's.. not really true at all.

Modern software does tend to have a lot of data validation, sanity checks, and error-catching elements, so that anything really weird will (hopefully) generate a warning or error message instead of causing the software to lock up, exit, overwrite memory, or what-have-you.. but it doesn't have error-correcting capability. That's all down to bug stomping, which is a notoriously frustrating process with diminishing returns and no guarantees.

How much better are they going to be in the future when there is self-aware software able to correct its own programming when something doesn't work?

You mean when the self-aware software incorrectly identifies the problem and applies the wrong fix? Especially if it fixes the part that does the fixing and creates a feedback loop of cascading errors?

I'm perfectly willing to accept that 90 years of computer science and programming advances will create systems that are more robust than modern systems with a similar level of complexity, and I'm willing to accept that they can create fantastic things like self-aware AIs that pretty much work the way they're supposed to pretty much all the time - but to suggest that everything works perfectly just because it's The Future.. well, that's just science fiction. ;)

[lachimba]

Originally Posted by Xplo

Of course, can all your video sources be trusted, or have some of them been modified (possibly in realtime) to add, change, or delete crucial details? Ah, intrigue.

Thats one idea for when it comes up. Plus also maybe some of the data is just simulated based on old data or errors or maybe key details are missing like what the suspect was doing after the camera went off for 5 mins.

[jason taylor]

Originally Posted by lachimba

Is it a reasonable assumption to assume in THS that intelligence and law enforcement agencies have banks of 10s of 100s of shells running NAI to monitor video feeds etc for persons of interest and sepecific events monitored by LAIs and then at a higher level by an SAI or human?

So more or less they have a high chance (50%+) of picking up anything/anyone that isn't well hidden or off the system so long as they were actually looking for it in the first place.


Also is it also reasonable to assume that when Law enforcement needs to reconstruct a scene they might use gathered video feed from ordinary citizens, cameras, other sensors etc to construct accurate VR simulations of that time period over a specific area? Such that they can walk in the simulation and shadow someone after the fact so to speak?

Actually analyzing the information gathered is impractical today. Computers don't have the intuition to analyze on any other basis then number crunching which has it's limits even with the most sophisticated computers. Humans have a large capacity for getting bored, which means, among other things, that the data most closely analyzed will be the data seen when a given analyst gets on his shift first(when I proofread business letters for my father for instance, I am most sure I am right about the first page). Presumably professional analysts are better trained, but there is more information to process. Humans also have a limited ability in the quantity of information each human can process, so that each given analyst can only see the information he is working on, and not the information another analyst is working on.

In a hypothetical transhuman world in which a human mind can have the flexibility and subtlety of a human and the speed and relentlessness of a computer things might be different. But then technology for concealment might be as well.

[lachimba]

Originally Posted by jason taylor

Actually analyzing the information gathered is impractical today. Computers don't have the intuition to analyze on any other basis then number crunching which has it's limits even with the most sophisticated computers. Humans have a large capacity for getting bored, which means, among other things, that the data most closely analyzed will be the data seen when a given analyst gets on his shift first(when I proofread business letters for my father for instance, I am most sure I am right about the first page). P

But an NAI should be perfect for just watching feed.

The slave mentality and other disadvantages help keep it focussed.

And looking for a particular face or just 'weapons' isn't analysis. It is a simple instruction with success based on a per roll.

[jason taylor]

Originally Posted by lachimba

But an NAI should be perfect for just watching feed.

The slave mentality and other disadvantages help keep it focussed.

And looking for a particular face or just 'weapons' isn't analysis. It is a simple instruction with success based on a per roll.

Weapons is easy enough for a computer to handle. To recognize faces they need to be already in the database. Which is fine as those are presumably the ones most wanted.

The problem is simply processing all the tons of data that can come in through a camera.

[jason taylor]

Originally Posted by lachimba

Is it a reasonable assumption to assume in THS that intelligence and law enforcement agencies have banks of 10s of 100s of shells running NAI to monitor video feeds etc for persons of interest and sepecific events monitored by LAIs and then at a higher level by an SAI or human?

So more or less they have a high chance (50%+) of picking up anything/anyone that isn't well hidden or off the system so long as they were actually looking for it in the first place.


Also is it also reasonable to assume that when Law enforcement needs to reconstruct a scene they might use gathered video feed from ordinary citizens, cameras, other sensors etc to construct accurate VR simulations of that time period over a specific area? Such that they can walk in the simulation and shadow someone after the fact so to speak?

Now that I think about it, videos on street corners is not necessarily as worrisome as we might think; it is simply the state claiming the right to observe people walking in public territory, a right any individual citizen can claim. What is more big brotherish is the possibility of snooping on private communications and electronic devices.

[Phil Masters]

Originally Posted by lachimba

Is it a reasonable assumption to assume in THS that intelligence and law enforcement agencies have banks of 10s of 100s of shells running NAI to monitor video feeds etc for persons of interest and sepecific events monitored by LAIs and then at a higher level by an SAI or human?

As others have said, this is more or less reasonable in itself (defining "shells" as mostly mainframes for this purpose, of course.) The snags are, (a) that NAIs lack any capacity for original or creative thought - they can look for "weapons", or "criminal behaviour" on certain simple definitions, or specific people, but telling them to look for "trouble" or "clues" in general will get nowhere (and LAIs are only a little bit better), and (b) cameras are so dirt cheap and ubiquitous that deciding which feeds to watch could be a serious challenge.

And (c), many governments have to contend with a hundred years' worth of privacy and civil rights legislation created in a world of cheap CCD cameras (and monitoring LAIs with honesty programming), so they can't necessarily get away with everything.

Originally Posted by lachimba

So more or less they have a high chance (50%+) of picking up anything/anyone that isn't well hidden or off the system so long as they were actually looking for it in the first place.

Subject to the above caveats...

In my experience GM'ing TS, the first thing that a bunch of PCs will do when they discover that something significant may have happened in a somewhat public place is look to see whose external security cameras seem to overlook that place, then go sweet-talk the owners. It helps if the PCs are government employees, of course, but it's not required. "There's been a break-in at the warehouse next to yours, and we're employed by the owner - feel free to check with her if you like. Could our AI skim through your security camera records for last night?" If the space in question is public, it's hard to see why the answer should be No.

In my Europe on Mars campaign, the fun is that the PCs are E.U. agents in Port Lowell, so they tend to plot the areas which they want to check, then plot which adjacent businesses are European-owned. It's not that they have police powers of any sort, but "Could the consulate request a favour?" from somebody with Diplomacy-15 tends to accomplish the desired result.

(The other thing those PCs do is look on MarsTube and research peculiar hobbyists, like the ones who collect sightings of interesting cybershell designs. Why bother asking anyone about that bar brawl when three customers uploaded imagery to MarsWeb within thirty seconds of the fight starting? Why go to the effort of tracking that heavy construction shell when the bot-spotter sites do your sifting for you?)

The fact is, somewhat ubiquitous surveillance is one of the components of the setting. You don't have to assume a panopticon society to think that most urban public spaces are, de facto, observed.

(Reasons for running a Mars-based campaign #1; there's still some wilderness.)

Note, however, that such records are almost entirely useless as evidence in a court of law. The flipside of a century plus of digital camera and AI development is a hundred years of video editing and digital animation development. The only digital records which a court would trust are those which have been held on encrypted storage with all sorts of certification.

Originally Posted by lachimba

Also is it also reasonable to assume that when Law enforcement needs to reconstruct a scene they might use gathered video feed from ordinary citizens, cameras, other sensors etc to construct accurate VR simulations of that time period over a specific area? Such that they can walk in the simulation and shadow someone after the fact so to speak?

As necessary, and that's a roll against various skills... Tracking and acquiring all the relevant records can take time, effort, and legal clout, note.

[Xplo]

As a kind of postscript...

My job is not entirely dissimilar to the surveillance scheme. I enforce editorial policy for a well-known SEM (internet advertising) business. How this works, generally, is that when new ads or keywords are entered into the system, they're examined by some algorithms and compared to some lists. Potential issues are flagged for human review. There is also a second scheme, running parallel to this, where humans are assigned to examine ads and sites associated with certain words that are likely to lead to policy violations and that are frequently used in searches.

When a human finds a violation, depending on the scope of the violation and how clearly it violates policy, the violation will either be handled by that human directly, or escalated up to higher-level agents who handle the tricky questions and grant lower-level agents permission to take action on large-scope violations.

Most violations are necessarily escalated due to the guidelines on who can take action on what.

The system works, sort of, but it's terribly complex, inefficient, and not a little dysfunctional.

Previous posters have noted that NAIs will only be able to pick out the really obvious signs, such as the presence of weapons, which I think is probably accurate. If you really want human-like surveillance, I think you're going to need a lot of SAIs looking at the data to determine which things are suspicious and which aren't - which will be subject to interpretation, of course - and then more SAIs (smarter ones, theoretically) to make high-level decisions about what to report or take action on based on guidelines that do not cover all contingencies and that may be vague or contradictory.

That's putting aside the possibility of technical problems that I raised before.

In short: yes, it can be done, but it won't result in anything resembling omniscience. At best, you'll get what you would get if you had a lot of unobtrusive beat cops watching people.