Crime and Punishment not fit for purpose - needs overhauling

[Ian Doncaster]

Boolean logic is simply not enough here, a state machine with a far more refined language is required, also a notion of time and with certain amount of memory.

"State machines with memory" are hardly a high-tech feature requiring the latest hardware.

Oolite's C&P system [1,2] had that sort of thing added to it over a decade ago to allow the reaction of all sorts of ships to the player be far more fine-grained than "are they locally Clean in this system" - a pirate might run because you've built up a rep as a feared bounty hunter, or they might have heard of you as a trader and move to intercept, and different pirates might have heard different rumours. A cop seeing you in a fight with another Clean ship will compare your reputations before deciding whose fault the fight was and allocating bounties accordingly. Keep a low profile, and you might sneak past that mission assassin because they were expecting someone else. The higher your Pilot ranking gets, the more your fame and infamy spreads - but you can always move on to a new cluster of stars to try to outrun it.

(This all runs, like the rest of Oolite if you don't turn the graphics settings up too high, perfectly happily on a computer which can't even start ED Legacy)

What specific benefit are you seeing from having a super-complicated buzzword-heavy solution requiring cutting edge hardware that you can't get with something like that?


[1] It's been years since I last played and I can't now find a page in the documentation which actually describes the whole thing at once. But essentially it's just one short file defining a set of roles a player or NPC could have, a bunch of events elsewhere in the game which adjust the weightings of which roles apply to the player, and a bunch of ship behaviours which say "if you think your target has role X, do this".
[2] I'm not saying ED should go down this specific route. Oolite's crime detection and response system is very different in other respects.

 

[Wily Rites]

"State machines with memory" are hardly a high-tech feature requiring the latest hardware.

State machines with memory are PDA or pushdown automata, the next step before Turing machines in language hierarchy.

Oolite's C&P system [1,2] had that sort of thing added to it over a decade ago to allow the reaction of all sorts of ships to the player be far more fine-grained than "are they locally Clean in this system" - a pirate might run because you've built up a rep as a feared bounty hunter, or they might have heard of you as a trader and move to intercept, and different pirates might have heard different rumours. A cop seeing you in a fight with another Clean ship will compare your reputations before deciding whose fault the fight was and allocating bounties accordingly. Keep a low profile, and you might sneak past that mission assassin because they were expecting someone else. The higher your Pilot ranking gets, the more your fame and infamy spreads - but you can always move on to a new cluster of stars to try to outrun it.

(This all runs, like the rest of Oolite if you don't turn the graphics settings up too high, perfectly happily on a computer which can't even start ED Legacy)

What specific benefit are you seeing from having a super-complicated buzzword-heavy solution requiring cutting edge hardware that you can't get with something like that?


[1] It's been years since I last played and I can't now find a page in the documentation which actually describes the whole thing at once. But essentially it's just one short file defining a set of roles a player or NPC could have, a bunch of events elsewhere in the game which adjust the weightings of which roles apply to the player, and a bunch of ship behaviours which say "if you think your target has role X, do this".
[2] I'm not saying ED should go down this specific route. Oolite's crime detection and response system is very different in other respects.

I've only dipped my toes into Ooelite, but am aware of all the extra functionality that it has, quite an amazing project, I'll perhaps go back to playing it again in the future with a more critical eye, I'd not noticed the in game C&P but was not playing it for that long either.

Of course you don't have to have dedicated hardware to compute state, you don't need floating point arithmetic hardware to use floating point numbers either, but at some point it is worth putting a floating point unit on board.

Dwarf fortress is another with amazing in game richness to its interactions, those fellows were not very good at the 3D side of game programming though, despite all their best efforts; I'm just saying that it is possible is all and that the tech available for assisting in generating it is leagues ahead of what was available to those who have already ventured there; Which is really promising, exciting even.

Buzzwords are not really my thing, but I do try to discuss topics in as legible way as possible, for any reading. AI has been buzzing around as a term since the 70's if not earlier, it was the raison d'être of both the GNU project emacs and I think Lisp languages more generally, that once this kind of usage, which is to say in games (game theory) for assisting in 'tracing' language specific domain problems, once it has been well established, it will be time for the next AI winter.

 

[AlSki]

the domain that seems the most likely to me is, AI in game ...

"AI" is applied excessively to any computer logic that is effectively a calculated line of best fit over many axes. i.e. For a 4K image, if every pixel is mapped to an axis, and each colour is a value on the appropriate axis, then there exists a function where result = coefficient1*value1 + coefficient2*value2 ... + coeffecient8million*value8million where result will be how close it looks to a cat. If you "train" to find out what those 8 million coefficients should be (effectively directed trial and error changes of the coefficients against many cat pics and many non-cat pics so that result improves accross iterations), you end up with an algorithm that was not coded by a human, but instead generated artificially.

The marketing types that seem to think this shows a form of intelligence, those who work in the industry all seem to share a healthy dose of skepticism. The problem is that there is no understanding of "cat" or "pic", only that the values in the array of bytes happen to match the algorithm.

... the use of AI in generating assets

All that really does is speed up the initial design processes. Your designers still need to tailor the assets generated for your game, to avoid someone else generating the same assets again in the future from the same model.

Where AI is useful is in following complex patterns like natural language. e.g. https://www.nvidia.com/en-us/geforce/news/nvidia-ace-for-games-generative-ai-npcs/ but even then you still get most of that effect much more simply with

outcome = random(1,6);

picking pre-recorded lines.

 

[Wily Rites]

Geez AI is not new, as I just said it's a buzz word that has been buzzing since the 70's. I'm drawing upon that, not the new crazy. The reaction to the use of this term due to the current state of affairs is quite simply a reflection of just how limited the English language is for discussing anything of any significance. That there are now models that auto generate language is an obvious advance in this area and clearly not a buzz. Heck one of the founders of the key science behind neural nets was just awarded the Nobel prize for physics.

This ought to be indicative of the importance of the advances and a heads up that it is not just buzz.

 

[Rat Catcher]

Geez AI is not new, as I just said it's a buzz word that has been buzzing since the 70's

In the 70's the concept was known, initially, as "Expert systems" with the aim of responses from the program being indistinguisable than those given by an expert in the programmed field.

So, 50 years on, and is the LLM indistinguishable from a 'human' expert if asked all kinds of questions related to the field of alleged expertise?

 

[Wily Rites]

"AI" is applied excessively to any computer logic that is effectively a calculated line of best fit over many axes. i.e. For a 4K image, if every pixel is mapped to an axis, and each colour is a value on the appropriate axis, then there exists a function where result = coefficient1*value1 + coefficient2*value2 ... + coeffecient8million*value8million where result will be how close it looks to a cat. If you "train" to find out what those 8 million coefficients should be (effectively directed trial and error changes of the coefficients against many cat pics and many non-cat pics so that result improves accross iterations), you end up with an algorithm that was not coded by a human, but instead generated artificially.

The marketing types that seem to think this shows a form of intelligence, those who work in the industry all seem to share a healthy dose of skepticism. The problem is that there is no understanding of "cat" or "pic", only that the values in the array of bytes happen to match the algorithm.


All that really does is speed up the initial design processes. Your designers still need to tailor the assets generated for your game, to avoid someone else generating the same assets again in the future from the same model.

Where AI is useful is in following complex patterns like natural language. e.g. https://www.nvidia.com/en-us/geforce/news/nvidia-ace-for-games-generative-ai-npcs/ but even then you still get most of that effect much more simply with

picking pre-recorded lines.

Personally, it is my belief that like procedural generation, it is awful in the hands of the wrong type of programmer, but extremely elegant in the hands of a competent one, the same stands true for AI and the use of it, that the two will combine beautifully.

 

[AlSki]

The reaction to the use of this term due to the current state of affairs is quite simply a reflection of just how limited the English language is for discussing anything of any significance

I'd disagree. The reaction is due to the overuse of a highly emotive term by people who don't understand what it is being applied to, conflicting with people who have implemented it and discovered its limitations. It's not the English language at fault, just the misuse of the language.

 

[Wily Rites]

In the 70's the concept was known, initially, as "Expert systems" with the aim of responses from the program being indistinguisable than those given by an expert in the programmed field.

So, 50 years on, and is the LLM indistinguishable from a 'human' expert if asked all kinds of questions related to the field of alleged expertise?

There has most certainly been a large step forward, just when the next AI winter will set in remains to be seen.

 

[Wily Rites]

I'd disagree. The reaction is due to the overuse of a highly emotive term by people who don't understand what it is being applied to, conflicting with people who have implemented it and discovered its limitations. It's not the English language at fault, just the misuse of the language.

What exactly is emotive about the the term; If it is emotive to any, then we should ask why that is?
AI in my understanding is an extension of state machines grammar and language, which also stretches out into cosmology in some schools of thought. Rather than being emotive, I find that it fascinating reflective of a system of grammar that is capable of generating a cartography of that very same system that generates emotional responses. Make of that what you will, but please don's suggest that I'm tossing around buzz words, that is a little bit rude, more importantly it is demeaning of the subject.

Addendum: That said, I also imagine that it is a complete nightmare of a domain today, when facing office types who want to exploit new technology without any understanding of it.

 

[Rat Catcher]

There has most certainly been a large step forward, just when the next AI winter will set in remains to be seen.

I agree, there have been massive leaps in progress, yet much more is needed to ever reach that mighty goal set 50 years ago.

Machine learning will continue to improve, I'm certain, but whilst the LLMs are trained by scraping internet sources, their knowledge base will be, let's be nice - and say "interesting"

 

[Wily Rites]

I agree, there have been massive leaps in progress, yet much more is needed to ever reach that mighty goal set 50 years ago.

Machine learning will continue to improve, I'm certain, but whilst the LLMs are trained by scraping internet sources, their knowledge base will be, let's be nice - and say "interesting"

I think one of the key things that they will bring, is a better understanding of the different nuances of human intelligence.

 

[tolm]

I think I would simplify things a little … I believe focusing on the concept of “insurance” that covers our ship rebuys can hit most of these aims.

In the event we lose a ship, we pay 10% of the cost of the ship to “rebuy” it: that’s our “excess” on the policy. The other 90% comes from some mythical space insurance company.

So, if one CMDRs destruction was caused by an unlawful action by an NPC or another CMDR you can bet your asp the insurance company will be going after that party to cough up.

Thus, I would propose the “increased rebuy” to the criminal player is simply the 90% not covered by the excess paid by the destroyed CMDR, which ailment going to refer to as an “insurance levy” for want of a better term right now.

Whether the criminal CMDR actually pays that or lets it grow across multiple kills - converting from a fine to a bounty at some point - is up to that CMDR. Should they be caught / destroyed, the full outstanding “insurance levy” would be added to their mandatory costs and would have to be paid, even if choosing not to rebuy their ship. This mechanic would persist even after notoriety decays to zero.

Additionally, should a CMDR have notoriety their own insurance would become invalid: ain’t no insurance company paying out to a wanted criminal! But it should become valid again as soon as their notoriety decays to zero.

Finally, the insurance company would be offering a fee to anyone who helps recoup their costs. Thus the bounty for destroying a wanted CMDR would include a “finders fee” of 25% of whatever outstanding “insurance levy” that CMDR owes.

I think this covers the key aims of the OP:

• Increased costs due to victim ship value (90% “insurance levy”)
• Increased repercussions (own rebuy becomes 100% whilst notorious)
• Increased rewards for eliminating wanted players (“finders fee”)

 

[Ian Doncaster]

Thus, I would propose the “increased rebuy” to the criminal player is simply the 90% not covered by the excess paid by the destroyed CMDR, which ailment going to refer to as an “insurance levy” for want of a better term right now.

For PvP content, tying the bounty size to the ship cost means:
- shooting beginners in their Freewinders: those things don't even have a rebuy, go nuts! (quickstart ARX ships are fair game too)
- shooting beginners in their C-rated Adders: okay, those have a rebuy, but the total ship cost is about 500,000 credits, you can still practically kill as many as you want
- shooting returning explorers in paper Asps: that 6A fuel scoop is going to add some but you're still only down maybe 25 million credits, and they're down a lot more than that in lost data
- shooting PvP-fit combat pilots in fully A-rated Cutters with maxed armour: that's about a billion credits per kill, stop picking on people who can fight back you griefer
- wearing your own super-sized Cutter down to 1% hull and lurking in a docking bay: 50 million credits to you to land some trader who was going a little too fast with a billion credit loss? Priceless.

(Frontier did experiment at least for a while with making the bounty higher if the killer's ship was significantly more expensive than the victim's - but a small engineered combat ship can outfight a much more expensive big freighter so that's not a great metric either, and it doesn't help for Sidewinder station rams)

Additionally, should a CMDR have notoriety their own insurance would become invalid: ain’t no insurance company paying out to a wanted criminal! But it should become valid again as soon as their notoriety decays to zero.

Notoriety is far too blunt an instrument for anything like that - you can take certain PvE missions at almost any port from almost any faction, the completion of which will involve picking up some notoriety; if the consequence for that is potentially "and if you die in the next six hours, even doing something entirely unrelated, you're potentially down half a billion credits" those missions are going to need a much higher payout than they currently have.

 

[Leadfoot]

AI is the next frontier for gaming hardware, and this is exactly the area that could be massively improved by CPU's having neural nets built in. It is reasonable to ask: 'what will game design focus upon when photorealistic rendering is one of the choices?', the domain that seems the most likely to me is, AI in game and the use of AI in generating assets.

And eventually your ship gets destroyed almost everywhere in every system on every planet when you least expect it to happen. And in the background you can hear something mumbling "Destroy all humans" in Bender's voice.

 

[Screemonster]

Notoriety is far too blunt an instrument for anything like that - you can take certain PvE missions at almost any port from almost any faction, the completion of which will involve picking up some notoriety; if the consequence for that is potentially "and if you die in the next six hours, even doing something entirely unrelated, you're potentially down half a billion credits" those missions are going to need a much higher payout than they currently have.

Personally I think notoriety works best in its capacity as a bounty-multiplier, and it should be jurisdiction-based - maybe superpower? Power territory? Hell, track it separately for minor factions, powers and superpowers and have it so that incurring a point increments all relevant buckets and your "effective" notoriety is just whichever is the highest wherever you are right now.

If we brought bounty timers back it's also what I'd base bounty durations on rather than the raw value of the bounty, but I went into that earlier in the thread.

the tl;dr of my feelings on C&P is that it should be easier (or, in fact, possible at all given the current panopticon system) to get away with crimes, that a bounty should be something you're encouraged to live with and lie low rather than just ditching it at the first interstellar factors you find as soon as your notoriety expires, and that both these things should open the gate for bounties at the lower end to be actually worth a damn. As it is, if I get a fine that I can't pay off because of notoriety, I'll go to a surface settlement and shoot a guy in the face then turn myself in - the 1000cr murder bounty is often less than the 25% fee I'd pay at an IF.

While I'm at it, fines for carrying stolen goods are kinda ridiculously large too, being based on the value of the items - it's extremely easy for a guard scan to become a matter of "this single item in my backpack will get me a 200k fine and turn the guard hostile, orrrrrrrrrrrrrrrrrrr I could just shoot him in the face for half a percent of that while he's still scanning." If there were literally any kind of statute of limitations system baked into the game, some people might choose to eat the fine and hope they can wait it out.

 

[Ian Doncaster]

While I'm at it, fines for carrying stolen goods are kinda ridiculously large too, being based on the value of the items

Which sort of makes sense, because otherwise you'd just sell them and treat the fine as a cost of business.

(Really what would make more sense is to only have a small fine ... but if you dock in that jurisdiction with that fine still active, or let the guard complete the scan, the goods get confiscated. Refusing/resisting the confiscation ups the offence to a bounty and normal violence applies.)

 

[Screemonster]

Which sort of makes sense, because otherwise you'd just sell them and treat the fine as a cost of business.

(Really what would make more sense is to only have a small fine ... but if you dock in that jurisdiction with that fine still active, or let the guard complete the scan, the goods get confiscated. Refusing/resisting the confiscation ups the offence to a bounty and normal violence applies.)

YOUR STOLEN GOODS ARE NOW FORFEIT

-> resist arrest

 

[Rat Catcher]

Additionally, should a CMDR have notoriety their own insurance would become invalid: ain’t no insurance company paying out to a wanted criminal! But it should become valid again as soon as their notoriety decays to zero.

That would be slightly unkind to puritans like me who only commit crimes against NPCs, and always for a good reason... Didn't like their name, colour of their ship, their ship itself, the day had a "Y" in in... All valid...

 

[metatheurgist]

- wearing your own super-sized Cutter down to 1% hull and lurking in a docking bay: 50 million credits to you to land some trader who was going a little too fast with a billion credit loss? Priceless.

Can't say I feel anything for the trader not obeying the laws there. Especially when grinding some PvP Cutter into dust at regulation speed for free would be even more priceless.

 

[Ian Doncaster]

Can't say I feel anything for the trader not obeying the laws there.

The best bit is the trader then gets notoriety (for the kill), shot down by the station (for the bounty), and is then on the hook for the full cost of both ships.

Especially when grinding some PvP Cutter into dust at regulation speed for free would be even more priceless.

The most expensive I think a ship can get is https://s.orbis.zone/qB9K - just over 2.1 billion credits if all the parts are bought from iSola Prospect by a non-Elite commander - which is many things but not a build I'd expect to win many PvP fights.

(Obviously people would only use this tactic in Power-controlled space where they wouldn't have to pay the 5% fee either)