Hacker News Reader

Comment by nikanj | Hacker News Reader

Source: "My friend who works w/ games knows someone who's at CDPR"

Rumour has it, management got extremely ambitious with AI. Every actor in the game world was to have a personality, daily routines etc.

The devs tried to explain how absolutely unfeasible that is, but management was convinced it's possible - and would indeed be ground-breaking.

After a large number of years of toiling with the problem, the writing on the wall was finally seen. The NPC AI was scrapped and replaced with crude "we'll just spawn the cops around you so they don't have to have a chase mode"-level hacks, to allow them to ship in 2020.

Now they're stuck in limbo, because the players are still waiting for those marvellous AI miracles that were promised, but they remain absolutely impossible to deliver. CDPR can patch things like the car physics, but people's expectations were thoroughly mismanaged and they'll forever be disappointed

View on HN

That's amusing because a game like Deadly Premonition managed to pull that off in a decently convincing way. We're not talking about hundreds or thousands of NPCs, of course, but everyone had a daily routine that they went about and the various places in town had opening hours. Most of the story would only progress if you were in the right place at the right time.

Not the most polished game but easily the best murder mystery and Twin Peaks style life sim I've ever played.

So if CDPR couldn't pull it off, they must have been asked to do it at an unreasonable scale or the requirements themselves were unreasonable.

A lot of other games (the first one that comes to my mind is Zelda:BotW) also have NPC routines, opening hours etc. but if you look closer it's just a simple O(1) mapping between time ranges and expected state. It gets much more complicated when you actually try to pull off NPCs that may decide to do different things based on their modeled traits, react to in-world events or player actions etc. which I assume is what grandparent was talking about.

Most players just want a decent looking, mostly bug free, interesting Witcher-3-but-it's-Cyberpunk game.

Exactly. I don't expect to be able to follow an NPC and see a life-like routine, just something convincing enough.

LOL, this reminds of an episode of Mythic Quest where the lead engineer (Poppy) imagines "real-time, persistent, globally-reflected environment changes" for the MMO they built, gets told that it's unfeasible by the entire dev team, mows ahead anyway and ends up with a disastrous demo at the end when the system basically collapses on itself.

Funny how reel-life and real-life collide into each other! :D

---

As an aside, after Ted Lasso (which gets all the well-deserved hype), Mythic Quest is a great trope-filled show esp. for anyone who's worked in tech in general (and game dev specifically).

Sounds vaguely inspired by Ultima Online

I’m a developer, though not in game dev. This sounds feasible, what am I missing?

I'm guessing you'd need Discord level tech to propagate changes in a multiplayer environment which is a bit of a distance from "core competencies".

A game today which is struggling to accomplish this is Star Citizen, and they did a video on what they plan to do here: https://youtu.be/TSzUWl4r2rU

Anyways, I think the big challenge compared to real time apps like Google docs is that game simulation is significantly more complex than synchronizing bits of text. One problem is that at a minimum, you have the round trip to the server as latency between “submit command” and “receive simulation result”. And this has to happen many times per second so that it looks realistic. And then on the server, it has to be processing the changes from N connected players M times per second, and then simulate what happens in the universe. And if simulation (M) drops too low, you get desync problems where players are rubber-banding and the AI gets very dumb. (An example is that star citizen appears to have great AI in demos, but in the live server, the servers can’t process changes fast enough for the AI to work well.)

And then, if you want a lot of connected players or a big universe, this just gets more and more complex, and you’ll eventually out-grow the ability of a single server to process changes. At which point you also need to handled data replication among multiple servers. For example, star citizen game servers don’t have capacity for more planets at the moment, either because of memory or cpu constraints. So sharding the universe into multiple servers is a pre-req for the next stage.

So to more directly answer your question, it’s definitely possible to implement. Many games have basic versions working. Minecraft is an example, but it can’t scale to hundreds/thousands of players. New World is another recent example, but even its servers have player caps. No Man’s Sky is sort of an example, but it keeps players spread out so that they rarely encounter each other. Eve Online is also an example, but I think it reduces the amount of nitty-gritty simulation details. (So it might not have a very in-depth physics simulation, or first-person-level details.)

So it’s basically a scaling problem, and even then still has limitations based on the speed of light (for internet latency) which limit how good it can get.

That reminds me of the early work on the internet arguing that the amount of global connectivity wouldn't scale due to power laws. In a way, they were assuming a system like these global world sims. The equivalent of static routing (players that want to play together enter same server) exists, but the "dynamic routing" field is mostly a game of selecting which illusion to present the player (sometimes, it's okay to just stick people on a different instance but make sure their friend group is preferentially on the same one, etc).

I'm not sure you understand. Cyberpunk is a single player game.

Star Citizen and the fictional MMO in Mythic Quest are multiplayer.

I was responding in the context of a grandparent comment, which was talking about features for a fictional MMO which turned out to be infeasible, and then someone asked why they were infeasible

I keep forgetting that Mythic Quest exists. I need to watch that.

> Rumour has it, management got extremely ambitious with AI. Every actor in the game world was to have a personality, daily routines etc.

They received some AI-related grant[1] from the local government and did whatever they could to avoid penalties.

[1] https://www.cdprojekt.com/pl/media/aktualnosci/30-mln-zl-dla... - original source but in Polish

Found the English version of that source[1]. Seems to have had five criteria[2] with AI being one such. Would be interested if you had a link about the penalties you mention regarding GameINN.

[1] https://www.cdprojekt.com/en/media/news/cd-projekt-group-sec...

[2] https://www.pwc.pl/pl/pdf/alerty-innowacje/13-PwC-Alert-GAME...

I have found[1] the government research grant program "Intelligent Development Operational Program for the years 2014-2020" ("Programu Operacyjnego Inteligentny Rozwój 2014- 2020 działanie 1.2") and funding number/code "POIR.01.02.00-00-0105/16" but am Ill-equipped to search surrounding[2] PDFs for penalties.

It was, and still is, my assumption that there are penalties.

Also, this seems to be another grant - "City Creation" is documented separately[1] and with different amounts from "Animation Excellence", "Cinematic Feel" and "Seamless Multiplayer"[3]

[1] https://archiwum.ncbr.gov.pl/fileadmin/gfx/ncbir/userfiles/_...

[2] https://archiwum.ncbr.gov.pl/programy/fundusze-europejskie/p...

[3] https://archiwum.ncbr.gov.pl/fileadmin/gfx/ncbir/userfiles/_...

What's really weird about it to me is that games like Oblivion and Skyrim were doing sandboxy open world AIs with daily routines over a decade ago. That franchise also had more than its share of bugs, but the AI did more or less work.

That CP2077 couldn't even reach that level remains kind of baffling.

was probably that they didn't have time after scrapping everything else and trying to hit deadlines

In Oblivion and Skyrim IIRC, the characters were mostly individually configured by the level designers. That works fine for a world where you have what boils down to a few small towns but I don't think the same approach could be applied to a place the size of Night City.

Bethesda's Elder Scrolls and Fallout series have much less NPCs in their worlds and far fewer densely populated areas compared to a game like Cyberpunk, where there's crowds of people around every corner. If they really did intend for every single NPC actor to be on a TES level schedule I can see where the difficulties come from.

The conceit being that I want to watch some McDonalds employee brush his teeth before heading out to work, or get the chance to talk to some guy about his mortgage at a bar.

I don't know why you'd build a game simulating more depth of interest in background characters than people actually possess, and it's not like occlusion culling is a foreign idea.

In Skyrim you could place poisoned apples in your enemy's chests so that they're going to eat them and die some days later... And it was 2011. Bethesda will need a miracle, because the expectations for TES6 are too high.

Bethesda oversold their AI all the same though. Google for Oblivion and "Radiant AI", pretty big drama back in the day.

> Bethesda oversold their AI all the same though

The big difference was that Radiant AI was fun. You could join the Warriors' guild, jump across the road to the Mages' Guild, use an enrage spell on them, the run out into the street. Instant chaos ensued.

And the largest city in Skyrim has 74 inhabitants. This kind of stuff has price.

Gothic (the game) introduced that concept even 20 years ago for me.

Can go much further back to e.g. the Ultima games, especially Ultima 7. The world was incredibly big and you had a decent approximation to a living world.

Another one from same time frame (1989-1992) was https://en.wikipedia.org/wiki/Virtual_Theatre Lure of the Temptress

I am not sure what you laid out is unfeasible, but perhaps the way they approached it was unfeasible and left them with no time to iterate or pivot.

CDPR doesn't iterate and test their code in the optimal way. Most of the time they just go at it full on and patch things up at the end. This was (un)fortunately for them a project where the complexity rose so high that their approach couldn't handle it (also covid didn't help), and they were left with intertwined systems that were causing severe performance bottlenecks everywhere (think polynomial or even exponential). AI the way they planned it meant spawning hundreds of NPCs, all interacting with one another and the player, with badly designed systems for that interaction (because why spend upfront time on this if you can floor it), and almost no consideration on how that would impact the graphics and physics sides of the game. The whole development of this game is a case study of how to fail at keeping your externalities in check.

That doesn’t make a lot of sense though. In my country first year CS students build game of life versions with thousands of not millions of individual entities all going about their business and interacting with each other in some shitty math based programming language, using math they typically get from Google because it comes from biology.

How can that be a thing of what you say is true?

I would say first, that's the power of exponential complexity. Second, AI in games tends to consist of somewhat complicated goal-based & fuzzy logic behaviors. Third, game AI agents have interactions that take place within a 3d world that has to be modeled accurately. That means lots of expensive computations for things like line of sight, pathfinding, and so on. Finally, given that games are soft real-time and include remarkably expensive 3D rendering, the compute budget available for AI functions I would guess would be somewhere around the order of 2ms per frame (on a 16.7 ms frame, for 60fps).

Though, two through four can usually be handled without issue, provided the devs know what they're doing & architect the engine correctly. (edit: And keep the scope in check, which ultimately is the responsibility of the leadership.) Unfortunately, this appears to be where #1 messed things up for them.

You don't really need to rerun your AI logic every frame though. It's more than enough if you do it every ~second, and can compute it on a separate thread.

That depends on the type of game. The game I am working on would not work with a AI update every second.

are we talking about the same game of life? in conway's, the update logic for a cell depends only on the immediate neighbors of that cell. so the time to update an entire frame is O(m*n), which is not hard for any matrix size that would reasonably fit on a laptop screen.

I am pretty sure that given the constraints (must operate reasonably well on last generation consoles and minimum specs with decent graphics) the problem space is pretty much impossible.

why? SimCity-like games do this already to a degree.

definitely not trivially easy, but impossible!? nah.

SimCity games do not simulate individuals. Even when they do, the indivduals are not very complex.

Cities Skylines seems to, but just like for the vast majority of systemic gameplay, it's going to be an approximation of some kind and that's fine. CDPR needed to find that compromise between full simulation and believable approximation, but it seems like they ran out of time to get it working.

Star Citizen's "Subsumption" and Quanta system is very similar in goal, and if it goes to plan will be one of the most engaging "AI population" simulations to date. It does't individually simulate every person in real time, but instead simulates a set of tasks and demands to be done, all the resource nodes, and an approximate population of simple entities carrying out those tasks, and when you spawn near said tasks, it spins up a "meta NPC" to represent in full 3D space, an agent of one of those simplified population entities.

So for example, you fly to a mine site, the system knows there are currently 40 miners here mining aluminium to take to the factory, it knows at what point they are at with their tasks, and so it generates 40 NPCs to represent those miners. At that point you can interact with them and influence the simulation, say steal their ship and minerals. That updates the simulation but the simulation need not care about the specifics of the NPC. Just that he has no minerals and he is no longer capable of mining.

You fly away with their ship, and all the other NPCs return back to simplified entities in the graph.

Star Citizen is notoriously widely considered to be vaporware; I wouldn't really give it as an example of "achievable with time, budget, and limited resources."

Yeah, probably not the greatest example of a lean operation. But the argument was that it's impossible on current hardware according to GP, which I don't think is true.

In regards to Star Citizen being vapourware, my recommendation would be to take a critical look at their timeline and what state the game is currently in and ask if you could really see it going any quicker. It's a controversial project, but I think calling it vapourware is just the industry being dismissive.

They had to build the team from nothing, now 700 people, and they are building two games not one, both are pretty groundbreaking in features and scope. That considered it is actually moving pretty quickly. They have already delivered a bunch of those groundbreaking technical hurdles to players. Of course the whole thing could implode yet, but it's looking good.

They deliver consistent progress reports and updates to the playable alpha, which has enough content for me and friends to play regularly for fun now. Updates every three months is the schedule, plus any in game content events that get run.

>Every actor in the game world was to have a personality, daily routines etc There's a similar system on the stalker saga and maybe they could have built upon it? Of course making every NPC do it for itself instead of a sort of hive mind (I don't have a better term for it) sounds very difficult but sounds rather weird that they can't do some sort of similar implementation, right?

They can make it work in a barren hellscape with less than a hundred characters, but not in a bustling urban setting as densely populated as Manhattan, I wouldn't call that weird at all.

> Every actor in the game world was to have a personality, daily routines etc.

I tried this on a limited scale in a text-based MUD in the 90s, and even then it quickly became an ungodly nightmare. What sounds good in your head isn't always feasible in reality.

Risking hubris:

With current GPT-3 technology (or even GPT-2, accessible by then), it should be possible to create each agent unique. But if the research started earlier, I am not surprised by the dissonance between marketing and devs.

The release date is not when work on the game starts, it's when work on the game ends.

Reply on news.ycombinator.com