Larian Studios - Swen on his obsession with Ultima VII

shamroxor said:

I'm looking forward to what tricks developers will use when advances in graphics aren't enough to grab the headlines.

If they can get the program to take care of how many loaves of bread are made in the local bakery, how many npcs desire loaves that morning, if any npcs are likely to riot if denied loaves, how many guards are around to deal with the riot etc then track the inventories of the dead bodies after the riot, the families of the dead now hungry cos no loaves

The possibilities are endless if all that reactivity can be offloaded to a 'virtual developer' working within a bunch of rules set up by the human developers.

It's crazy developments like that I'm looking forward to!

Click to expand...

That didn't answer my question.

Perhaps you are mixing machine learning up with general artificial intelligence?

Perhaps if it is his ultimate wish, that someday he and his studio create a product that approaches the greatness of Ultima seven, then this all makes a lot more sense to me. I'm a big fan of setting a decent goal, then doing your best to get there or, perhaps with some extra effort, actually exceed even your own expectations.

Morrandir said:

That didn't answer my question.

Perhaps you are mixing machine learning up with general artificial intelligence?

Click to expand...

Well, I guess, whatever technique which takes the nightmare of marsharlling all those elements to work in a consistent way and not descend into chaos once the player does something the programmer didn't take into account for.
The sheer number of permutations is the real killer in building a 'living breathing world' automate that stuff and the humans can spend more time filling it with the cool stuff.

Its been years since I played U7. I remember it with fondness though as I do with all the Ultima games. There are a lot of elements that I would like to see implemented in an RPG turn based game. Some of them have been done in MMORPGs but not in regular RPGs. Others are conceptual that I have not seen yet - like a damage shield (similar to druids) for clerics that damage undead only. But I am an idea man.... not an implementation man.

Lue.

Hmm there was a funded kickstarter called Corven - Path of Redemption.

Based on some of these topics and threads some of you might enjoy it.

Corven is a story-driven, open world, single player RPG inspired by our favorite old-school RPGs, especially Richard Garriott's Ultima series. We miss Ultima and its design philosophy and so we set out to create something very similar. Corven will let you explore an interactive open world full of secrets and surprises. Maybe even more important than that, Corven will deliver an engaging story with many interesting NPCs. Our goal is to make you want to know "what happens next?" in the main quest. Or in side quests for example"why is the, married, mayor of this town sneaking over to this other woman's house almost every night?"

Click to expand...

shamroxor said:

Well, I guess, whatever technique which takes the nightmare of marsharlling all those elements to work in a consistent way and not descend into chaos once the player does something the programmer didn't take into account for.

Click to expand...

Well, as a rule of a thumb you can say that machine learning requires a lot of data. For example when you want your machine to learn how to identify pictures with traffic lights you need a lot of (> thousands) pictures with (and without) traffic lights. When the pictures (or parts of them) are labeld with "has traffic lights" and "has no traffic lights" we have so called "supervised" learning. (That should sound familiar.)

Artificial intelligence on the other hand can be everything. For example just simple rules written by humans. So simple scripted NPC routines can be considered as being artificial intelligence. The term unfortunatly isn't well defined.

The sheer number of permutations is the real killer in building a 'living breathing world' automate that stuff and the humans can spend more time filling it with the cool stuff.

Click to expand...

That's a problem that's typically solved by good testing. Here automation is possible in way that every permutation is tried and then it's checked whether the (game) world is in some strange undefined state.
But it's not sorcery. As said it's just more work.

Sad to see posts about "rose tinted glasses" and "u7 did a lot of things but did them badly." It's like mocking early films because they were black and white with no sound or scrutinizing early cars because you had to crank the engine and they weren't durable or very reliable.

I agree that playing U7 today requires that a great deal of annoyances be ignored. I particularly dislike inventory management. But these kinds of critiques miss the point that the early Ultima series peaking at U7 did things nobody else was doing and it struck many gamers, me included, as very exciting and fun.

If Swen ever makes the "rpg of all rpgs" in the spirit of U7, my hope isn't that he repeats poor execution, my hope is that he takes the concepts and evolves them to 2020 production values. Few games have evolved or cared to evolve what U7 did so long ago.

Morrandir said:

As said it's just more work.

Click to expand...

That's kinda the reason Swen gave during Original Sins development when day/night npc cycles were dropped, that's why I find these advances in AI/Learning so exciting; its getting close to removing that excuse.

shamroxor said:

That's kinda the reason Swen gave during Original Sins development when day/night npc cycles were dropped, that's why I find these advances in AI/Learning so exciting; its getting close to removing that excuse.

Click to expand...

Wouldn't that dramatically increase the time spent in validation? First, I don't see how the day/night cycle would benefit from AI, it's really not necessary, scheduling is pretty basic as I said. It would be fun to make it very complex with a free scheduling, if that's what you mean, but it would be very CPU-intensive, and not predictable enough, that's why developers would prefer to keep NPCs as simple state machines.

IMHO it would also make it unplayable, because the player would be completely lost.

Secondly, that's not something you can use to make the validation easier, so I don't see how that would "remove that excuse". On the contrary, the "more work" comes from complexity.

Some AI could be added so that the NPC can reach specific goals. Pathfinding is a good example, but it could go beyond that, such as giving basic fetch quests to a companion.

Redglyph said:

Wouldn't that dramatically increase the time spent in validation?

Click to expand...

Wouldn't it be nice if there was an artificial 'director' pulling those strings allowing the development team to feed the raw information items, goals, win/lose states safe in the knowledge this 'director' will make those elements play nice with each other in a logical way? Something which plays the role of a team of testers?

I'm not talking about tech now which left to its own devices makes goofy scenarios play out and requires human intervention to knock in to shape or remove completely.

As much as I'd love Ultima 7 level 'Life' in modern games, the variables are so daunting even for teams in the hundreds, I don't blame developers for keeping it simple.

Redglyph said:

Some AI could be added so that the NPC can reach specific goals. Pathfinding is a good example, but it could go beyond that, such as giving basic fetch quests to a companion.

Click to expand...

Those specific goals can conflict with each other snowballing in to the loaf apocalypse I mentioned earlier! That's always fun but would grind your game to a halt.

I'd like an artificial director knowing there's only 10 loaves in the bakery so no more than 10 npcs are hungry in that 8 hours, if I steal/buy 2 loaves, 2 npcs suddenly aren't hungry anymore instead going to work thus removing conflict.

Many scenarios could be playing out in parallel overseen by a silent artificial observer in the glorious gaming future but again I dread thinking about the work required for humans trying to tie those elements together in a fluid logical manner.

*Out of curiousity I wonder if Larian programmers use Exult for Ultima VII or DOSBOX.

While I agree that Ultima VII is an aspirational game that tested the limits of simulationism in an open world, I actually think Ultima V broke the most ground in that department minus the graphics side of things.

Larian are already on the road to using AI more btw. They have an AI player test the game to find bugs and to free up programming time for game development tasks. I would like to see them develop an AI set designer (i.e. layout a dungeon with props etc) which designers can then tweak for better gameplay. The beauty of that is you can refine the algorithm to generate better and better results.

shamroxor said:

I'd like an artificial director knowing there's only 10 loaves in the bakery so no more than 10 npcs are hungry in that 8 hours, if I steal/buy 2 loaves, 2 npcs suddenly aren't hungry anymore instead going to work thus removing conflict.

Click to expand...

Yeah, it would be nice - but think about it: What sounds easier to you: Creating a generalized machine-learning based AI and feeding it enough structured input data about life, the universe, and everything else to figure out even the simplest of rules (such as, NPCs need food to survive) - or simply hardcoding the rule that you formulated above yourself?

Creating and training a *biological* intelligence (which comes with a lot more neurons and has a world of input data to draw from) to become a game developer capable of doing what you describe above takes something like 20-30 years, after all.

In general, if the rules by which a system works are known, it's a LOT more efficient to simply hardcode them than to develop and train an ML-based AI. That's why, up to today, rule-based AIs, aka the highly scripted NPC routines of Ultima VII or Gothic II, feel a lot more convincing to me than any of the "radiant" AIs or "emergent" NPC behaviours that have appeared since then.

Is this abomination still a thing?

Atrachasis said:

Creating and training a *biological* intelligence (which comes with a lot more neurons and has a world of input data to draw from) to become a game developer capable of doing what you describe above takes something like 20-30 years, after all.

Click to expand...

… I'm talking about a 'Havok' type, off the shelf solution. 'OK virtual director dude, this world area our game team has provided has npcs who are passive/hostile, require food, 4 hrs work, 8 hrs sleep 4 hrs wandering around town - make them play nice'

Ultimate 7 and Gothic have great worlds so why isn't that degree of 'life' present in more modern games? It's hard work!

I think Ultima 6 might have been my favorite - that was the one with the gargoyle race that everyone thought was evil, and then you found out they were just misunderstood etc... right? I *think* the combat system was still turn based in that one?

Damn. I should check out some Ultima titles on GoG. I had a lot of fun with those games. Ultima 5 was good, too, with the Shadowlords and all; but Ultima 6 was when I really fell in love with the series as I recall.

I think I might have been one of the very, very few people who actually didn't like 7 as much as I liked 6.

FurtiveNyctophile said:

I think I might have been one of the very, very few people who actually didn't like 7 as much as I liked 6.

Click to expand...

Ultima 7 gets a lot of criticism for its combat when compared to 6, its just that everything else was so easy to get in to in comparison to the older titles. 6 has plenty of fans
Its like Mass Effect 2 streamlining some rpg elements which annoyed a bunch of fans but ending up eclipsing the first one's success. I still prefer the first one!

shamroxor said:

Wouldn't it be nice if there was an artificial 'director' pulling those strings allowing the development team to feed the raw information items, goals, win/lose states safe in the knowledge this 'director' will make those elements play nice with each other in a logical way? Something which plays the role of a team of testers?

Click to expand...

I'm not sure I understand what you mean or what you'd like to see, to be honest. When you develop, you start by defining the validation tests, and you make sure they pass, even after further modifications. You can only give this task to a human, it's too high-level.

First I thought you meant to use AI in the implementation, to allow for more complex environment or at least make sure the environment auto-regulates. Now you mention AI for the validation process? Or both? I'm a bit lost

If something is meant to auto-regulate, you have to validate it by throwing all possible external inputs. In the case of a game, that's the player actions, and we know that's the most unpredictable of all. Validating so many unpredictable inputs with a game that changes its rules based on the past sounds like a nightmare.

shamroxor said:

Those specific goals can conflict with each other snowballing in to the loaf apocalypse I mentioned earlier! That's always fun but would grind your game to a halt.

Click to expand...

The same happens with AI, you need conflict resolution. Check for instance planning algorithms like Graphplan, they have heuristics to make their paths and reach a goal, and check that the sub-goals don't conflict with the final goal.

Mutual goals will often conflict by they nature, and that's even the base for games against adversaries. Your example with loaves makes me think of simcity-like games, it's resource management, do you mean to let the user manage that, or have a self-managing economy in the game? If the latter, you simply have to provide the resolution with simple economics equations, for example with offer/demand prices, or by increasing the production of loaves. It's more efficient than recreating the whole process of economics in an AI (I say again - users won't have the resources to run such a program in real-time anyway).

See that from a user experience point of view, too. What game are you talking about? Is it an RPG? If so, is there really any added value to put a big resource regulation loop in the game? The players are interested in a good story, they don't want to spend their time searching for a loaf so they wouldn't starve, it's too low-level to captivate their motivation.

Redglyph said:

…You can only give this task to a human, it's too high-level.
..AI for the validation process? Or both? I'm a bit lost

Click to expand...

Dude, i'm talking about cool stuff which expedite the creation of virtual playground like world, hence the 'wouldn't it be nice' I don't see the solution here now, i see it as an avenue for future rpgs once graphics cease to be such a selling point!

Redglyph said:

… The players are interested in a good story, they don't want to spend their time searching for a loaf so they wouldn't starve, it's too low-level to captivate their motivation.

Click to expand...

One of the most common memories, the thing which blows people's minds from ultima 7 is grab flour mix water place on oven wait couple seconds, pop! Bread!
There's plenty of gamers who marvel at 'boring stuff' like that.

One thing to add to discussion about machine learning for NPC behaviour in RPGs: the problem is that's it's difficult to describe what the desired outcome is. But a metric for this is needed.
A related problem is "NPC" behavior in shooters or any other game where you have an actual competition. E.g. you can let the AI play against itself many million times. Then by learning which actions lead to a desired bahaviour (=win) you get bots who are good… against other bots, and perhaps good against humans.
But this method (reinforcement learning) can not be applied to NPCs in RPGs. You just lack the reward function.
Also it's questionable if you really want to have bots who "play really well" and win against you or if you actually want bots who play like humans.