Mücke's Musings on MMORPG Making

"Die Spieler machen das Spiel."

Newest Postings
Table of Contents
I can be contacted as hobold at this domain name.


Newer Postings


#47. Analysis: Exploring the Lands of Diablo II vs Diablo III

That was quite a detour into information theory and the origin of life. However, now I can explain rigorously why some uses of randomness lead to a more engaging game that provides a lasting, long term motivation, while other, very similar uses of randomness cannot deliver the same experience to the players. Please don't let the imposing buzzwords scare you away! I am merely trying to be as thorough and as convincing as I can be.

Recall my earlier comparison between randomized level layouts in the sequel games Diablo II and Diablo III. When you look at game guides written by players, you will notice that for D2 they use phrases like "enter from the south / sweep clockwise / move until you find some landmark, then turn". Such phrases describe rules which exploit patterns in order to find specific places more quickly, despite the randomization. Apparently, there are some subtle underlying truths to be learned about the game world in D2.

Guides for D3, in contrast, talk about "go there / find entrance / proceed to". Such phrases refer to very specific patterns. They imply that either there is only a single path, or that exhaustive search is the best known method of finding something. Either the locations are completely determined ahead of time, or they are fully randomized, without an underlying rule or pattern. In either case, there is no need to describe anything in more detail. There is little or nothing to be learned for the players.

So in Diablo II maps, there is a medium amount of correlation between different features of the area. Finding one landmark reveals some information about other places - players can make informed decisions as they travel around. The contrast are Diablo III maps, where the correlation only assumes the extreme values 1, for things with a fixed placement, or 0, for the fully randomized locations - neither requires or rewards active decision making. That means Diablo II keeps presenting information to the players as they explore the lands of Sanctuary, while Diablo III offers only redundancy and entropy in that very same situation.

I can only speculate whether Diablo II's informative level layout is an artifact of the particular maze generation algorithm used, or if the designers specifically tweaked the procedural generation of content with the explicit goal of putting such correlations in. But the end result is very consistent. In almost all zones of Diablo II, even when different environments require different constraints on the maze generation algorithm, there are these recognizable correlations.

Speaking for myself, Diablo II's level layout was always interesting, and kept rewarding my attention even after years and years of fighting my way through the lands of Sanctuary. Sadly, Diablo III lasted only for a month or two before I had learned everything there was to know about the geography of the game world.


#46. Too Advanced: Information and the Origin of Life

This entry's headline is the title of a book by Bernd-Olaf Küppers. I wouldn't dream of claiming that it is required reading for creators of virtual worlds (so I jokingly categorized these recent postings as "too advanced"). Nevertheless, the author presents a fascinating idea in a very plausible and scientific way: that the processing of information is central to all life.

This starts from the observation that life began when the first molecule managed to reliably make copies of itself (assuming, hypothetically or not, a lack of intervention by a godly creator). From there, the molecules moved on to more complex forms of information processing: mutation and selection, that is: evolution. So eventually there were many different molecules that could be reproduced in a controlled fashion. Not every molecule could literally copy itself, but because the molecules could rely on each other's presence, they could collectively store and process more information. Thus the task of replication could be delegated to a higher level of organization. So life advanced to form the first cells, each capable of dividing itself into two smaller, but identical copies.

Here we transition from chemistry to biology.

This guiding principle of aggregating more and more information, and ever higher levels of organization, can be traced throughout the evolution of life. After simple cells came complex cells. After single-cell organisms came simple multicellular life forms. The simple creatures developed a more complex structure, with specialized organs. Some of these organs were sensors that could collect data about the environment. As those sensory organs became more numerous and more complex, a large subgroup of organisms began to develop specialized organs to process all the perceived information: the animals got nervous systems, and eventually brains.

Some of the animals continued the aggregation process as they lived in groups and cooperated to survive in their respective biotopes. This required transmission of information between individuals. This marks the transition from biology to explicit communication. For the first time, information was converted from one form into another, across different media, if you will. The dance of honey bees is one famous example.

One species was first to develop a fully generic, general-purpose form of signaling: our distant ancestors developed speech. This was a disruptive change, because until this point, all information processing on earth was directly tied to the necessities of survival. Human speech, however, could be used for an unlimited number of non-existential topics as well. It was the beginning of culture and civilization.

Over time, humans invented the written word and later the Gutenberg press. Information processing began to become a little independent of humans, it became partly automated. Electronic computers then automated it to a very high degree. And today, the internet as an unimaginably large collective of autonomous computing devices, has almost fully automated the processing of information. This is another disruptive change: machines begin to make decisions for us, which we can only review after the fact (not necessarily in an evil, big-brother-machine kind of way; many phenomena are simply emerging without us steering them). The processing of information, foundation of all biological life, begins to leave biology behind.

Interesting times. :-)

This is relevant for game design in the following way: all humans, in fact all life forms, are made to collect and process information. Most relevant is everything that helps to survive, to prosper, to multiply. This is why white noise, raw chaos is not informative to us: we cannot learn anything useful from it. Likewise, too much redundancy makes us lose interest as well. We quickly learn what we can, but then we have nothing more to gain.

This is so ingrained in us that we keep following the very same principles even with information that is non-existential. Repetition (i.e. redundancy) is boring, and chaos (i.e. random noise with no discernible pattern) is frustrating. It is the imperative of life that directs our attention to the grey area between, where there is enough redundancy that we can recognize recurring patterns, but not so much that we are quickly done with deciphering what we can see.

Life itself is the reason why entropy and information are different things.


#45. By the Way: Virtual Philosopher for Hire

I recently flirted with the idea of making a living as a seeker of knowledge, i.e. as some sort of scientist or researcher at an academic or educational institution. It took me a while to accept that wish, because for most of my life it sounded too good to be true. Not long ago, it turned out to indeed not be true (i.e. not all that good in practice). It then took me another while to give up this nice dream.

That means I am now seriously looking for other interesting work. The design of virtual worlds is certainly interesting. What we can learn of the human condition by setting people loose in these virtual worlds is even more interesting. If you, dear reader, are in the business of making and running and evolving virtual worlds, and if you think that my babbling here in this blog makes any sense, then feel free to contact me at the e-mail address that is mentioned (in a slightly obscure way) in the header and footer of every page here.

If you happen to be working at particular companies, I might already be a dusty skeleton in the database of your human resources department. But my application will never get past those recruiters, because nothing I ever did for a salary involves game design. Nevertheless, you can find my real name somewhere on this web site in case you want to cross-check.

No matter whether you can or cannot offer a job to a game analyst doing basic research of the foundations of virtual worlds, I'd still be interested to hear from anybody who thinks my musings here are worth reading. Which of my points are or are not convincing? Is there any data publicly available that I could use to verify or falsify some of my hypotheses? Has anybody been thinking along similar lines before and published it? Do you have a particularly interesting question that you want to discuss?

I can offer an inquiring open mind with an academic background in math and computer science. I grew up among curious and gifted people, so I picked up tons of "useless knowledge" as a kid from archeology over music theory to zoology. If you have followed this blog, you will know that my interests cover a fairly diverse, perhaps even chaotic collection of topics. If I have any exceptional strength, then it is to intuitively recognize relevant connections between these unrelated topics, and then to back up my intuition methodically with thorough analysis.

Perhaps I have even been able to clearly communicate my findings ... but that is for you to judge.


#44. Too Advanced: A Mathematical Theory of Communication

The headline of this entry is the title of a famous article by Claude E. Shannon. It laid the foundation of things that are pervasive in computing: checksums, error correcting codes, data compression, encryption. The idea of a bit as a unit of information was introduced here. I will need a few concepts from this theory to explain my claim of players wanting information from their games.

Model of Data

One main concept of Shannon's theory is that there is always a model of the kind of data that is meant to be communicated between a sender and a receiver, and both parties know that model in advance of the communication. In everyday terms this means that two people who are communicating with each other must share quite a few expectations about the topic that they want to talk about. Otherwise, communication will be inefficient or impossible.

This starts with basic things that we never think about, like speaking the same language. But it also covers background knowledge about, for example, third persons that both sender and receiver know well. Or it is implied context, which helps people to distinguish between apple, the fruit, and Apple, the music label, and Apple, the computer vendor. The model can even influence the language itself, as people learn words for new concepts, or even invent their own jargon.

This blog entry is an example of me trying to establish a shared model, by introducing you to concepts and their names, which I will need for some communication later on.


The aforementioned model of data allows to make predictions of possible future data. Or in plain English: the listener's expectations of what the speaker may say enable guessing. Anything that can be accurately guessed in advance, does not really add any information when it is finally spoken. This lack of surprise implies a lack of information. Such bits of data are called redundant.

What redundancy is strictly depends on the model currently being in use. The better a listener's expectations can guess upcoming data, the higher the redundancy will appear to him, for any given body of data. Or, in technical terms: the better a model matches actual data, the better said data can be compressed.

Loosely speaking, high redundancy implies low informational content.


This term originated in physics, as a more formal way of distinguishing between highly ordered states of matter, such as crystals, and very chaotic states, such as gases. Low entropy in that sense means order, and high entropy means chaos.

For our purposes, entropy is what's left when all redundancy has been removed from data (i.e. redundancy is all the orderly, predictable, non-surprising bits). For example, if you are taking notes in a hurry, you will only write down the parts that surprise you, that are new to you, and skip over everything else. You try to preserve the entropy, without wasting effort on the redundant parts.

You might suspect that entropy is a measure of information. That is not wrong, and certainly holds for our computers. But entropy is a technical concept that cares nothing for the meaning of data. Raw chaos, white noise, has high entropy and low informational content at the same time - it is not informative. To return to the example above: if you were taking notes from a speech, and suddenly all that comes from the loudspeaker is static noise, what could you possibly write down?

The mathematical theory sidesteps the issue of meaning by hiding all the messy human interpretation within the model. The model defines what redundancy is, and thus it also defines what entropy is. So in order to finally explain what information is, I am going to have to delve a bit into biology in the next posting. Our background as organisms informs everything we do and everything we think.


#43. Analysis: Randomness in Diablo II and Diablo III - Level Layout

Earlier, I touched on the topic of randomness in game design. Since then, Blizzard released Diablo III. I was surprised that D3 turned out to be a disappointment (to myself, probably not to Blizzard's sales department) after the novelty wore off. I loved Diablo II so much back in the day, how could I be disappointed by that classic game's successor?

I know that I am not alone in this, despite the record breaking sales. I distinctly remember that this type of games, this genre, used to be called "Diablo Clones". Ever since the release of Diablo III, people started to refer to this category of game as "Action RPG". It seems as if Diablo III itself is suddenly unworthy of the name Diablo Clone; it even destroyed the meaning of the name Diablo.

I had a feeling right away that this had to do with how both D2 and D3 used randomness at the heart of their game design. The original creators of the franchise, no longer employed by Blizzard and not involved in the D3 project, went on record with public statements about their own surprise regarding D3's use of randomness.

I think I finally understand where the problem really lies. A few posts down the road - sorry, this is going to be a longer chain of thought - I will present my conclusion that the players are really after information. Since the game designers cannot include an unlimited amount of information in their games, they have been using randomness as a surrogate. Whether this actually works depends critically on how exactly random game elements were designed.

So let's look at how the two games (Diablo II and Diablo III) are using randomness.

Level Layout in Diablo II and Diablo III

First the similarities of D2 and D3: both games are structured into acts, which in turn are subdivided into chapters. This is directly reflected in the structure of the virtual landscape. Each act happens in a visually distinct context, like a specific climate zone, or a particular geological type of landscape. Each chapter takes place in a contiguous area, say, a specific biotope, or around some interesting landmark. Architectural styles, too, as well as distinct populations of monster species further add to the distinctiveness of virtual places as the story progresses.

Each chapter zone usually has one entrance and one exit. Many zones have optional side paths in the form of a dungeon, that has a separate entrance somewhere in the zone. The overall story of one act is a linear sequence of chapters, with the occasional bit of side plot interspersed. Subsequent chapters are typically mapped onto adjacent zones, but the placement of the zones does not have to be linear.

Now the differences: in D2, outdoor zone entrances and exits are positioned randomly. When a player experiences a particular chapter for the first time, he might travel through the respective zone from west to east. On a subsequent play through, though, that same chapter could mean a journey from north to south. In D3, outdoor zone entrances and exits are not randomized. Each chapter of story is coupled with traveling through a specific zone in a specific direction.

In D2, outdoor zones are maze-like in that they can randomly have narrow chokes between areas, or long impassable barriers. A player encountering such an obstacle must decide in which direction to turn. This decision is meaningful, because one side might lead to a dead end, or to an optional dungeon, while the other side would progress the story.

In D3, outdoor zones are open plains, with the occasional story-driven choke point in a fixed location. That plains is randomly dotted with relatively small, compact plots of land that may be an obstacle, a single building, or a dungeon entrance. Turning left or turning right makes no difference, as either path is equally short, and will reliably bypass the obstacle.

Indoor zones in D2 typically are mazes, built from relatively small tiles (much smaller than screen size), so there is a lot of variety and not much repetition. The paths often fork to dead ends, or long detours, so players continually make decisions as they go.

In D3, indoor zones look maze-like at first glance, but they are built from much bigger tiles (at least two by two screens, I guess). Within each tile, forks are not very meaningful, because players quickly memorize the alternative paths. The number of tiles to choose from is rather small, and there are constraints on which tiles have matching edges. The end result is that D3's indoor zones are much more repetitive, and not much decision making is needed when navigating on subsequent play throughs.

In order to explain the significance of decisions, and of repetitiveness, I will have to refer to a bit of mathematical theory next up. Don't worry, there will be no formulas.


Older Entries:
36 - 40
33, 34
(not numbered)
31, 32
27 - 30
24, 25
16 - 23
0 - 15


Newest Postings
Table of Contents
I can be contacted as hobold at this domain name.