More Notes on Procedurality research

[Versão em português ainda não disponível para este post.]

WIRED published an article on procedural generation (procgen), as part of it’s fourth annual trends report “The WIRED World in 2016”. In this post I’ll comment on some parts of the article which I believe to be common misconceptions about the field of procgen.1

Read another post on procedurality here.

ProcGen ≠ Hand-crafted

First, the suggestion that procgen can in some way substitute hand-crafted content is misleading. There is no magic – an algorithm can never generate the same kind of thing a human artist could create, since it can’t be truly creative (RIP Lady Lovelace). There are things that procgen can do that can’t be done manually, and vice-versa – these are both valid, but different approaches.

There are some specific cases in which procgen may help in some manual creative work, but these are exceptions. The fact is that games like Minecraft or Terraria will never generate a world with the same hand-crafted feel of the cities in Grand Theft Auto, for example; and GTA will never invite exploration in the same way that those games do. It’s apples and procedural oranges.

So, ProcGen is “Cheap”?

This brings us to the cost argument, which says that procgen allows developers to spend less making their games. Although partly true, this idea greatly misses the point. Procgen doesn’t apply to all types of games (or all types of game elements/aspects). Also, even if (when?) a programmer (or team) is able to create an algorithm capable of generating a world with the same hand-crafted feel as the one in Skyrim, for example, this will probably require either (a) a long time or (b) a lot of money, which would invalidate the argument about procgen being cheaper in the first place. Nothing comes “for free”.

Procedural Generation, not Creation

Regardless of how much procgen a game uses, the algorithms are (have to be) designed and/or written by humans. Nothing can ever be 100% created by a machine. The difference in procgen is simply that it’s metacreation (creation of a system that “creates”, or generates, stuff).

None of the games mentioned in the article are “developed by algorithms” – they use procgen in certain elements and content, but that’s it. When procgen is used to “develop” a game, the result is something more like the generic outputs from Ian Bogost’s Game-o-Matic.

There is a reason it’s called procedural generation (procgen), and not procedural creation. In a general sense, it’s misleading to say an algorithm “created” (or “developed”) something, since this implies human intention (which is why I insist, for example, that Deep Blue does not “play chess”).23

ProcGen is not Random

The WIRED article also contains the common confusion of associating procgen to randomness, which is also missing the point. Randomness is not the basis, neither the most important element in procgen – algorithms are.

It’s true that there will always be the matter of equilibrium between randomness and creativity – when is something intentionally expressive or meaningful, and when is it just randomness? However, this is not what is being referenced when discussing procgen (or generative art in general).

When a game designer decides that the rules for positioning trees is leading to a “random” look, what is happening is not necessarily a conflict between randomness and design. It can be a much more complex situation, involving a series of factors such as the human perception of what “looks random”, the artist’s intentions, and of course randomness itself.4

If a painter decides to use fewer strokes, for example, this does not necessarily mean that there is “less art” being made. The same goes for an artist working with code. The decision about which elements or aspects will be generated by the algorithm, which will be randomized and which will be fixed (or static), above anything else, are creative choices.


The WIRED article mentioned here is mostly fine, as are most of the articles on this topic. For example, despite it’s title, eventually the article does say that the algorithm creates “the worlds”, which is more correct. Also, it does suggest the metacreative nature of procgen in saying that “developers won’t be sculpting, but playing God.” I enjoyed the opposition made between procgen and the linear and fixed nature of film as well.

My intention here is just to point out specific elements or aspects that I believe to be confusing or misleading about the topic.5

Hopefully one of the good things that will come from the popularization of procgen is that the media will stop thinking that it’s “astonishing” that a game like No Man’s Sky has [Huge Number] planets, and begins focusing on the fact that what matters is not the quantity of content, but it’s quality (in the sense of what makes it original or specific to that designer/artist). Any programmer can write a code that generates [Humongous Number] “things” – but few designers are able to make these “things” be interesting and original.

  1. This post is supposed to be a quick reply, so I didn’t do much in terms of formatting, adding links or references. If you have any corrections or comments, please contact me..
  2. In my own research, I use the term “Procedural Expression”, but I make it clear that the expression comes from the artist, and that procedurality is the language being used.
  3. Another recent example is this TED Talk in which the speaker asks if a computer is able to “write poetry”, suggesting some sort of intention. The talk presents several examples of computer generated poetry and asks the audience to differentiate among them, but fails to mention the fundamental differences that exist in the process behind each one of them. A code that generates poetry based on existing text is very different than one that does the same without the source material (they are both valid approaches – I just mean that this information is relevant to the exercise proposed in the talk).
  4. I read somewhere that originally the characters in Will Wright’s The Sims had a more complex AI, however it was later simplified because in testing players felt that their behaviour felt… random.
  5. Other parts of the artice that I think are problematic are: the (cool) cartoon used to illustrate the piece is misleading, since it suggests that procgen is similar to branching plot systems, when these are actually totally different approaches; and the closing quote, which I enjoyed, but also suggests that the idea applies to procgen in general, when it actually only makes sense for No Man’s Sky (or similar games).
See more theblog.

No comments »

No comments.

RSS Feed for these comments. Trackbacks

Leave a comment