Simulationism and IF (long)
James Mitchelhill
found quite interesting. The following is an attempt to sum up some of the
responses that have emerged and clarify a few of my own thoughts about
deeply simulated IF, what the problems are and how they might
be solved.
1. What is "simulationism"?
All interactive fiction more complex than "Choose-Your-Own-Adventure"
games involves some level of simulation. Various physical properties of
the world are simulated in conventional IF. Almost all IF involves
simulation of "containment" - the idea that one object can be inside
another. Similarly, most systems provide some simulation of time, whether
this is the simple turn-by-turn model or more sophisticated models that
involve hours, minutes and seconds.
Many IF games go beyond the basic level of simulation, but only in
specific domains where a deeper level of simulation is necessary for the
story. A game involving liquids will have a much deeper simulation of
liquids than a game where this is not important. There is no point in
implementing divisible and mixable liquids if this is never used for
anything within the game.
Simulationism is the tendency towards deeper and less abstract simulation
of physical (and possibly emotional) properties of the game world, not for
limited domains that the author has chosen, but as a general framework.
Additionally, the "physics" of the world are likely to interact with each
other leading to unforseen consequences.
2. Why simulationism?
In conventional IF, a player who tries something the author has not
anticipated will either place the game into a broken state or be
disallowed from taking that action. If an author has included an axe which
will eventually be used to cut the rope that ties the heroine to the train
tracks, then a player will try to use it to cut other things. The player
who tries to use the axe to break down a door instead of giving the golden
egg to the guard in exchange for a key will probably be told that
"violence isn't the answer to this one".
While a simulationist system would allow the player to break down the
door, it does not offer any easy solution as to what happens after that
point when the plot the author had envisioned has been short-circuited. It
has been suggested, however, that the way plot is implemented using
conventional systems is a result of the properties of those systems. A
simulationist system would call for different techniques. The use of
puzzles to pace a game, which is a popular technique in conventional IF,
would obviously be inappropriate in a simulationist game.
Simulationism is not a solution to problems facing conventional IF. Such
a system would almost be a different medium. Whether a simulationist
system would be able to tell different stories to contemporary systems is
a matter of speculation, but some feel that this is a question worth
investigating.
3. Problems with simulationism
A number of problems with simulationism have been raised. These generally
fall into three categories: Implementation issues, authoring issues and
philosophical objections to simulationism as a whole. The following is not
an exhaustive list, but contains most of the major points.
3.1 Implementation Issues
3.1 Generating human-like text
Interactive fiction systems already face the problem of generating
human-like text to describe situations arising in games. The list of
objects in a drawer is generated from the underlying world model. The
problem with simulationist IF is that this becomes a magnitude more
complicated. In conventional IF the contents of a drawer might exist in
the world model as a list of objects held as a property of the drawer
object:
> LOOK IN DRAWER
There are three coins, a pencil and a sheet of paper in the drawer.
A simulationist system faces the problem of generating its description
from a less abstract internal representation. The internal representation
would differ depending upon the details of the system. A simulationist
system may not implement containment, but a spatial model.
> LOOK IN DRAWER
A sheet of paper is at the bottom of the drawer. A pencil is half-covered
by the sheet, while two coins rest on top of it. Another coin is at the
back of the drawer.
And this becomes more complicated when other things are happening within
the game model. If the player had set the pencil on fire, the above
description would seem inappropriate as the PC looking in the drawer would
be sure to notice this.
> LOOK IN DRAWER
A sheet of paper is at the bottom of the drawer, charring from the heat of
a burning pencil that is half-covered by it. Two coins rest on top of the
sheet, while another coin is at the back of the drawer.
It is certainly possible to code such a system, but the amount of work
involved for the author is huge. However, if we want a simulationist
system to present information to the player in a way that does not rely on
appending a list of properties to each object mentioned, it is a problem
that must be solved. Ideally, such a system would take hints from the
author as to the language used in its description.
This is not a simple problem and some have argued that it is intractable.
3.1.2 NPCs
It is not immediately clear how NPCs would fit into a simulationist
system. One possibility is that their emotional states are modelled via an
"emotional physics". One attempt at this kind of modelling was Chris
Crawford's Erasmatron, which was widely derided by the IF community on its
initial release. Another attempt, which received a much better reception
was Facade.
How NPCs should be implemented is, however, a problem in conventional IF.
Much of conventional IF has relatively one-dimensional characters, while
IF that is character focused usually has to deal with conversation, which
demands a greater level of simulation. Emily Short's Galatea, which has
become something of a benchmark for deeply implemented NPCs, takes a
simulationist approach to conversation while retaining a very strong level
of authorial control.
3.2 Authoring Issues
3.2.1 Combinatorial explosion
In conventional IF, an author has to deal with only the interactions that
he can conceive of happening. In most cases, if the author has not
specifically included a response to an action, it will not be allowed. In
contrast, simulationist IF would be much more permissive - the author
would have to specifically disallow actions. The problem is that the
author of a simulationist work does not initially know the range of
actions that will be possible in his game and has every action to
consider.
One possible solution to this may be a sufficiently sophisticated IDE,
which would allow the author to prune the list of possible actions to a
relevant and important subset. Whether this is possible and what
"relevant" means in this case is unclear. Another solution may be to
abstract the problem, allowing the author to refer to ranges of actions
rather than (as in conventional IF) to focus on one action at a time,
although this may require generating human-like text.
3.2.2 Defining properties of objects
Conventional IF defines a limited number of properties for objects. Using
the containment model, the author will define a "location" property
defining where the object is in the game world. Simulationist IF, by its
nature, would require many more properties to describe objects and,
depending on the system, may require that objects be split into parts that
are further defined. The amount of work an author would have to put into
making a single object may be prohibitive.
A solution suggested for this has been to provide a sufficiently large
collection of pre-defined objects that the author could then refine to
describe the object they intend. Whether this is a practical solution is
debatable.
3.2.3 Complexity of the library
One consequence of the higher level of simulation of a simulationist
system is that the programming library provided would be much more
complicated than the (already quite complicated) libraries of conventional
systems. This in itself has led to criticisms that such a system would be
much more difficult to customise and that there would be a higher learning
curve than in conventional systems.
This may be inescapable, although a successful simulationist system would
attempt to minimise these problems.
3.3 Philosophical issues
3.3.1 Simulation is uninteresting
The main objection to simulationist IF is that such a system would not be
capable of producing interesting games. This is tightly related to the way
that plot and simulation would interact in such a system. The argument is
that a story arising from the physics of the world model would not be
interesting. Adam Thornton has summed up this viewpoint with the phrase
"Greek tragedies". The Erasmatron was one system that is, perhaps, guilty
of this.
This objection seems only to apply if the story *does* arise from
simulation. However, a simulationist system that allowed more authorial
control than this would seem to bypass these problems. While some have
suggested that a simulationist system could itself generate plot, a more
realistic approach would be one where the author had complete control over
the plot and the simulation provided ways to traverse between (possibly
diverging) plot points.
Whether simulation does ultimately prove to be an uninteresting avenue for
interactive fiction is a question that will only be answered when a
simulationist system exists.
4. How to build a simulationist system?
The most promising approach to such a system seems to consist of defining
different levels of abstraction and enabling authors to efficiently move
between these in programming their games. This was an approach taken by
Facade. In a simulationist IF system, the range of abstraction would run
from objects to overall plot, but there are certainly other levels of
abstraction between these. An interesting question is which levels of
abstraction would be most useful?
5. Further reading.
Facade: <http://www.interactivestory.net/>
"AI and Authorship", Michael Meteas, _Grand Text Auto_
<http://grandtextauto.gatech.edu/2003/07/16/ai-and-authorship/>
Erasmatron: <http://www.erasmatazz.com>
"All that Erasmatazz", Neil deMause; _XYZZY News_
<http://www.xyzzynews.com/xyzzy.14f.html>
--
James Mitchelhill
ja...@disorderfeed.net
http://disorderfeed.net
rgrassi
Will have a closer look in the next days.
Rob
samwyse
> There's been a couple of recent threads on simulation and IF, which I've
> found quite interesting. The following is an attempt to sum up some of the
> responses that have emerged and clarify a few of my own thoughts about
> deeply simulated IF, what the problems are and how they might
> be solved.
Great post! This needs to be copied over to the if-wiki.
[...]
> 3. Problems with simulationism
>
> A number of problems with simulationism have been raised. These generally
> fall into three categories: Implementation issues, authoring issues and
> philosophical objections to simulationism as a whole. The following is not
> an exhaustive list, but contains most of the major points.
I suspect that simulationism works best with FPS games. If anyone even
invents a holodeck, it will work well there as well. A 3-D GUI
sidesteps the whole issue of generating human-like text, at the risk of
excluding a large sub-population of IF players.
The Nostagia Channel shows movies with a voice-over that describes the
action for the vison-impaired. Automating that voice-over is probably
just as difficult as generating text descriptions from a 3-D model.
Perhaps the author could tag the importance of various things and their
properties, so consise but relevant discriptions could be generated.
> It is not immediately clear how NPCs would fit into a simulationist
> system. One possibility is that their emotional states are modelled via an
> "emotional physics". One attempt at this kind of modelling was Chris
> Crawford's Erasmatron, which was widely derided by the IF community on its
> initial release. Another attempt, which received a much better reception
> was Facade.
The Erasmatron was IF's version of the Michelson-Morley experiment, a
spectacular but respectable failure. Crawford is a very talented game
designer/implementor and had both the time and the money to make it
work. That he couldn't tells us a lot.
> This objection seems only to apply if the story *does* arise from
> simulation. However, a simulationist system that allowed more authorial
> control than this would seem to bypass these problems. While some have
> suggested that a simulationist system could itself generate plot, a more
> realistic approach would be one where the author had complete control over
> the plot and the simulation provided ways to traverse between (possibly
> diverging) plot points.
>
> Whether simulation does ultimately prove to be an uninteresting avenue for
> interactive fiction is a question that will only be answered when a
> simulationist system exists.
Until then, LARPS and murder mystery dinner theater are probably the
best existing simulations of simulationism. ;-) Note that both do have
plots, but not many puzzles.