[MUD-Dev] Simulation Revisited

Dave Rickey daver at mythicentertainment.com
Sat Oct 13 16:35:46 New Zealand Daylight Time 2001

There's an interesting article by Harvey Smith (designer on Deus Ex)
at the IGDA website, apparently adapted from his keynote speech.  It
seems far too long to be pasted in here, so I'll just paste a link
and let the moderator decide.


Essentially, it focuses on the issues the list has been discussing
lately, on game as story vs. game as world, from the perspective of
single-player games.

--Dave Rickey

<EdNote: Copied below for archival purposes.  Its long, but not
*that* long.>

		     The Future of Game Design:
	  Moving Beyond Deus Ex and Other Dated Paradigms

			  by Harvey Smith


As an art form, immersive games are in a transitional state,
currently positioned on the cusp of something almost unrecognizably
different. Future games will employ deeper simulation in order to
achieve far greater levels of interaction and complexity, while
simultaneously simplifying the learning curve for new players. Most
game environments of the past have been based on crude abstractions
of reality, limiting player expression and requiring users to learn
a completely new vernacular in order to play. The games of the
future will rely heavily on much more complex, high fidelity world
representations that will allow for more emergent behavior and
unforeseen player interactions. Taken together, these
next-generation design paradigms are not simply improvements over
older models, but represent a fundamentally different approach to
simulating real-world physics, handling artificial intelligence and
interface usability.

Using the award winning and critically acclaimed game Deus Ex as an
experimental foundation for discussion of these new design
paradigms, come explore the theories that will bring about the
renaissance of the next-generation of interactive exploration.

(This lecture was first presented as the keynote address of the game
track at the Multimedia International Market, in Montreal, October

I - DX and Me

Hello. I'm Harvey Smith from Ion Storm Austin, an Eidos studio. I
was lead designer of Deus Ex and I'm project director of Deus Ex
2. This is intended as a lecture concerning the ways in which
increasingly complex simulations will lead to richer gameplay
environments in the near future. This is my first trip to Canada and
the first time I've attended the MIM conference. I'm glad to be
here. Prior to working for Ion, I worked at two other game
companies: Multitude, where I was lead designer of a game called
FireTeam, and Origin Systems, where I worked on several games in a
variety of roles. I started in the game industry as a quality
assurance tester in 1993.

Deus Ex, the game our studio finished last year, was a hybrid game
that attempted to create an environment in which the player was
calling the shots as much as possible. The game mixed a variety of
genre elements, including:

    * The action and point-of-view of first person shooters.

    * The story, character development and exploration of
      role-playing or adventure games.

    * The economic management and strategic expression of strategy

Deus Ex tried to provide the player with a host of player-expression
tools and then turn him loose in an immersive, atmospheric
environment. We wanted to do this in a way that did not limit the
player to a few predefined choices, but instead allowed the player
to come up with his own strategies within the flexible rules of the
environment. We wanted to allow the player to approach the game from
the direction of his choice, employing his own play-style cobbled
together from the options we allowed. Sometimes we succeeded;
sometimes we fell back on more traditional (more limited) means of
providing interactivity. The desire to give this talk today was
largely fueled by seeing both moments in Deus Ex.

When we did succeed in implementing gameplay in ways that allowed
the player a greater degree of freedom, players did things that
surprised us. For instance, some clever players figured out that
they could attach a proximity mine to the wall and hop up onto it
(because it was physically solid and therefore became a small ledge,
essentially). So then these players would attach a second mine a bit
higher, hop up onto the prox mine, reach back and remove the first
proximity mine, replace it higher on the wall, hop up one step
higher, and then repeat, thus climbing any wall in the game,
escaping our carefully predefined boundaries. This is obviously a
case where - had we known beforehand about the ways in which these
tools could be exploited - we might have capped the height or
something. Most of the other surprise examples I'll mention today
are going to be 'desirable' examples of emergence or emergent
strategy. But I thought I'd start with an undesirable example
because that's one of the things you have to watch for in attempting
to create flexible game systems that behave according to implicit,
rather than explicit rules. In any case, we were delighted at the
flexibility of the system, of the ingenuity of the players and of
the way that the game could, in some ways, be played according to
the player's desires, not the designers'.

When we failed in our attempt to implement gameplay according to our
lofty goals and instead fell back on some special case design,
players sometimes felt robbed if their actions caused a situation to
'break' or if we failed to account for some desired play-style. For
instance, many times we included three paths through a map and each
corresponded heavily to a play-style like stealth, combat or
high-tech hacking. If a player typically resorted to some other
play-style (like seeking out water passages and using SCUBA gear to
get past obstacles), then that player acutely felt the limitations
of what we had offered. Instead of feeling like he was operating
within a flexible simulation with consistent rules, suddenly the
player felt as if he needed to figure out what the designer wanted -
what the designer had explicitly planned as the 'right way' to
negotiate a part of the game. This problem was even further
exacerbated in the few cases where we provided only a single
option. For instance, at one point in the game (for plot purposes),
we wanted the player to set off a security alarm in one particular
research lab complex. There was no way to avoid setting off this
particular special case alarm, even for the player who had spent
most of his in-game time and resources on playing as a
counter-security specialist. Players felt completely robbed. This
was a forced failure in Deus Ex, created by a special case break in
the consistency of our game rules.

The success cases in Deus Ex tended to rely on the interaction of
flexible sub-systems within the game (and were about what the player
wanted to do). The moments that I perceive as failures tended to
rely on special-case triggering or scripting (and were more about
what the designer wanted the player to do). The experiences we had
working on DX1 motivated us to move further toward more deeply
simulated game environments. I'll return to Deus Ex off and on, but
first let me briefly outline my talk.

III - Simulation Overview

A simulation is a representational model. Computer and video games
have obviously simulated aspects of the real world (or some skewed
version of it) from day one. Early on, most of the simulations
involved were fairly simple. For instance, in Pitfall - the classic
Atari 2600 game - the notion of gravity existed; if the player
leaped, he moved up and forward, then fell, in a crude approximation
of gravity. On the other hand, you could point to Lunar Lander (and
a few other space games) as example in which a concept like gravity
was modeled in much greater detail, accounting for planetary mass,
directional thrust and momentum.

Modern examples of representational game systems abound, from crude
models to overly complicated models. For instance, some first person
perspective games have allowed the player to get into vehicles. In
some of these cases, the vehicle physics simulation is too crude
with regard to the way it interacts with the terrain, allowing the
player to get stuck on small hills that it seems like the vehicle
should be able to negotiate. On the opposite end of the simulation
scale, Trespasser is probably a game that, despite any innovations
or strengths it might have had, could be said to have failed because
it featured overly complex simulations without the requisite control
and feedback. So the vehicle stuck on a small bump is a symptom of a
simulation that's too crude for the game; conversely, Trespasser's
problems were a symptom of a simulation that was too complex for the

In the past, games have been mostly about branching paths. The
designer manually sets up a number of outcomes or interactions and
allows the player to pick one. This merely equates to a handful of
canned solutions to a particular game problem. (Some hypertext
writings refer to this as "multilinear," or allowing simultaneously
for multiple linear options of equal value.) Deus Ex featured some
options for player expression that were facilitated by systems of
coarser granularity. (Good examples here might include our branching
conversation system or a critical room that could be entered at only
three specific spots, each representing a different approach.)
Manually setting up solutions to game problems requires a lot of
work on the part of the team, can result in inconsistencies and
generally only equates to a small number of possibilities for the
player. However, Deus Ex also featured options for player expression
that were facilitated by systems of finer granularity. (Good
examples might include some of the player-tools that we provided
that were tied into analogue systems like lighting or sound, such as
the ability to see through walls or dampen the sound of
movement. These tools interacted with our enemy awareness models in
numerous, fairly complex ways. They could be activated at any time
in a very wide range of situations, incorporating distance, facing,
enemy type, etc.) The finer-granularity systems required more
feedback and introduced some uncertainty that equated to some
interesting degenerative exploits; but the freedom players felt more
than made up for these costs.

Essentially, almost all games involve representational models of
reality. So why talk about simulation? What's happening is that the
models are becoming finer in granularity. We're talking about a
scale here, with incrementally more weight being added to the sim
side. We're slowly moving toward games built upon much higher
fidelity conceptual models, with greater control or
self-expression. At some point, the scale will tip. There will come
a point (in part, an arbitrary point) at which gameplay in the
average game will be much richer because the player will be
presented with a vastly larger range of expressions. Yes, we're
moving incrementally along, but at a certain point, the systems
become flexible enough to allow for emergence, at which point the
experience is more about the player's desires.

Example list of slow progress metrics toward more complex

    * Example one: Birds fly up when player enters trigger
      radius. This is somewhat interactiveit requires player to
      approach specific spot at least.

    * Example two: Birds fly up when player draws within range or
      when specific events occur. For instance, weapons are
      explicitly told to broadcast a "birds scatter" type message.

    * Example three: Birds fly up in response to dynamically
      generated stimulus based on lower-level relationships between
      the unit and the stimulus. For instance, sight of enemy,
      loud/sudden sound, bright/sudden light, rapid motion. This
      version could get increasingly complex, depending on how you
      model the stimulus created by the player (or other in-game
      agents) like light or sound, as well how you modeled the
      birds' perceptions.

This brings up the question: Why should we continue to attempt to
build games around higher fidelity simulations? Why is a wider range
of expression better? Multiple reasons:

    * Simulation allows for more emergent behavior on the part of
      the game's systems and more emergent strategy on the part of
      the player. New gameplay is possible and a larger/deeper
      possibility space is created. Basically, this means that the
      player will have more than "a few canned options," which
      provides the game with greater potential to be perceived by
      players as interesting.

    * Games typically have more consistency when response to player
      stimulus springs from the interaction (according to rules
      about relationships) of the elements of a simulated system (as
      opposed to when response to player stimulus is derived from a
      bunch of special case, designer-driven instances).

    * As a labor-cost benefit, a better-simulated game environment
      requires less time to create content. This saves money, but it
      also allows designers more time to focus on tuning the
      gameplay. For instance, collectible card games feature an
      individual card's rules-of-play on the face of each card. The
      cards have been categorized into a system, with each card
      falling into a subclass. As a result, the rules written on
      each card do not have to explain how the card works with every
      other card created for the game; instead, each card's rules
      only explain how it interacts with a card subclass (or
      multiple subclasses). To be more specific, imagine a card for
      the Harry Potter card game (if that thought is not too
      painful) that stated, "Affects the following cards" This would
      require designer consideration of each card, it would require
      lots more space and lots more writing, plus it would preclude
      our example card from working with any future, unplanned
      cards. By instead using a system-with global rules governing
      the relationships between subclasses of cards-the game a) does
      not require the designer to consider every possible
      permutation, b) it allows the card to function with future
      card releases and c) it allows for emergent strategy. (Which
      leads to our next consideration)

There are also a couple of side effects of setting out with the goal
of creating games around deeper simulations:

    * Emergence in games is mostly a benefit with potentially
      wondrous ramifications, but also something of a cost. In a
      flexible system in which designers don't attempt to provide an
      explicit relationship for every element in the system,
      uncertainty is introduced. This often leads to interesting
      implicit consequence - players can formulate plans that spring
      from indirect interactions of the rules system. For instance,
      in the online strategy game ChronX, a player can obtain and
      use one of the game powers to enhance an organic unit (like a
      human soldier), making it a more powerful 'mech' unit (or a
      sort of cyborg). Normally, making an enemy more powerful is
      not something you'd want to do. However, if he has access to
      it, this player can then use another game power - one that
      steals enemy mech units - to cause the now-more powerful,
      now-mechanized enemy soldier to switch sides. The first card -
      normally played on a player's own units to enhance them - does
      not have an explicit relationship with the card that steals
      mechs, but they work well together if the player sees and
      exploits this emergent strategy. Unfortunately, the
      uncertainty introduced by this approach can also lead to
      exploits that break the game. Bulletproofing against these
      exploits requires time and effort. (The Deus Ex 'proximity
      mine climbing' method I mentioned earlier is a good example of
      such an exploit that we didn't catch.
    * Another side effect: Purely on the downside of the flexible
      rules system approach, better user feedback is required to
      avoid confusing the player, since a more complex simulation
      usually equates to a more granular range of player
      expression. For instance, some games have emulated enemy
      awareness using directional facing. In other words, an enemy
      unit can only see what is in front of it, within its field of
      view. Thief (by Looking Glass Technologies) came along and
      introduced a much deeper awareness model, involving complex
      sound propagation and lighting that acted as stimuli that the
      enemy could perceive. Since understanding lighting and shadows
      was key to the player's success as a thief, the player needed
      a really good indicator as to how well lit he was at any given
      time. Since Thief is a first person perspective game, the
      designers added a "light gem" feedback device to inform the
      player as to his current light-based visibility. Thief asks
      the player to understand a much more complicated model, but it
      also helps the player out by offering some information germane
      to that model. Using concepts like noise and shadow, and
      elements like thieves and guards, Thief also puts things into
      a familiar, realistic context. While 'realism' itself is not
      always the goal in a game, using game settings and elements
      that relate to the real world - with which the player has
      great familiarity - often helps make the game inherently more
      intuitive, sidestepping some of the additional cost. For
      instance, if you use elements like "fire" as a part of your
      game systems and if it actually behaves like fire does in the
      real world, players will probably have an immediate
      understanding of this element without requiring the game to
      educate them.

IV Game Simulation - Specific Systems

I've talked some about specific systems in passing - Thief's sound
propagation and lighting, for instance. Now let's get more specific:

Sound/Light and Unit Awareness:

Many games model 'enemy awareness' in some way, attempting to
simulate the real-time gathering of information. In most combat
games, for instance, enemies perceive hostile or suspicious
events. I think we're at a point where traditional models for
perception are just not enough - relying on such models is having an
increasingly negative impact on overall gameplay.

For instance, in DX1, sound propagation worked like this: A sound
event was broadcast in a sphere outward from a source, ignoring
wall/floor surfaces (as if the sound were generated in an empty
space). Taking distance into account, units within the broadcast
would be alerted (i.e., would 'perceive' the sound). A different
model was used to determine whether or not to play a sound for the
player (involving a line-of-sight check to fake dampening a sound if
it was playing through a door, for instance).

By contrast, let's look at our plan for sound propagation in DX2
(which we think is the next step in the direction undertaken by
Thief): A sound event is broadcast in a sphere outward from a
source. In cases where the sound hits a surface, we bounce the
sound, taking into account the material applied to the surface. (So
that carpet muffles the sound, for instance.) The number of bounces
is capped. Taking distance into account, units 'perceive' the sound
if the sound reaches them, directly or by bounce. The same model is
used for both player and game unit (or guard) to determine whether
the sound is perceptible. Certain acoustic aesthetic effects are
ignored on the AI side, but these have nothing to do with whether
the AI perceives the sound.

The first model (the one used by DX1) did not always allow the
player to predict whether a game unit (like a guard) would hear a
sound or not, which led to some really unsatisfying occurrences:
Either a guard would hear the player (when the player assumed that
he was acting 'quietly'), or the player would make sound that he
assumed a guard should hear (but the guard wouldn't, making the
game's awareness model feel broken). We think the second model (the
one being used for Thief3 and DX2) has the following benefits: We
can unify player-related and enemy-related sound propagation, which
will allow for a more intuitive game environment. The player will be
able to make assumptions about whether a guard will hear him or not
based on the player's own perception of sounds in the
environment. We also hope that the higher fidelity model will equate
to a more 'fair' gameplay model; guards will not hear sounds that
are blocked by multiple thick walls. (Again, this will allow the
player to make some strategic assumptions, closing a vault door
before operating a noisy tool, for instance.)

Anecdotally, I want to mention that DX1 players already do things
like closing doors before taking actions (because that is the
intuitive thing to do - something we learn from childhood forward,
trying to trick our parents and siblings). If players do this, but
realize that the system does not take something like a closed door
into account, they feel cheated or let down. If they're going to do
it anyway, it makes some sense to model the game according to their
intuition and assumptions; we don't want to pass up the chance to
squeeze in an interesting, intuitive game dynamic. (This is a good
example of a deeper simulation leading directly to more player
expression, more gameplay.)

Realistic Physics:

Currently physics is useful for establishing player-action
capabilities - limitations related to movement speed, falling
damage, gravity, etc. But over the last few years moving toward more
realistic physics has had other significant gameplay ramifications
as well.

First, a comment about the word "realistic":

In games, realism is not necessarily the goal. But if the world
seems to behave consistently and in ways that the player
understands, it seems that the player has less difficulty immersing
himself in the environment, suspending his disbelief. In this way,
realism in games is related to intuitiveness and player
expectation. (It's also worth noting that if you set up an
environment that seems familiar (and thus is intuitive) then you
thwart the player's expectation of that environment, the player
often finds it extremely jarring. For instance, we included
telephones in Deus Ex and gave them limited functionality. Their
presence helped the player identify, say, an office space as a
familiar, real-world location. However, we could not possibly make
the phone in the game as flexible and powerful as a real-world phone
is, and the lack of functionality in the game-phones served to
immediately remind the player that the office space was "fake." It
might have been better to leave the phones out altogether. So
realism is not the point (even though it can be useful).

Continuing with "realistic physics": The first game I played that
allowed me to realistically bounce grenades around corners was
System Shock. Bouncing grenades around corners is an example of
"physics as gameplay." It's one step less direct: Instead of going
toe-to-toe with an enemy, the player can take up a safer (more
strategic) vantage before attacking. The player suddenly had new,
interesting options. It also makes the environment more dynamic: If
someone moves a crate out into the center of the room, a grenade can
then be bounced off the crate. Obviously, collision physics that
allow for grenade bouncing gameplay have been around for a
while. But the more thorough and more realistic physics simulations
of the next generation of games should have interesting
ramifications. To cite some examples:

    * New gameplay tools: If we track mass and gravity, for
      instance, we can arm the player with a tool that increases
      mass, allowing for all sorts of interesting effects. This is
      one of the goals of our studio-to continue to widen the range
      of gameplay tools beyond "more guns." Not because we dislike
      games with guns, but because we are looking to make the game
      more interestingto expand the possibility space.

    * More intuitive environment: "Of course paper should burn." (In
      today's games, casual players might be baffled by the physics
      of the world: Only explosive barrels and bodies burn,
      sometimes pieces of light furniture cannot be moved around,
      the player-character can often not perform simple tasks like
      climbing up onto a desk and sometimes glass does not
      break. Why *wouldn't* this harm accessibility? To play, you
      must re-learn the physics of the world, like a child.) When
      the world works in a way that makes sense to a human
      (non-gamer), because it functions in ways that reflect their
      lifelong experience, the average person is more likely to find
      the game environment "intuitive" even in fantasy realms and
      alien dimensions.

General Game Systems: Tools and Objects

In the past, gameplay tools (including weapons) had to have explicit
relationships with any other elements of the game in order to affect
those elements. So a weapon class, for instance, specifically
contained code listing all the things it could affect. For instance,
to use a simplistic example, if you wanted the bullets from a gun to
break a window, you had to set up a direct relationship between the
weapon entity and the glass entity. Now, there's an additional layer
of abstraction between the two: The weapon projects a bullet
entity. The bullet entity carries with it information about what
properties it carries (like ballistic damage, heat or electricity,
for instance) and the glass is a stimulus-receiving entity. When the
bullet meets the glass, the game's object/property system looks up
the effect of the bullet's properties on the glass entity. There is
a set of rules about the relationships between these general-case

How is this different, from a pragmatic standpoint? The latter, more
flexible approach (with the layer of abstraction between the bullet
and glass game elements) has the following benefits:

    * Global consistency: Game environments now include thousands of
      object types. Using the old method - involving direct, special
      case relationships - it would be easy to fail to create a
      relationship between something, say, like a potted plant and a
      bullet. So the bullet might ricochet and fail to break the
      potted plant. This counter-intuitive physical interaction
      between the plant and the bullet might break the user out of
      the experience by defying his intuitive expectations. In the
      more flexible system (in which the bullet merely carries
      stimulus properties to which damageable object subclasses can
      respond), everything is more likely to be covered, instead of
      only the things that were manually given stimulus-response

    * Time saved: Also, since we're talking about an environment
      hosting thousands of objects, instead of hard-coding
      everything, programmers can build tools that allow designers
      to attribute properties to any new object class via a simple
      tag. So this model saves development time.

    * Emergence: In Deus Ex, we found that players (initially just
      in QA, but later among the game's fans) were using an emergent
      strategy that had never occurred to us. One of the unit types
      (an MJ12 soldier character) exploded upon death. Our idea was
      that this would cause the player to react strategically,
      switching away from a pointblank weapon when fighting this
      unit. In a more traditional game systems model, we would have
      created an explosion entity with an explicit relationship to
      the player, damaging the player if he was within range of the
      explosion. However, in our more flexible system, we simply
      spawned a generic explosion with properties related to
      concussive/ballistic damage. Players figured out that they
      should lead this unit near a locked container before
      delivering the final blow. When the explosive unit blew up, it
      inflicted damage on the locked container, opening it up. (We
      did not plan this or even foresee it - it just worked.) In
      this way, players were exploiting the system in order to open
      locked doors and safes (without spending any lock picking
      resources). We were delighted.

It's largely due to hardware limitations and the nascent state of
interactive entertainment that games have by necessity relied on
cruder models in the past. No single game project of which I've been
a part, including Deus Ex, has fully taken advantage of all the
opportunities to provide the player with as much exploration and
expression as possible. With that qualifier, I will relate the
following example:

Recently at one of the game industry's conferences, I had an
opportunity to see the demo for an upcoming game. I've been excited
by this game for quite a while. It's essentially an adventure or
role-playing game that allows the player to explore a fictional
world, building up his power so that he can face increasingly tough
threats, while uncovering new pages of the game's plot. This is a
traditional conceptual model, but a popular one that has provided a
lot of enjoyment over the years. This new game looks and sounds
beautiful; I fully expect it to be a lot of fun. (I'll be buying it)
But after talking to one of the developers and watching him play the
game, I cannot help but point out how I think that the designers
have missed some opportunities. The game seems to feature an
extensive set of player tools and powers. However, most of them are
purely related to inflicting damage. The rest of the environment is
modeled in a very simple way. The game uses a traditional paper
RPG-style 'spell' system, which should allow for a great number of
interesting player expressions, even if you restrict your thinking
to the tactical arena. So, during the demo, I inquired about types
of spells that, in paper RPG's, are often exploited in interesting
ways beyond toe-to-toe combat. For instance: Can the player freeze
the water pool (in the cave featured as part of the demo) as a way
of creating an alternate path around an enemy? Can the player
levitate a lightweight enemy up off the ground and thus get by it
without resorting to violence? Can the player take the form of a
harmless ambient animal and sneak past the goblin? Can the player
create fake sound-generating entities that distract the enemy? I
believe the answer to all these questions is "no." The game was
designed around pre-planned, emulated relationships between
objects. Had the game been designed around a more flexible
simulation, these sorts of interactions might have just worked, even
if they had never occurred to the designers. (All of this still
might be possible in the special case emulation model, but would run
the risk of a great deal of inconsistency, would require tons of
work and would not as likely produce emergent results.) Had the game
been built around more thoroughly simulated game systems, creating
more interesting (less combat-centric) tools would have been easier
- the game's possibility space would have been greatly enlarged.

By contrast, let's look at the gameplay tools given to the player
for the game System Shock 2 (by Irrational Games and Looking Glass
Technologies). There was a web post about a player who, when under
attack (by a mutant and a turret) and completely out of ammo, used
psi-telekinesis power to pull an explosive barrel toward him, moving
it through the firing arc of an attacking turret. The turret blew up
the barrel, destroying the turret and killing the mutant. No one on
the System Shock 2 development team explicitly set this area up with
this outcome in mind; these things emerged from the game's
general-purpose approach to gameplay tools interacting with the
other elements at the whim of this (clever) player. This is a really
good example of a flexible, consistent set of rules, very similar to
our bullet/glass or collectible card game examples from earlier:
Rules about the relationships between the game's objects and tools
had been established at a high level. No code or scripting
specifically related to the idea that the player's psi-telekinesis
could pull barrels in front of turrets; instead the psi-telekinesis
was set up to affect moveable objects, the barrel was tagged as a
moveable object, the turret projectiles were set up to affect
explosive objects and the barrel was set up as an explosive
object. And everything just worked.

Again, as a downside, in attempting to create flexible game systems
(that behave according to implicit, rather than explicit rules)
problems are caused by undesirable exploits. So efforts must be
undertaken to bulletproof against anything that outright breaks the

Unit Needs and Behaviors

Most game units have very limited awareness of their state, needs or
environment. They generally don't need any greater awareness:
Imagine a racing game in which one of the drivers was distraught or
suicidal because his Sim-girlfriend had just broken off their
relationship. Sounds ridiculous. But imagine a racing model in which
the drivers were intelligent agents who were aware of their car's
current fuel needs. That sounds interesting (to me). And, to
integrate some of what we've talked about, imagine that this
self-driver then uses the game's thoroughly modeled aerodynamic
system to 'draft' behind another racer to conserve fuel. For all I
know, people making racing games might already be doing this - my
point is that the deeper simulation in our hypothetical model
provides a much larger possibility space. The self-aware driver
provides a more interesting AI opponent and the wind-drag model
allows the player to take more strategic elements into account and
act upon them.

The deeper simulation of additional aspects of a game does not
inherently make the game more fun. But if you choose the 'right'
aspect to simulate, you can make the game more interesting. For
instance, DX combat featured units that would run away if they
realized they were badly wounded. This did not make combat more fun,
but it made it one step more interesting than a toe-to-toe
shootout. Players remarked at how it prompted ethical decisions:
Track him down and shoot him in the back, or let him go, since he is
no longer a threat? For DX2, we're thinking of ways of expanding
upon this idea, allowing units within a group to maintain awareness
of group needs as well as individual needs. These leads to some
obvious ideas: A medic squad member, a commander, etc.

Another direction in which we're trying to move for DX2 is real-time
IK-based movement. People talk a lot about it, but we want to use it
for gameplay-specific purposes. With IK pointing, touching and head
movement, suddenly character movement is not limited to what an
artist has pre-defined. With IK, we can model more on the unit's
response-to-environment side. The IK will let a unit flexibly act on
its desires. For instance, if a bystander thinks that the left door
is the one that the police should open, it can point in real-time to
the left door. (While not a needs-based behavior, the IK facilitates
expression of this behavior. The IK 'body language' communicates AI
state and change to the player.)

V - And Beyond

What comes next? Clearly we're moving along a curve of greater
hardware capability, more elaborate software systems and a more
sophisticated understanding of our nascent art form. What's the next
revolutionary gameplay angle someone will exploit by figuring out a
deeper, more interesting way to model a game system? I can't say
with certainty, of course. But I can look at the last cycle of games
and point to two interesting, noteworthy examples:

    * Thief looks on the surface like a shooter. However, the game
      design team at Looking Glass decided to model sound
      propagation, lighting and AI awareness in a much more complex
      way. In doing so, they greatly expanded the possibility space
      of the first person perspective shooter. They were smart
      enough to know that their approach required them to provide
      the player with a great deal more feedback.

    * The Sims (by Maxis) created a character "needs" model that,
      while it seems fairly simple, is far more complex than
      anything used to represent the moods and needs of most game
      characters. (Most game units, of course, have no concept of
      anything much more than whether they can see an enemy. Even in
      all the games that rely heavily on the game industry's
      meat-and-potatoes of faux combat, units generally fight until
      they drop dead (instead of running away when badly wounded),
      fail to intelligently switch weapons (based on the situation
      or upon enemy defense), and lack any significant amount of
      tactical awareness with regard to their squad mates. In
      creating their character needs model, Maxis created a sandbox
      of possibility that was entertaining to explore,
      conceptually. It didn't feel like a game - in that there were
      no hard-and-fast victory conditions and little in the way of
      artificial conflict - but through its flexible system it
      allowed the player a lot more expression than most games.

Someone in the next cycle, I hope, will pick out a new area, model
it in a high-fidelity way that can be made interesting for the
player, and will contribute their own part to the revolution. Maybe
they will leapfrog from Thief's sound/light/awareness simulation in
another stealth game, or maybe they'll pick up where The Sims left
off and create characters that seem remarkably alive, with feelings,
moods and relationships.

But what lies beyond the short-term? (This part is for fun -
something to embarrass me in the future, like an insulting note from
my past self.) How will games be different a decade from now? Here's
some hopeful and perhaps provocative speculation:

    * Speech Synthesis and Dynamic Game Conversations: Imagine if
      the game could assess a situation based on a long series of
      relevant player inputs, string together some responses and
      construct a convincing verbal response using a speech synth
      system. Suddenly, vastly more interactivity is possible. Once
      again, instead of a few canned responses (provided by the
      designer), the game could allow for a much wider range of
      responses - games might someday be able to analyze voice input
      and formulate a conversation that never had to be written by a
      designera conversation of much greater relevance to the
      player's actions. (And when speech synthesis is combined with
      true artificial intelligence, narrative games will finally
      become truly interactive.)

    * Long-term persistent games: The player starts a game and plays
      it for years (or his entire lifetime) as it wraps itself
      around his choices. The more he plays, the more unique the
      game gets.

    * Auto-generated content: At some point, games might dynamically
      generate terrain and architecture, creating entire cities on
      the fly, based on some parameters. Also, units (or characters)
      will be created in the same way. Building all this around
      player input - or past player decisions - will allow games to
      spin out alternate futures based on the player's initial

    * Intelligence vs. Multiplayer: Most of us have accepted MP as
      the future. But if AI entities were as smart as people,
      wouldn't narcissism dictate the desire for SP? Would you
      rather have 4 obnoxious roommates or a really good dog? Some
      experiences might be better qualitatively in a SP
      environment. For instance, is it spookier to explore a haunted
      house alone or with 100 people? Also, MP games currently use
      fairly static, traditional environments and rely on the agency
      of other players to create interesting (or emergent)
      interactions. Immersive sims are SP games with huge emphasis
      on creating an interesting (dynamic, interactive) environment
      and an expressive set of player tools, hopefully
      (increasingly) built using simulations. Imagine if you
      combined these two.

VI - Summation

This wraps up the lecture. I hope you enjoyed listening as much I
enjoyed preparing. Actually, I hope you enjoyed listening a great
deal more than I enjoyed preparing.

As games continue to rely on increasingly realistic or complex
simulations, obviously we'll have a bunch of problems to solve
related to uncertainty and user feedback. But the end result, if we
solve those problems, will be unprecedented possibility in
games. Striving for finer granularity in the representational
systems we create for games should allow players much more freedom
of expression and should make the 'game' experience more about the
player and less about the designer. We want players evaluating their
environments, considering their tools and formulating their own
strategies with as little regard as possible for what we as
designers might have wanted them to do. Older game genres might be
completely reinvented when built upon deeper
simulations. Additionally, new game forms will emerge. Even though
this approach involves the designer surrendering some control of the
game's emergent narrative to the player, ultimately this should
prove much more creatively satisfying; our goal is to entertain, to
allow players to interact and express. In the future, we might only
be "designing" games at a higher level, establishing parameters and
allowing the players and the game's intelligent agents to work out
the details.

Lastly, before I stop talking, I'd like to offer special thanks to
the people who have taught me what I know about design and
development, without whom Deus Ex would never have been made: Doug
Church, Warren Spector, Marc LeBlanc and everyone at Looking Glass
and Ion Storm Austin. Thank you and goodbye.

Author Bio


Harvey Smith is Project Director of the Deus Ex 2 team at the
award-winning game studio ION Storm Austin. Recently, he worked as
Lead Designer of the genre-bending game Deus Ex, winner of numerous
Game of the Year awards, as well as the 2000 BAFTA for Interactive
Achievement. Harvey, along with Warren Spector, was awarded the 2001
Game Developers Choice Award for Excellence in Game
Design. Previously, he worked at Multitude on FireTeam, a team game
that allowed players to co-operate using real-time voice over the
Internet. Prior to Multitude, Harvey worked at Origin Systems in
roles ranging from Lead Tester to Producer on several ambitious game
projects. He is a member of the Advisory Board for Game Developer
Magazine. Beyond video games, Harvey's interests include comics,
dogs and writing. He lives in Austin, Texas.

Harvey can be reached at: witchboy at ionstorm.com

MUD-Dev mailing list
MUD-Dev at kanga.nu

More information about the MUD-Dev mailing list