L’AI Terrible:

A funny thing happened on the way to the Oblivion Gate

Open world games operate on laws that are an amalgam of game tropes and simulated agents. The Elder Scrolls IV: Oblivion (2006) is a prime example of this interplay. Oblivion is hard to classify. It is the product of a long history of game convergence and cross-pollinating many mechanical traits of past games. It could be simply called a single-player RPG, or an open world/sandbox game, or a massively single-player RPG. Semantics aside, it is a virtual fantasy world that simulates combat, economies, biota, and many other things. By contemporary standards, it is an enormous game world with acres of programmed circuits that create the illusion of landscape, continuous space, humans, and creatures.

With Oblivion, developer Bethesda Softworks created a new engine for the franchise and many new systems to take advantage of their first foray into a new technological generation for games. Of primary concern is the game’s ‘Radiant AI system’, which imitates the impact of resource pressure, individual needs, and player interaction on a non-player character (NPC), resulting in a dynamic, procedural reaction. Radiant AI (RAI) contains a simplistic rubric for individual personality that it assigns to each AI. Based on a number of factors, characters posess a scaling value set that makes up their rudimentary personality, which is susceptible to change. NPCs have gradations of disposition toward the player and other NPCs.

For the most part, NPCs follow predictable paths in the game and fit into designated roles. These classes include:

  • Aggressive enemies: Usually creatures that attack the player on-sight.
  • Allies: Characters who defend you or work toward defeating a mutual threat.
  • Quest-essential characters: Issue vital directives or items to the player that are needed to carry out a mission.
  • Guards: Enforce in-game laws.
  • Merchants: Constitute the in-game economy by offering paid services to the player and other NPCs.
  • Citizens: Simply live in the world and commune with the player and other NPCs.

            However, RAI can sometimes influence a character to step out of their role and exhibit some unusual behavior that is motivated by a particular goal, or a response to another agent. Players seldom witness these behaviors, since they are always being procedurally generated throughout the game’s world. But there was a period during the game’s development that RAI had more liberty to act on impulse and react to a personal motivation. Game designer on Oblivion, Emil Pagliarulo, described some of this bizarre and remarkabl behavior:

“ In some cases, we the developers have had to consciously tone down the types of behavior they carry out. Again, why? Because sometimes, the AI is so goddamned smart and determined it screws up our quests! Seriously, sometimes it's gotten so weird it's like dealing with a holodeck that's gone sentient. Imagine playing the Sims, and your Sims have a penchant for murder and theft. So a lot of the time this stuff is funny, and amazing, and emergent, and it's awesome when it happens. Other times, it's so unexpected, it breaks stuff. Designers need a certain amount of control over the scenarios they create, and things can go haywire when NPCs have a mind of their own.

Funny example: In one Dark Brotherhood quest, you can meet up with this shady merchant who sells skooma. During testing, the NPC would be dead when the player got to him. Why? NPCs from the local skooma den were trying to get their fix, didn't have any skooma, and were killing the merchant to get it!” (Pagliarulo 2006) 

Other examples:

“ 1. One character was given a rake and the goal “rake leaves”; another was given a broom and the goal “sweep paths,” and this worked smoothly. Then they swapped the items, so that the raker was given a broom and the sweeper was given the rake. In the end, one of them killed the other so he could get the proper item. 2. In another test, a minotaur was given a task of protecting a unicorn. However, the Minotaur repeatedly tried to kill the unicorn because he was set to be an aggressive creature. […] While testing to confirm that the physics models for a magical item known as the “Skull of Corruption,” which creates an evil copy of the character/monster it is used on, were working properly, a tester dropped the item on the ground. An NPC immediately picked it up and used it on the player character, creating a copy of him that proceeded to kill every NPC in sight. 5. In one test, after a guard became hungry and left his post in search of food, the other guards followed to arrest him. The town people looted the town shops, due to lack of guards. Bethesda worked to fix these issues, balancing an NPC’s needs against his penchant for destruction so that the game world still functions in a usable fashion. In-game there are over 1,000 different NPCs, not including randomly spawned monsters and bandits. The result is that the AI in the release version is much reduced, only featuring NPC schedules.” (Sanjeev 2006)


After reading of these past incarnations of RAI, it felt like the NPCs in the release build of Oblivion had been lobotomized. Over the course of the 200+ hours I’ve spent playing the game, the only unusual RAI behavior I can readily recall witnessing was a battle to the death between two archers in a forest. There are only so many hours one can spend playing the game before the illusion wears off and the recursive Groundhog Day reality is exposed. Characters are anesthetized by strict scripting parameters, and feel all the more wooden and unsurprising. You’re stuck in a looping dollhouse, a kind of animatronic half-life where listless husks parrot the same awkward conversations, observe the same rituals, and follow the same itineraries of eating, strolling, and sleeping in the daily humdrum of their little lives. The world of Oblivion is a place of comatose wanderers, interned Alzheimers patients who serve as “information kiosks” (Abbott 2009).

I can understand why Bethesda worked to curb some of RAI’s past behavior. Most of their procedural expression took the form of destruction or theft, which in some cases lead to a devastated environment of routine mass murders and significant gameplay opportunities that were robbed of the player. However, the world that shipped is so sedated that this behavioral latitiude seems absent. Could RAI have been more effectively tempered without diluting it to its current state? Why were the past lives of NPCs sapped of variation? At the very least, why not include the option to activate prior RAI settings? Essentially, why were these behaviors not permitted to live? The developers maintain that these changes were made to “balance” the game, but I believe more imbalance and NPC latitude is needed in open world games.


Emergence and progression in the open world

The questions posited above find their roots in the Player-Game-World Problem. Within virtual worlds, what factors dictate how the game is played? What agents shape the experience of play? Is it the game itself, the player, or a combination of the two? I prefer to think of interaction as an aggregate of the latter. The laws of the game world erect certain constraints that impact the degree of potential interaction, but players project their own meanings onto play, participate in unique personal experiences in which cognitive arousal varies, and players can mechanically and observably change the game to produce unintended, unforeseen consequences. It seems that certain games are constructed with varying capacities for player manipulation and revision; some games are more malleable than others. Certain design traits influence this malleability, such as the number of game options, the contest between emergence and progression, the existence or absence of an explicit goal, whether or not the player is forced to complete it or meet all of its terms, and the number of ways a player can resolve a challenge.

In particular, emergence and progression are of chief concern. Game theorist, Jesper Juul says of the two,


“[…] Progression structures are heavily pre-controlled by the game, whereas emergence structures allow for much variation and improvisation that was neither anticipated by the game designer, nor is easily derivable from the rules of a game. However, this does not mean that players are a) completely free to do whatever they like or b) that their behaviour is devoid of pattern or regularity.” (Juul 2002)


According to Juul’s model, Oblivion is “a game of emergence, with embedded progression structures.” There exists a narrative, progression-based framework within the game, comprised of quests, or explicit goals the player can choose to seek out and complete, which occasionally offer multiple, optional clauses that allow the player to carry out the mission in a few prescribed ways. Then, there is RAI, an emergent framework that produces unique unforeseen and untelegraphed behaviors that impact gameplay. However, not all of RAI is emergent. For example, the game makes mention of the chance that if the player is asleep, they can contract vampirism if a vampire feeds on them. This is a random occurrence, but one that is anticipated and built into the game’s rules. On the other hand, NPC behavior such as the rake and Skull of Corruption incidents detailed above are emergent.

It’s clear that Emil and the other Oblivion developers value emergence, otherwise they never would have built the RAI system. Yet, I question why they would make the system so flexible, and then reconfigure it and introduce amendments to effectively make it less flexible. Since it is an emergent system, they could not foresee some of the behaviors that would occur, and most of which were not in accord with their idea of acceptable and balanced emergence. In a general sense, it seems like they take issue with those manifestations of emergence that disrupt progression. I do not intimately know how disastrous some of the byproducts of less restricted emergent AI were, but the fact remains that the world they settled on is a more reductive, restrained, and creatively-stifled idea of emergence than I would have liked. Where they see a problem, I see a benefit.

Within the game’s vast wasteland of recurring design-sanctioned emergence, the player is driven to escape the doldrums of this ultimately malnourished framework and undertake the progression framework. With the exception of a handful of questlines, the progression component of the game is almost uniform in its monotony. Most quests boil down to speaking with a character to receive another directive, killing designated NPCs, raiding specific locations, collecting and returning a particular item, or in more rare instances, protecting something or someone against an aggressor. Quest completion will yield either/or a.) narrative development b.) in-game currency, access to new areas and abilities, and/or commodities (e.g. armor, weapons, items, custom spells, etc.). The problem with the progression component is that it’s cyclical, formulaic, and safe. Gameplay remains resistant to meaningful, permanent change. Nothing is taken away from you, everything is taken for granted, worldly threats become empty, and your reward is seldom an interesting or unusual experience. Instead, you are fattened up with material, and gameplay becomes more trite and habitual as a result. This is customary of many RPGs and has become known as the Carrot and Stick approach to game design. The player plods along, trying to get a reward that is just out of reach. At the conclusion of a quest, the carrot is given to the player, while a fresh one is being reattached. When attempting to create a sprawling open world with an embedded progression structure, it would be absurd to expect the development team to make each quest offer truly unique experiences and radically diverse goals. However, with a more open treatment of emergence, the tedium of progression could have been alleviated, and a more dynamic world cultivated.


Repressed simulationist decisions

With the idea of emergence now introduced, we will examine the GNS Theory of role-playing games, with respect to Oblivion, its designers, and the player. The theory asserts that games, as well as players, promote or align with behavior that falls under three distinct categories: Gamist, Narrativist, and Simulationist.


“ Gamist refers to decisions based on what will most effectively solve the problem posed. These decisions are most common in games which pit characters against successively tougher challenges and opponents, and may not spend much time explaining why the characters are facing them.

Narrativist refers to decisions based on what would best further a dramatic story or address a central theme.

Simulationist refers to decisions based on what would be most realistic or plausible within the game's setting, or to a game where the rules try to simulate the way that things work in that world, or at least the way that they could be thought of working.” (GNS Theory 2010)


These are certainly principal components of engineering and diagnosing particular gameplay behavior. Using this rubric, Oblivion is a marriage between gamist and simulationist implementations, but with gamist traits possessing dominance. The gamist properties of Oblivion include: skill trees, superhuman abilities (such as casting spells and running on water), accelerated accumulation of experience, time of day manipulation, enchanted items that buff player aptitude in certain areas, recovering from mortal injuries by ingesting HP-replenishing food or potions, quest-essential NPCs going “unconscious” instead of dying, and fast traveling. The simulationist properties include: Drowning, fall damage, disease contraction, serving jail time for unlawful offenses, NPCs that cannot be resurrected from death, earning experience in a discipline through practice, melee combat that distributes damage based on parries, thrusts, where the blow lands, and armor resiliency, but most importantly, all of the simulated biological and goal-oriented needs that dictate RAI’s behavioral expression.

The team’s decision to tone down emergent NPC behavior suggests that it was motivated by a gamist approach. Many of these unexpected properties of RAI approximated human motivation, and thus, were closer to simulationist decisions, and as Emil mentioned, they produced unwanted outcomes that broke progression elements. Of course it is not commonplace for a person to kill another human being just to obtain an insignificant item (such as a rake), though it does happen. Even when humans use destruction as a means to an end, it is not enacted as orderly or unemotionally as RAI does, barring psychopathy. However, RAI provided a simulated motive and intention to killing, and other activities, albeit performed in both gamist and simulationist fashions. A myriad of intentions and motvations begets diverse, and more frequent emergent behavior. More scripted and assigned intentions, with suppressed NPC will, begets further predictable, repetitive, uncomplicated behavior, and subsequently spawns the incredibley wooden and thin AI that exists in the game.


Conclusion: Childproofing the playpen

We arrive back at the team’s aversion to those emergent, simulationist behaviors in question that “broke” the game. The question I posed earlier, “Why weren’t some of these behaviors permitted to live?” is more theoretical than practical. Bethesda had the ability to ship Oblivion with those RAI settings, but they decided against it. I can only speculate, but some people on the team may have been simulationists that wanted to see those exterminated behaviors make it to the release code. Emil himself displays a particular admiration and wonder when describing those now defunct emergent activities. However, I think they were more influenced in this matter by player expectation than designer intent. It seems the most plausible explanation is that the majority of players are gamists who hate when things break, bemoaning closed gameplay opportunities, agitated when they don’t have what they feel is complete agency, and even blaming their controllers for compromised performance. If Oblivion – God and designer forbid – attempted to make death and resource pressure a force of game nature and NPC will, I would guess that sales figures would be significantly damaged and the collective tantrums of the community would force Bethesda to release a costly RAI patch to mitigate their unpopular transgressions.

If games want to create convincing and compelling game worlds that foster true emergent behavior and consequence, they need to embrace the broken by introducing entropic agents. Reality is full of broken, irrepairable things. We are acted upon by a number of external agents, and chaotic processes. The tidy, control-obsessed attitude toward game worlds and entitled players is a limited aesthetic imperative that negates interesting dynamic behaviors. In some ways, Oblivion succeeds for a time at displaying a convincing, nuanced game world. But at a certain point abstraction and progression cannot mask a world that is too convenient. The narrative of the game is predicated on the idea of chaos infiltrating this world through “gates” that enable a hellish, barbaric dimension to crossover. It seems only fitting then that chaos and entropy be allowed to actually impact the world, so it is more of a reality than a fiction perpetuated by characters. In the game, every chest exists just for you. You are essentially an expected guest, a destined hero impervious to restrictive forces that affect denizens of the real world.

Take my newest character in Oblivion for instance. As I do with most malleable open world games like this, I played my second time through not as myself, but as a moral extreme, often times antithetical to how I would ordinarily react. I had no interest in saving the world of Cyrodill in the main quest. I tried selling the Amulet of Kings, taking it off and discarding it in a bog somewhere, but the game didn’t allow me to. I had no interest in that responsibility. Now, if you wait long enough, why doesn’t an NPC step up to the challenge? All of Cyrodill is waiting patiently for you, and only you, special ole you. It would be more interesting if you had a sudden change of heart, continued progress with that quest of saving the world later on and found that an NPC beat you to it, or had already done most of the work and for once, you weren’t viewed as the glowing, hero, the harbinger of balance and harmony. You were viewed as that miscreant who was either too absent-minded or too self-absorbed to assist with this duty of saving everything. For once you’re not the hero.


This is a prelude to a larger piece that's been in the works for some time. The above is a skeleton, a research question with a rough argument. The next multi-part installments will not just identify what I see as a problem for the aesthetic potential of open world games, but also add the formulation of a new theory and aesthetic philosophy — a proposal, a plea to developers and players alike to entertain the idea of a decomposing game, a transforming, variable landscape where player empowerment is reduced, and non-player agents are entitled to the acquisition and termination of resources and progression avenues. 

Works Cited

Pagliarulo, Emil. “Developer quotes – Offsite Archives III”. Waiting for Oblivion, January 29, 2006. (Accessed April 17, 2010). http://www.waiting4oblivion.com/developer_quotes_offsite3.html#msfd53

Sanjeev, Anirudh. “Radiant AI – This is awesome and freaky”. Thought Outflux, October 23, 2006. (Accessed April 17, 2010). http://anirudhsanjeev.org/radiant-ai-this-is-awesome-and-freaky/

Abbott, Michael. “Fallout 3”. Pop Matters, February 13, 2009. (Accessed April 19, 2010). http://www.popmatters.com/pm/review/70438-fallout-3/

Juul, Jesper. "The Open and the Closed: Game of emergence and games of progression". Jesper Juul, 2002. (Accessed April 24, 2010). http://www.jesperjuul.net/text/openandtheclosed.html

“GNS Theory.” Wikipedia: The Free Encyclopedia. Wikimedia Foundation, Inc. February 26, 2010. (Accessed April 24, 2010). http://en.wikipedia.org/wiki/GNS_Theory