Featured Blog

Thinking About Elizabeth: Part 1

Thoughts on achieving the feel of multiplayer co-op with a fully AI controlled companion

This post is a speculation on how Elizabeth, the fully AI controlled companion in Irrational’s upcoming BioShock Infinite, might be designed in ways that create new player experiences. It’s really just an exploration of AI companion design in general, but Elizabeth was chosen as a useful sort of hypothetical case study.

The following thought experiment, then, is based on these premises: A) a new kind of experience will be reached when co-op play with a fully AI controlled companion has the same level of camaraderie as multiplayer gaming with a friend; B) whereas AI companions are the weakest aspects of most games of similar genre, it is the very fact that the player has no direct control over Elizabeth which makes realizing Premise A possible.

BioShock was praised for bringing about ground-breaking ways to tell stories in games, but I think Infinite has the potential in Elizabeth to do more—much, much more. So, let’s see if we can extract some of what makes co-op feel like what it is, and picture how that might apply to an AI controlled companion.

1. Job Relegation

High level cooperative experiences demand job relegation. If we look at experiences like WoW or versus Left 4 Dead, we find an explicit recognition that certain classes are meant to do certain jobs, and the other classes are meant to be enablers and co-contributors to that job.

If, however, the only job is DPS (co-op campaign L4D, for instance, as opposed to versus L4D), it becomes a competition for who gets the most kills, the most pipes thrown out, burns all the gas cans, etc. Obviously, this is bad for our purposes—we don’t want the player to have to compete with, or feel like his agency has been robbed by, a bot.

Job relegation therefore allows each team member to anticipate what the other team members are going to do, which allows forward planning and deliberate cooperation. But more importantly, it lets players enjoy supporting experiences other than simply just DPS all the time. Notably, properly enabling another is half the fun of co-op.

2. Communication

Job relegation is about providing other players the agency space (or opportunity space) required to do their job. For example, if we think of health as agency space, threat management gives other classes the space to do their tasks; upkeep of the tank’s health allows him to continue generating threat; etc. Similarly, a Charger that knocks down a group of Survivors gives a Spitter the opportunity to inflict real damage; a well placed boom blinds the Survivors and gives a chance for a Jockey sneak a Survivor away; one incapacitation makes another incap more likely, which further facilitates another; etc.

The point of communication, then, is to alert each other as to when and how a player is about to create such spaces. Which is to say, communication largely comes down to knowing how a map is played, and reading/working with how other players are playing the map.

The reason why Elizabeth is a useful example here is her obvious prominence—her status as an equal partner—in Infinite's gameplay (at least, from what can be gathered). The vigors/nostrums system and the fact that Elizabeth’s powers evolve over time (i.e., possibly tailors to player tendencies) make real, player driven job relegation possible, and we can definitely expect highly tuned, opportunity rich maps for job relegation to take advantage of.

Plus, BioShock’s focus on narration through gameplay bodes well for the stated goals. A friend you play multiplayer with will never have his life threatened, but Elizabeth will. Obviously, the emotional connection with a friend will be different in substance and longevity, but, if the kinds of experiences found in BioShock are anything to go by, perhaps not very drastically in quality.

So the main difficulties lie in figuring out how to create such connections so that Elizabeth doesn’t feel like a dead weight that holds the player back (that is, like a game-long escort mission, or frustrating artificial inflation of difficulty), and in making “communication” really work.


Interestingly, the things that make job relegation so enjoyable are the same things that make relationships between married couples sustainable—at least according to this research on meaningful marriages and relationships.

Individuals use a relationship to accumulate knowledge and experiences, a process called “self-expansion.” Research shows that the more self-expansion people experience from their partner, the more committed and satisfied they are in the relationship.

From the same article, this process was descriptively called “the ‘Michelangelo effect,’ referring to the manner in which close partners ‘sculpt’ each other in ways that help each of them attain valued goals.”

In other words, self-expansion means that for assets that we are missing in ourselves, we look towards our partners to provide or help obtain (you might even read the term as “expanding the self through another”). It was found that the closer the emotional connection between a couple was, the longer it took for them to distinguish between characteristics belonging to oneself and those belonging to one’s partner.

All of this is to say that a significant portion of the multiplayer experience will be successfully recreated if the player begins to experience self-expansion through an AI controlled companion. (Think of the relationship between Joker and EDI in Mass Effect 2. Of course, we’re not going to have real AI, but that’s arguably not necessary, especially within a controlled environment and narrative.)

BioShock Infinite, as far as we know from available published information, lacks the respec ability provided by Gene Banks in BioShock. What this means is that the player’s ability to try out different classes or take on different roles within the same game is restricted. Subsequently, it also means that player tendencies expressed through the course of the game are likely to remain, which gives the opportunity to collect a statistics based database from which to tailor Elizabeth’s own evolution and reward player decisions over time.

The important thing to draw from the above is that as long as Elizabeth develops in a fashion which works closely with the player’s own development/investment pattern, the player will experience a substantially real relationship with her (of course, assuming that the character of Elizabeth is well written and well programmed—at least, as much as can be expected from a BioShock game). Elizabeth is then no longer just a detached and unrelated hanger-on that has nothing to do with the player’s build, she is now an integral asset, an enabler and co-conspirator in more ways than just “here is a thundercloud, now use it to strike lightning, regardless of whether or not that was your goal.”

Get Out of Jail Free Problem

Perhaps counter-intuitively, the thing that will make self-expansion seem more organic is to reduce the number of deus ex machina type situations, such as the raincloud and metal ball in the gameplay demo for BioShock Infinite. It appears from the demo that Elizabeth only brings out those skills after DeWitt has already been using their matching power. But the fact remains that they significantly reduce the challenge of the situation, which is not necessarily what the player actually wants.

We might also call this “the Fatman Problem.” When the player has access to something as powerful as the Fatman in Fallout 3, the player tends to refrain from using it. (This player tendency was actually discussed in the first episode of Irrational’s regular interview series, which featured Mr. Todd Howard.)

The tendency arises from two factors: 1) A game is less fun when there is less challenge to be won (that is, when there is less agency to be built), especially when the player is already acclimated to/expects a higher level of challenge (this is in the same way that a deus ex machina resolution to a story often feels cheap and unsatisfactory). 2) Game balance demands that overpowered devices such as the Fatman necessarily be restricted in number of uses, which reduces their presence in regular play. So much so that the Fatman does not figure into most player builds at all, which means that they become an intrusion into something the player has been fighting to establish throughout the course of the game.

Going back to the raincloud example in the gameplay demo, we can see that DeWitt is diligently using electro bolt/shotgun to try to stem the tide of incoming mobs. Elizabeth then appears and provides an essentially overpowered solution based on electro bolt. This means (unless the player was already expecting and waiting for that cloud) if the player wants to get his fill of the challenge, the player actually needs to stop what he is doing and switch out from electro bolt to something else.

If Elizabeth’s use of powers then has a physical or other type of cost (as is widely speculated), not only do we have a situation where the player’s game has been interrupted, we now have wasted resources on top of that, which together would seriously break any sort of self-expansion projected onto Elizabeth. A similar situation in multiplayer games would either be a complete breakdown in communication, or a failed depletion of opportunity space.

Actually, the get out of jail free card is not particularly a useful parallel to understand the above situation. Because the get out of jail free card relies partially on chance for its occurrence, without a hard limit on how many times it can occur, it has an illusion of being part of regular play.

With Elizabeth, we can’t resort to this type of randomization because this prevents reliable job relegation and communicative coordination (if the player herds a bunch of mobs expecting a raincloud only for Elizabeth not to use it, that becomes a serious breach of trust). Moreover, simply limiting the number of times she can use the raincloud by resource restriction has the effect of pulling it out of regular play (like the Fatman). So the trick is to figure out how to sensitively incorporate such abilities into regular play as an integral part of the player’s build without breaking game balance.

Part 2 will examine possible solutions to the problems explored in this post, from how we might implement job relegation through player investment, to how properly handling player initiated changes in pacing can help solve communication difficulties.


Explore the
Advertise with
Follow us

Game Developer Job Board

Game Developer


Explore the

Game Developer Job Board

Browse open positions across the game industry or recruit new talent for your studio

Advertise with

Game Developer

Engage game professionals and drive sales using an array of Game Developer media solutions to meet your objectives.

Learn More
Follow us


Follow us @gamedevdotcom to stay up-to-date with the latest news & insider information about events & more