This piece is written as a supplement to a talk of the same name, delivered at the Norwich Gaming Festival in April 2015. The full talk can be seen in the video above.
As you would expect, the adoption of AI within games is a big talking point here on AIandGames.com, we’re often interested in what techniques are implemented in commercial games and the subsequent performance of these systems. However, one topic we seldom discuss on this site is what do the designers seek to achieve in the adoption of AI within video games? How are AI systems adopted and what (potentially new) experiences do they help to provide? It’s a really interesting question, given that AI can allow for some really exciting experiences when used in a sensible fashion. However, many game players (and the larger genreal public) are often unaware of these innovations. Indeed, the drive to make this apparent to people is part of the mission of this site. One of the biggest stumbling blocks in highlighting the impact AI can have on video games, is that gaming as a medium – or at least how it is popularised through marketing – is one with an arguably unhealthy obsession with violence.
A History of Violence
Indeed video games are renowned for being violent in some fashion or another, but there is a larger purpose behind much of the functional aspects of these mechanics. This is not an argument over the potential (if not negligible) impact of the glorification of violence in video games on the minds of consumers. Nor is this an argument of whether video game violence should be curbed/censored/restricted etc. Rather, what we’re interested in is why do video games frequently associate with violence and what does this mean for AI systems?
Let’s wind the clock back to the early days of game development and focus specifically on titles launched in the early 1980’s. One of the big transitions during the Atari-era, was to move away from the notion of requiring two players at the same time. Many games began to focus on single-player experiences, with a variety of games popular at that time, notably games such as Yar’s Revenge (1981), Missile Command (1980), Pac-Man (1980), Centipede (1980) and Space Invaders (1978), having enemy characters that are controlled by the software itself. While still rather simple in many respects, these are nonetheless AI implementations. What all of these implementations share with each other, is that beyond all else, their function is to be destroyed by the player. Their subsequent destruction is means by which to achieve an increase in score. This achieves means of player empowerment: whereby the player not only feels rewarded for their efforts due to a sense of progression, but ultimately they will feel good about themselves in the process.
Now player empowerment can prove to be an interesting problem in itself. There are all sorts of implications that these player empowerment frameworks can cause, particularly if the themes, artistic style and design of games lead towards specific political or sociological philosophies. However, the main issue we have with this, is that this form of empowerment often leaves AI implementations trapped with a rather contrived design space.
The video provided above serves to prove a point: in each video game shown the AI is provided in the game world to act as the players folly. This can result in AI that is shot, stabbed, decapitated, set on fire, hit with shells, arrows to the knee and all sorts. The level of detail and whether this violence is glorified will obviously vary, but the key thing is that video games have a bad habit of using AI solely for NPCs. This of course fails to acknowledge some of the amazing work achieved in implementing these AI systems, such as the use of behaviour trees in the Halo series.
Now this argument is a little cynical and also, thankfully, far from the truth in a lot of modern video game development. Fortunately over the last decade, we have seen an increasing change in how AI systems are used for the purposes of player engagement.
An Optimistic Outlook
While using AI as NPC’s is still largely the norm, there are some really exciting innovations taking place in both the AAA sector, as well as the indie scene. To name just some of the games that come to mind that adopt AI in an interesting way:
- Left 4 Dead: The director system manages the gameplay experiences by adding and relieving the pressure on players throughout their run.
- Forza: Use of the ‘drivatar’ system allows for players to play their friends offline, by replicating their friends behaviour through player modelling.
- Killer Instinct: The shadow system allows players to train a clone of their own gameplay, akin to that of Forza’s drivatar.
- Alien: Isolation: A game where the AI, the xenomorph, is the core of the experience, reversing genre conventions by having the first-person control be an instrument of fear, rather than power.
- Titanfall: A first-person shooter game that drops non-player characters (NPCs) into online matches. This allows not for novice players to attain a sense of empowerment even if they struggle against human players, but also add strategic value in attaining score. In addition, players can use the AI to control their ‘titan’ robots in combat rather than pilot the machine themselves, leading to new tactical opportunities.
- Borderlands: A game about a players quest for their favourite gun. The procedural generation system behind the series can generate of 14 million different types of weapon. The true quest for the player is not to complete the story, but find the <1% of weapons that make this game even more exciting.
- No Man’s Sky: A procedurally generated universe full of galaxies and planets to explore.
- Spelunky: A platforming roguelike whose intense difficulty and PCG level generation adds an air of mystery to an otherwise rather charming game. Undoubtedly one of the main reasons for its continued popularity.
These are of course but a fraction of some of the really exciting innovations now happening in game development. The key element that all of these games share is that they exploit AI in an interesting way beyond traditional means. The exceptions to this are Alien: Isolation and Titanfall. Alien: Isolation deliberately subverts the traditional first-person gameplay model; with the NPC alien the dominant force and the player is submissive to its power. Meanwhile, Titanfall subverts multiplayer design convention by adding single-player gameplay tropes (and their AI implementations) to enhance the online experience.
AI-Based Game Design Patterns
Much of the argument that drives this piece arose from a discussion on the adoption of AI as a design pattern at Dagstuhl seminar 15051: Artificial and Computational Intelligence in Games – Integration. One of the issues being discussed is how often AI is adopted where it is placed at the forefront of the experience for players. In addition, what is the larger function of that AI within the system?
The discussion group at Dagstuhl was a collection of AI games researchers, many of whom with experience in game development ranging from commercial titles to small works of their own. Through these experiences, a collection of patterns shown in the table above were established (Treanor et al., 2015). While several of these are common place, some of these patterns are rather foreign or alien to many designers, with few examples of their actual adoption. We wouldn’t hasten to argue why that is (too difficult, doesn’t work etc.), but it does mean there are still opportunities found within them.
Some examples of actually exploring these ideas can be found in (Cook et al., 2015), where a subset of the discussion group decided to make games inspired by these ideas. Naturally, they’re not fully polished (or even complete) games in some cases, but they are a reflection of what can be achieved by adopting these design patterns as a guiding principle behind AI adoption in games. This includes Contrabot: a game where you must decide on a cypher that allows for a smuggler to recognise and grab packages of importance. All the while attempting to avoid AI inspector in the middle, who becomes increasingly suspicious of the codes stamped on the packages.
The potential for future game innovations is massive for AI provided we move away from the conventions we have established and maintained for so long. The real power of AI adopted in games is that it can illicit the most human aspects within ourselves: joy, hilarity, exhilaration, fear; these can all manifest courtesy of a system that ironically, is incapable of these feats. That’s the power of AI-driven game design: systems that reveal the most human aspects of play.
Treanor, M., Zook, A., Eladhari, M.P., Togelius, J., Smith, G., Cook, M., Thompson, T., Magerko, B., Levine, J. and Smith, A. (2015) “AI-Based Game Design Patterns” Foundation of Digital Games Conference, June 2015.
Cook, M., Eladhari, M.P., Smith, A., Smith, G., Thompson, T., Togelius, J. and Zook, A. (2015) “AI-Based Games: Contrabot and What Did You Do?” Foundation of Digital Games Conference, Playable Demo Track, June 2015.