The Future of AI NPCs

In this article, game designer Sande Chen describes what happens when players have a supercomputer for a mentor and what this development means for games in the future.

Showcased in last week's session "Playing Medical Minecraft with IBM Watson" at the Games For Change Festival, Medical Minecraft is more than just a student mod of Minecraft transformed into a first-person shooter with disease enemies like malaria.  Players get a strong ally in the form of a NPC powered by IBM Watson. IBM Watson is the computer system known mostly for beating its human competitors on "Jeopardy!"

IBM Watson By Clockready via Wikimedia Commons

Using AI for NPCs has been done before, most notably in the game, Facade, whereby the player can interact with a couple on the brink of divorce. YouTube is littered with examples of players messing with Facade.  The funniest tales are when the players refuse to role-play the domestic scenario and instead type in outlandish inputs such as impending global thermal nuclear war or a grisly car accident.  The couple continues on with the conversation without acknowledging the dire circumstances.

In Medical Minecraft, players can quell their curiosity about various diseases.  IBM Watson playing the knowledgeable mentor character answers the questions not by doing a massive brute force search of the Internet, but by relying on its training in determining what is the most likely answer.  Moreover, players do not have to confine themselves to certain phrases or words as in previous computer games.  They can carry on a Q&A conversation with IBM Watson.  It isn't like the classic game ELIZA at all, where the computer played a psychotherapist and tended to repeat words back to the player. 

What does this mean for the future of games?  Instead of a limited script from a NPC or a database the player must search through for information, the player can interact with a NPC in natural language.  Although the example above is limited to medical knowledge, it very likely could be expanded for history or other sciences.  And that's truly exciting stuff.

Sande Chen is a writer and game designer whose work has spanned 10 years in the industry. Her credits include 1999 IGF winner Terminus, 2007 PC RPG of the Year The Witcher, and Wizard 101. She is one of the founding members of the IGDA Game Design SIG.

Lost in Game Space

In this article, game designer Sande Chen describes certain game deficiencies that lead to player frustration and how better storytelling may provide the solution for one of the problems.

In the past few years, I have served as a judge for multiple game festivals and competitions. There are several reasons why some games don't make the cut.  Beyond the technical complications of not being able to get a game running, I find a similar failing may be in not having a strong enough tutorial, i.e. a player shouldn't be confused about how to play a game.  Struggling with controls or an interface is just frustrating and not the experience you want for a first-time player.  I recall there's an infamous transcript of a WWII Online player griping that it was easier flying a plane in World War II than trying to do the same in a game!

In other games, I find a beautiful world that I would like to explore, but I am directionless as to what would be my goal.  Free-form exploration and self-direction are fine as long as there's enough interesting content to support it indefinitely.  In most cases, due to production costs, this is simply not true.  Therefore, there needs to be a way to guide the player to the more interesting content rather than leaving the player to trod through the same loop of scenery.

A prehistoric storyteller describes a hunt.

Luckily, stories provide context and player motivation.  If I know I have to find a way off the island, then I'm not going to spend my time admiring sparkly fish.  Moreover, human beings crave stories.  Even in prehistoric times, cave dwellers conveyed tales of great hunts.  Stories tell us about ourselves and the human condition.

In this age of game making, it might seem like emergence or AI is the solution, but it's not enough.  Emergent stories could be interesting, but they could also be not interesting.  As Alex Toplansky said at the panel, Writing for Horror Video Games, even in systemic games, "a writer needs to come in and stack the dice."  Dramatic storytelling, whether linear or non-linear, is a crafted experience.   

As for AI, while there have been advances in computer algorithms generating stories, poetry, and news articles, sometimes a human touch is warranted.  To escape the redundancy of randomly generated "Rescue X at location Y" quests, players of the now-defunct The Matrix Online banded together to create an epic storyline that gave their characters more motivation.  While the quests did give the players specific goals to complete, the randomness did not generate an interesting story for players.

What's the lesson here?  As I have written in my article, Towards More Meaningful Games, don't leave your narrative design choices to chance.  Yes, a game still needs to feel open enough to allow for meaningful player choices but that doesn't mean that players should be left confused as to what they ought be doing.



Sande Chen is a writer and game designer whose work has spanned 10 years in the industry. Her credits include 1999 IGF winner Terminus, 2007 PC RPG of the Year The Witcher, and Wizard 101. She is one of the founding members of the IGDA Game Design SIG.

Learning to love the narrative puzzle

In this article, Professor Clara Fernández-Vara argues that narrative puzzles don't need to be convoluted, but rather, they require a more conscientious and responsible game design strategy.

One of the recurring questions in narrative design is how necessary puzzles are, since they seem to be the staple of adventure games since their inception. There are plenty of examples of how adventure games may not need puzzles. Game makers such as Telltale Games or Choice of Games have explored how to engage players through choice design, while indie darlings such as Dear Esther, Gone Home, The Stanley Parable or Kentucky Route Zero show how exploration can also become satisfying gameplay. Some advocate leaving puzzles behind and focusing on problem solving.

Thing is, solving puzzles is problem solving. Puzzles get a bad rap because of years of players facing puzzles that only make sense to their designers, and the infinite patience of players who kept trying random things until they bumped into the hyper-contrived solution. Examples of bad puzzles are legion while, when a good puzzle is good, it is often seamless because it makes sense, so most players do not notice there ever was one. See for example how game critic John Walker mentions how the puzzles are perfunctory in Act 1 of Broken Age, pointing out that he may have expected something more complex but that the puzzles are there to help the narrative. Half of his review of Act 2, however, is a tirade against the tediousness of the convoluted puzzles.

We don't need to banish puzzles from our games. They can help us learn more about our world, set up character, and get the player to be in a specific place when we need it. What we need is conscientious and responsible puzzle design, understanding the range of what works and what doesn't.

Puzzles are problems that require a solution (hence invoking "problem solving" as an alternative may not be that useful), and most of the time there's only one valid solution. The issue with narrative puzzles is that they often have only one way to get to that solution, even if the player can think of multiple ways to tackle the challenge. In game design terms, we designers can follow three strategies:

• Offering players more than one way to solve the puzzle: since puzzles in general have more than one way to achieve the solution - there are different strategies to solve mathematical and logic puzzles, jigsaws or crosswords. So why don't we try to provide more than one way to get to the solution? Games like Deus Ex or more recently Dishonored are famous for taking this design approach. Thing is, the challenges that the player faces tend to be physical problems; challenges that involve human behavior and psychology, for example, are out of the question. Puzzles that are more grounded in the narrative that have multiple paths are still a challenge. A glorious example of how multiple paths can backfire is Zack McKracken and the Alien Mindbenders, the one LucasArts game nobody remembers because it does offer several ways to complete it, but it does not tell the player that if you solve a puzzle a certain way, other paths will be locked out.

• Setting up the puzzle so that solving it seems an adequate challenge: the key to satisfying puzzles is that the player achieves insight at the moment of solving them. Some puzzles feel invisible because their solution feels natural to the player; there is a problem but the solution seems logical. This was the case with most of the puzzles of Act 1 of Broken Age, for example. There is set of questions that will help us set up each puzzle: 
•  How can the player tell there is a puzzle that they need to tackle? 
•  What information does the player need to solve the puzzle? 
•  If it's not information that relates to everyday life (such as opening a door, or doing an economic transaction), where in the world is that information? 
•  Is the information available in one place or several? How close is it to the puzzle itself? Can the player revisit the information? The more pieces of information the player needs, and the further they may be from the puzzle (whether it is in terms of space or time), the more difficult the puzzle is. 
• How can the player tell that they have found the wrong solution to the problem? Does the player get more information to solve it?  
• How can the player tell they have solved the puzzle?  
• What does solving the puzzle mean in the game? Does the player learn about the world / story? Does the player obtain something out of solving the puzzle? (I've talked about this topic before at length; you can watch one of my presentations here.

• Design an esoteric puzzle for hardcore puzzle solvers: there are games that are geared towards hardcore players who want their puzzles to be extremely challenging. If you use the checklist above, it turns out that hardcore puzzles are missing one of these elements, from letting the player know that there is a puzzle, to requiring esoteric knowledge to solve it, for example. The line between a badly designed puzzle and a hardcore one is very thin; the definition depends on the audience. The rule of thumb is that the logic of the puzzle must still be there. You still get insight if you check the solution. That's why omitting elements can be okay, because it's up to the player to fill the gaps. When the puzzle feels random, unjustified, or the challenge consists of reading the designer's mind, then the puzzle falls apart. Examples of games with difficult narrative puzzles are The Hitchhiker's Guide to the Galaxy, Maniac Mansion or, more recently, Device 6. The puzzles in Broken Age Act 2 aim at being hardcore (perhaps because some players thought the puzzles in Act 1 were too easy), but often the logic seems absent: at one point, I had to go to a location in order to trigger a cut scene that allowed me to obtain an object, which is the kind of random access to information that infuriates players.

There may be other design options that technology may facilitate in the future. For example in the games Symon and Stranded in Singapore, we used procedural generation, so that whenever you started a game the puzzles would be different. Our method of generating the puzzles was not particularly complicated, so there was still one single way to solve each puzzle; perhaps in the future we can have AI that can create and check multiple paths. What worked in the design of Symon (a bit less so in Stranded) was that the game design focused on creating a system of relations between objects, rather than just specific puzzles. Players could not look up a walkthrough that gave details on each puzzle. Instead, they had to figure out the relationships between objects according to specific qualities, thus showing the potential to understand the world as a whole, rather than puzzle to puzzle. What I want to highlight here is not that future technology will solve our design problems - although it will probably help - but we that we need to change our paradigms and the way that we design narrative games. We need to shake off our nostalgia a bit and start pushing for new design patters - a sentiment that I'm not alone in sharing.

Puzzles in narrative games are just a part of the design lexicon, and we need to expand the vocabulary of narrative in games. The possibilities of choice and exploration are now gaining popularity - although they have been around for a while - and the future of narrative in games looks bright. But let us not completely dismiss narrative puzzles yet. We should banish badly set up puzzles with unsatisfying convolutedness that do not help the narrative. The solution is realizing puzzles also require game design.

Clara Fernández-Vara is a game scholar, designer and writer, and she is an Associate Arts Professor at the NYU Game Center. Her area of expertise is narrative in games and how it can integrate with game design, which she has explored both in games for research and in the commercial sphere. Her first book, Introduction to Game Analysis has been published by Routledge.

IGDA Webinar: Interactive Fiction - An AI Based Overview

In this video, writer Emily Short reviews different ways to provide player choice in interactive fiction and how AI can be part of this process.