AI: the smart way to go
AI: the smart way to go
Aug 26, 2005
Artificial intelligence -- or "AI" -- is the Rodney Dangerfield of video game design. It gets no respect when it's working great, as when it contributed to "Halo 2" and "Half-Life 2" becoming the hugely successful games that they are. But when game characters start walking into walls, everyone knows to blame the AI.
According to John Funge, high-quality graphics may be what attracts a player to a game, but it's the AI and the gameplay that holds their attention. Funge is a co-founder and one of the leading scientists at iKuni Inc., a Palo, Alto, CA-based startup focusing on AI effects for video games. He's a previous contributor to research labs at Sony Computer Entertainment America and Intel, and is the author of "Artificial Intelligence For Computer Games: An Introduction". In a chat with Hollywood Reporter columnist Paul Hyman, Funge talks about why designers ought to think about AI when turning their IP into games, and how AI has the potential to become the new driving force behind video game innovation.
The Hollywood Reporter: "Artificial intelligence" (AI) refers to the programming that acts as the brain for everything that isn't controlled by the human player, is that correct? So that includes what gamers call "NPCs" -- or "non-player characters" -- which are those characters that are controlled by the computer?
John Funge: Exactly. Among other things, including the equivalent of what would be the director in a movie. Someone -- or something -- has to decide where to place the "camera" in a game to enable the player to see what's going on. For example, in a football game like "Madden," unless the camera is pointing at the ball, the player will never be able to see it to catch it. That's actually one of the hardest AI problems there is in games, trying to decide what's interesting and where to place the camera. Generally you'd want to point at the ball, but that's not always true. If I'm going to pass to my colleague on the other side of the field, you want the camera to pan out to see the person you're going to pass to, to give a sense of anticipation. Basically, all the problems you have in a real sporting event with placing cameras you have in a computer game of that sport.
In a first-person shooter, like "Doom," typically you'd want the camera to be behind the player's character. But then, you may want to swivel the camera as you come around the corner so you can see the character coming around the corner. All of these rules have to be written in advance because there's obviously nobody actually directing the camera in the game. If programmed incorrectly, you wind up with bad AI which is when the camera ends up behind an obstacle so you can't see what's going on. Everybody knows that's bad AI; what they don't recognize is great AI.
THR: What are some examples of great AI.
Funge: Well, I noticed in "Halo" and "Halo 2" that when I shot somebody and injured them, they'd turn and run away. You'd say that's pretty obvious, but it's also realistic. It gives the illusion of intelligence. (Click here to view "Halo" developer Bungie's white paper on the game's AI). In some games, a character will just keep coming at you until you shoot them so many times that they die.
That's very similar to what impressed people about "Half Life" when it first came out. If you attacked a creature, it would take cover and wouldn't just stupidly walk towards you. It would react to you and do vaguely sensible things.
Or, a few years ago, Lionhead Studios developed a strategy game for Electronic Arts called "Black & White: Creature Isle" in which the goal was to create a creature and teach it how to behave. The AI was up front and center in that game.
One of the best things about a game called "Seaman" on the Dreamcast was that, at one point, the creature would "look" at your memory card to see what other games saves you had, and then it would make a comment like "Oh, I see you like to play shoot-'me-ups." Can you imagine how spooky that was? Definitely a great moment in AI.
Currently, there's a game called "Forza Motorsport" that uses a specific AI -- Microsoft calls it Drivatar technology -- to enable you to train your driver to use a specific driving strategy. That is really pushing the AI envelope.
And in MS Game Studio's "Fable," you'd have conversations with characters who seemed to remember what happened earlier in the game. If you kicked a chicken in the beginning of gameplay, for example, later on, people in the crowd would call you a "chicken kicker." A player may not even realize that's the result of good AI, but it certainly adds to the realism of the game.
THR: When people talk about the next generation of console games, they always have a laundry list of what's going to make them better than today's games: more intense graphics, better sound and music, more realistic action. But you don't hear much about AI. Is that because no one expects AI to improve in next-gen games?
Funge: No, that's not true at all. It's just that, for whatever reason, AI hasn't really captured the public's imagination to the same extent as graphics. Partly because it's a harder thing to appreciate. The graphics are very obvious to the player. Interestingly enough, that is changing. Nowadays, it's difficult to see the difference between this year's game graphics and last year's, so you're getting into an area of diminishing returns. Which is why I think AI will start to become more and more important as a way to differentiate games.
THR: By making them more realistic?
Funge: Absolutely. In the old days, all the characters in a game like "Space Invaders" would just move according to very predefined rules; they weren't reacting to the player at all. But, today, in a game like "Grand Theft Auto," the game world is a big open space filled with so much content. In order for there to be a convincing experience, you want AI that will make the NPCs seem more intelligent so they can be like actors who are improvising in response to the player's actions.
THR: Is that something we're going to see in the next-generation of games?
Funge: Yes, you'll see better AI out of necessity; as the game worlds become more complicated, it becomes harder for the programmers to anticipate every single possible situation that will arise in these enormous open worlds. The NPCs will have to become more intelligent in order to handle whatever the gamer decides to do.
THR: Would you expect that gamers will recognize this as better AI?
Funge: Most likely not. It's sort of a Catch-22. The better the AI gets, the fewer people will recognize it as AI. They'll just see it as realism that gives them a more fun experience. Talking about ways to differentiate your games, that will be a prime example.
THR: I'm looking for an analogy here. Would you say this is sort of like a Broadway play in which you don't notice that the lighting is adding to the enjoyment of the show unless it's so bad that you can't see the actors?
Funge: Exactly. And in the film industry, everyone knows the importance of great cinematographers, for example. But they aren't the ones who get the headlines the day after the Oscars. That's because the public is far more interested in, say, Brad Pitt winning the award for best actor. I think the next step will be that the games industry appreciates the importance of AI, but how far that will spread outside the industry is anybody's guess. For the foreseeable future, I suspect AI will only be visible by the effect that it has on the game and not for the AI itself.
THR: In your book, you say that AI has the potential to become the new driving force behind video game innovation. How so?
Funge: Once players become excited about AI and the effects that AI can give a game, they're going to demand more and better AI. Obviously the problem of creating intelligent agents with AI is incredibly difficult, but as we keep on making improvements, I think that's going to help drive the games industry and drive new types of games.
THR: Let's talk about what's driving the games industry today, which is games based on licensed IP like movies. How will AI make that type of game even better?
Funge: I understand that "The Godfather" will have an open world like "Grand Theft Auto," and that definitely presents a challenge for the AI, because you want the player to be able to go anywhere he wants with the NPCs reacting appropriately. Perhaps you try to run someone down with a car. Well, you want that person to remember that later in the game so that, when he sees you, he tries to take revenge. What you don't want is to go somewhere and have some sort of interaction and, later on, you come back and none of the NPCs have any memory of that interaction, for example. Incorporating AI into their games will definitely be a future challenge for designers, and many intelligent people are working on that today -- determining what are the AI elements that work the best and what new kinds of games will be possible with new AI capabilities.
THR: Is there any one game whose AI really knocks you out?
Funge: In terms of impressive AI content, there's a new concept game called "Facade" that's really more interactive graphical fiction. But you can download it for free and try it for yourself. You are a guest in the house of a couple who is having marital troubles. You interact with them by typing in questions; it's a free-form dialogue in which you can ask them anything you want. But depending on how the conversation goes, different scenarios unfold. And you can either get them to stay together or even convince one of them to run off with you. When I tried it, I kept getting thrown out of the house. It's a great example of what you can do if you take the time and effort to create lots of AI content. And there were only two people working on that game. Imagine what a big company could do with a whole team! That should indicate to game designers that they can do lots more interesting things than have characters walk into rooms and shoot everyone.
THR: So this is the direction AI will take gaming?
Funge: Possibly, among others. For example, games based on films are extremely popular these days. Imagine designing games in which the characters not only look like movie characters but also behave like them. And I don't mean just in animated cut scenes but in response to the gamer and his actions. For example, you might create a game from the movie "Reservoir Dogs" in which each character has its own distinct personality. So Mr. Blonde, for example, would always have a strong tendency to just start shooting and, if you were robbing a bank, you might want to leave him in the car because his shooting could alert the police too early. But then he might get annoyed with you if you keep him in the car, and that could be dangerous, too. Or Mr. Pink might want to wait until he was really pushed before acting. So you'd have to manage their personalities as you undertake various missions. I think that kind of depth of character and personality are, within reason, something we will hopefully start to see more of in next-gen games.
Paul "The Game Master" Hyman was the editor-in-chief of CMP Media's GamePower. He's covered the games industry for over a dozen years. His columns for The Reporter run exclusively on the Web site.
According to John Funge, high-quality graphics may be what attracts a player to a game, but it's the AI and the gameplay that holds their attention. Funge is a co-founder and one of the leading scientists at iKuni Inc., a Palo, Alto, CA-based startup focusing on AI effects for video games. He's a previous contributor to research labs at Sony Computer Entertainment America and Intel, and is the author of "Artificial Intelligence For Computer Games: An Introduction". In a chat with Hollywood Reporter columnist Paul Hyman, Funge talks about why designers ought to think about AI when turning their IP into games, and how AI has the potential to become the new driving force behind video game innovation.
The Hollywood Reporter: "Artificial intelligence" (AI) refers to the programming that acts as the brain for everything that isn't controlled by the human player, is that correct? So that includes what gamers call "NPCs" -- or "non-player characters" -- which are those characters that are controlled by the computer?
John Funge: Exactly. Among other things, including the equivalent of what would be the director in a movie. Someone -- or something -- has to decide where to place the "camera" in a game to enable the player to see what's going on. For example, in a football game like "Madden," unless the camera is pointing at the ball, the player will never be able to see it to catch it. That's actually one of the hardest AI problems there is in games, trying to decide what's interesting and where to place the camera. Generally you'd want to point at the ball, but that's not always true. If I'm going to pass to my colleague on the other side of the field, you want the camera to pan out to see the person you're going to pass to, to give a sense of anticipation. Basically, all the problems you have in a real sporting event with placing cameras you have in a computer game of that sport.
In a first-person shooter, like "Doom," typically you'd want the camera to be behind the player's character. But then, you may want to swivel the camera as you come around the corner so you can see the character coming around the corner. All of these rules have to be written in advance because there's obviously nobody actually directing the camera in the game. If programmed incorrectly, you wind up with bad AI which is when the camera ends up behind an obstacle so you can't see what's going on. Everybody knows that's bad AI; what they don't recognize is great AI.
THR: What are some examples of great AI.
Funge: Well, I noticed in "Halo" and "Halo 2" that when I shot somebody and injured them, they'd turn and run away. You'd say that's pretty obvious, but it's also realistic. It gives the illusion of intelligence. (Click here to view "Halo" developer Bungie's white paper on the game's AI). In some games, a character will just keep coming at you until you shoot them so many times that they die.
That's very similar to what impressed people about "Half Life" when it first came out. If you attacked a creature, it would take cover and wouldn't just stupidly walk towards you. It would react to you and do vaguely sensible things.
Or, a few years ago, Lionhead Studios developed a strategy game for Electronic Arts called "Black & White: Creature Isle" in which the goal was to create a creature and teach it how to behave. The AI was up front and center in that game.
One of the best things about a game called "Seaman" on the Dreamcast was that, at one point, the creature would "look" at your memory card to see what other games saves you had, and then it would make a comment like "Oh, I see you like to play shoot-'me-ups." Can you imagine how spooky that was? Definitely a great moment in AI.
Currently, there's a game called "Forza Motorsport" that uses a specific AI -- Microsoft calls it Drivatar technology -- to enable you to train your driver to use a specific driving strategy. That is really pushing the AI envelope.
And in MS Game Studio's "Fable," you'd have conversations with characters who seemed to remember what happened earlier in the game. If you kicked a chicken in the beginning of gameplay, for example, later on, people in the crowd would call you a "chicken kicker." A player may not even realize that's the result of good AI, but it certainly adds to the realism of the game.
THR: When people talk about the next generation of console games, they always have a laundry list of what's going to make them better than today's games: more intense graphics, better sound and music, more realistic action. But you don't hear much about AI. Is that because no one expects AI to improve in next-gen games?
Funge: No, that's not true at all. It's just that, for whatever reason, AI hasn't really captured the public's imagination to the same extent as graphics. Partly because it's a harder thing to appreciate. The graphics are very obvious to the player. Interestingly enough, that is changing. Nowadays, it's difficult to see the difference between this year's game graphics and last year's, so you're getting into an area of diminishing returns. Which is why I think AI will start to become more and more important as a way to differentiate games.
THR: By making them more realistic?
Funge: Absolutely. In the old days, all the characters in a game like "Space Invaders" would just move according to very predefined rules; they weren't reacting to the player at all. But, today, in a game like "Grand Theft Auto," the game world is a big open space filled with so much content. In order for there to be a convincing experience, you want AI that will make the NPCs seem more intelligent so they can be like actors who are improvising in response to the player's actions.
THR: Is that something we're going to see in the next-generation of games?
Funge: Yes, you'll see better AI out of necessity; as the game worlds become more complicated, it becomes harder for the programmers to anticipate every single possible situation that will arise in these enormous open worlds. The NPCs will have to become more intelligent in order to handle whatever the gamer decides to do.
THR: Would you expect that gamers will recognize this as better AI?
Funge: Most likely not. It's sort of a Catch-22. The better the AI gets, the fewer people will recognize it as AI. They'll just see it as realism that gives them a more fun experience. Talking about ways to differentiate your games, that will be a prime example.
THR: I'm looking for an analogy here. Would you say this is sort of like a Broadway play in which you don't notice that the lighting is adding to the enjoyment of the show unless it's so bad that you can't see the actors?
Funge: Exactly. And in the film industry, everyone knows the importance of great cinematographers, for example. But they aren't the ones who get the headlines the day after the Oscars. That's because the public is far more interested in, say, Brad Pitt winning the award for best actor. I think the next step will be that the games industry appreciates the importance of AI, but how far that will spread outside the industry is anybody's guess. For the foreseeable future, I suspect AI will only be visible by the effect that it has on the game and not for the AI itself.
THR: In your book, you say that AI has the potential to become the new driving force behind video game innovation. How so?
Funge: Once players become excited about AI and the effects that AI can give a game, they're going to demand more and better AI. Obviously the problem of creating intelligent agents with AI is incredibly difficult, but as we keep on making improvements, I think that's going to help drive the games industry and drive new types of games.
THR: Let's talk about what's driving the games industry today, which is games based on licensed IP like movies. How will AI make that type of game even better?
Funge: I understand that "The Godfather" will have an open world like "Grand Theft Auto," and that definitely presents a challenge for the AI, because you want the player to be able to go anywhere he wants with the NPCs reacting appropriately. Perhaps you try to run someone down with a car. Well, you want that person to remember that later in the game so that, when he sees you, he tries to take revenge. What you don't want is to go somewhere and have some sort of interaction and, later on, you come back and none of the NPCs have any memory of that interaction, for example. Incorporating AI into their games will definitely be a future challenge for designers, and many intelligent people are working on that today -- determining what are the AI elements that work the best and what new kinds of games will be possible with new AI capabilities.
THR: Is there any one game whose AI really knocks you out?
Funge: In terms of impressive AI content, there's a new concept game called "Facade" that's really more interactive graphical fiction. But you can download it for free and try it for yourself. You are a guest in the house of a couple who is having marital troubles. You interact with them by typing in questions; it's a free-form dialogue in which you can ask them anything you want. But depending on how the conversation goes, different scenarios unfold. And you can either get them to stay together or even convince one of them to run off with you. When I tried it, I kept getting thrown out of the house. It's a great example of what you can do if you take the time and effort to create lots of AI content. And there were only two people working on that game. Imagine what a big company could do with a whole team! That should indicate to game designers that they can do lots more interesting things than have characters walk into rooms and shoot everyone.
THR: So this is the direction AI will take gaming?
Funge: Possibly, among others. For example, games based on films are extremely popular these days. Imagine designing games in which the characters not only look like movie characters but also behave like them. And I don't mean just in animated cut scenes but in response to the gamer and his actions. For example, you might create a game from the movie "Reservoir Dogs" in which each character has its own distinct personality. So Mr. Blonde, for example, would always have a strong tendency to just start shooting and, if you were robbing a bank, you might want to leave him in the car because his shooting could alert the police too early. But then he might get annoyed with you if you keep him in the car, and that could be dangerous, too. Or Mr. Pink might want to wait until he was really pushed before acting. So you'd have to manage their personalities as you undertake various missions. I think that kind of depth of character and personality are, within reason, something we will hopefully start to see more of in next-gen games.
Paul "The Game Master" Hyman was the editor-in-chief of CMP Media's GamePower. He's covered the games industry for over a dozen years. His columns for The Reporter run exclusively on the Web site.
Share on LinkedIn








