The amazing adventures of RoboPlayer

[hyphz]

Introducing the new Inuwatabetai RoboPlayer 2000!  Yes, this fine device can sit in as a player in YOUR RPG sessions.  Its operation is perfectly simple:

-  Using GM's descriptions and game rules as input, construct a "database" of potential actions and how they relate to each other;

-  Using some information about a 'character' to play - either internally generated or supplied to it - together with a 'premise', 'social contract', 'group focus' and similiar, generate a "program" of judgment rules that return score values for potential action;

-  When time comes to act in the game, look at all possible/appropriate actions in the database, judge each action using the rules in the program, and emit the one that produces the highest score.

The questions:

-  Can RoboPlayer play this way?

-  Can RoboPlayer be a good player this way?  Could he be one if he had a really good program?

-  What can a human player do that RoboPlayer can't?  (One answer is "play author/director stance games", I think.  So, what can a human player do in an Actor stance game that RoboPlayer can't?)

-  Is it possible for a GM *not* to railroad RoboPlayer?

-  Is it possible for a GM *not* to railroad RoboPlayer if the GM knows RoboPlayer's program and the program contains no ambiguity?  How much of the program can the GM know before this problem arises?

[ethan_greer]

Apart from the obvious logistical problems of a Roboplayer implementation, I have a couple questions.

Are you serious?

I'm assuming this would be a program for use on a laptop or desktop computer?

As far as the GM railroading issue, careful attention would need to be paid to what info was fed to Roboplayer.  Leave something out (the location of a door, for example) and you run the risk of erroneously limiting Roboplayer's options.

Could Roboplayer be a good player?  Yeah, I think so.  Probably a little too good, in that it could tend to be fairly predictable without a sophisticated AI.

[Le Joueur]

I don't think the Roboplayer will be of much use.  It may seem like a great convenience, and a lot of people might buy it, but in the end I see them left out or still sitting in the closet.  Why?

Initiative.

What you've defined is the responsive ideal.  This player is totally reactive, not proactive in any way.  I watch humanity sometimes with an 'outside perspective' and one thing I can tell you is they want interaction, not just reaction.  Oh it's easy to find people who get a little too much interruption willing to extol the virtues of a simply reactive partner, but left alone with them, it pales.

The subtle thing at work is that, over the long term, no matter how attentive, obsequious, and sycophant-like, the Roboplayer will slowly give the impression that they just don't care.  Longer and it actually starts seeming insulting.  Everybody needs 'somebody on their side,' but in the complete absence of criticism, the 'disconnectedness' will read patronizingly.

Ron's friend Mario isn't like this, he takes an active role in supporting; I believe he'll jump in with...well, we call them the 'straight lines.'  Delivering 'straight lines' is a difficult art, but with timing and the ability to 'see where things are going' they are an invaluable tool towards heightening play.  Roboplayer doesn't do this, he merely reacts, does his part, fills in, smoothes things over; ultimately his play will 'flatten' the tenor of the game.

I really wanted to get at some of this in my Emergent Technique regarding Player Ambitions for Scattershot.  What you've described for Roboplayer is more or less one step below Passive Commitment and I'd argue that Roboplayer isn't really playing the game at all.  If anything he's 'playing the crowd.'

And if that isn't patronizing, I'm not sure what is.

Fang Langford

p. s. Actually, Roboplayer should be equally good at Author and Director Stance; stance isn't tied to initiative, but sphere of control.  There's no reason Roboplayer couldn't emit responses in control of larger components of the game.  I should think even Narrativism is possible given that 'the proper response' clearly includes the 'right' way to address the Edwardian Premise.

[Marco]

-  Can RoboPlayer play this way?

Not only can they--I think I have some in my group!

-Marco
[ More serious response: the railroading issue is a very good question that gets to the heart of the matter. I say "no" because (IMO) railroading implies disfunction which doesn't exist with robo-player. ]

[hyphz]

Ok, just to be clear here, I'm not talking about a real implementation of anything like this (the level of parsing required would be far beyond anything we have now, to start with).  Hence the posting in the theory group.  

What you've defined is the responsive ideal.  This player is totally reactive, not proactive in any way.  I watch humanity sometimes with an 'outside perspective' and one thing I can tell you is they want interaction, not just reaction.  Oh it's easy to find people who get a little too much interruption willing to extol the virtues of a simply reactive partner, but left alone with them, it pales.

I wonder why you assume this.  Nothing in RoboPlayer's system above bars the program from being one that gives proactive actions high scores.  Yes, it only performs actions that have been indicated as possible by the GM, but in a non-authorial game all players must do that, because in these games, unless the GM's described it or it can be reasonably entailed from the GM's description, it isn't in the world.

The subtle thing at work is that, over the long term, no matter how attentive, obsequious, and sycophant-like, the Roboplayer will slowly give the impression that they just don't care.  Longer and it actually starts seeming insulting.  Everybody needs 'somebody on their side,' but in the complete absence of criticism, the 'disconnectedness' will read patronizingly.

Again, why should it be obsequious and sycophantic?  You can just change the judgment program to return lower scores for these actions.

[hyphz]

As far as the GM railroading issue, careful attention would need to be paid to what info was fed to Roboplayer.  Leave something out (the location of a door, for example) and you run the risk of erroneously limiting Roboplayer's options.

That's true.  But are you then implying that you could CORRECTLY limit his options?

My real point with the railroading question was that, if you (the GM) know RoboPlayer's program and know it contains no ambiguity, then when you describe something in the world you can predict 100% how RoboPlayer will react - and thus, by choosing your description, 100% determine RoboPlayer's PC's behaviour.

And this gets interesting when you consider that, yes, some real gamers do play like RoboPlayer.  (In fact, some would argue that ALL gamers do but some have better programs than others.)

[ethan_greer]

To "correctly" limit Roboplayer's options would probably require some sort of program that will feed it info that has been painstakingly compiled by the GM beforehand.  Which would pretty much make Roboplayer useless in a regular tabletop game in which other humans are involved.  When it comes down to it, Roboplayer practically exists in various forms in various CRPGs with varying complexity.

[Marco]

I wonder why you assume this.  Nothing in RoboPlayer's system above bars the program from being one that gives proactive actions high scores.  Yes, it only performs actions that have been indicated as possible by the GM, but in a non-authorial game all players must do that, because in these games, unless the GM's described it or it can be reasonably entailed from the GM's description, it isn't in the world.

This is the sticky bit. In the real world I have tons of options. My most obvious options are often not the best (the reason creative problem solving is highly valued). If the program evidences creativity, it *is* a player (might as well be human). And (IMO), probably a good one.

If it doesn't then it's as valuable as a non-imaginative player is. If the GM must be explicit about options, no matter how good an evaluator it is, it isn't playing the same game I am:

The only time I've seen GM's explicitly spell out options is in very specific situations (you have a group of wittnesses to question, the GM asks which you question first--but you're free to question none of them if you have another idea). Or in very minor situations: "The waiter says 'smoking or non-smoking."

-Marco

[Thierry Michel]

Can RoboPlayer play this way?

Probably, assuming such a database is possible ( a big assumption).

Can RoboPlayer be a good player this way?

A successful player, yes (assuming the above). What's a good player ?

What can a human player do that RoboPlayer can't?

Enjoy the game.

Is it possible for a GM *not* to railroad RoboPlayer if the GM knows  RoboPlayer's program and the program contains no ambiguity?

If the program is suficiently complex why not ?

[Walt Freitag]

When I first saw the "Random Dungeon Generation" tables in the AD&D1e Dungeon Master's Guide, my first impulse was to try to create the corresponding tables for a "random player" ideally suited to explore the randomly generated dungeons. But the purpose was satirical, not an attempt to design an actual functional script.

Good thing, too, because it's harder than it looks, even for a constrained and stylized universe like randomly generated dungeons.

The short answer to the question "what can a human player do in an Actor stance game that RoboPlayer can't," assuming unlimited development resources but only presently commercially standard programming techniques (no Media Lab R&D stuff), is apply common sense to situations. This affects everything from RP's ability to understand what is being said to its ability to formulate a reasonable plan of action. For example: "Johnny saw a puppy in a pet shop window. He wanted it." Consider for a moment how much of one's "common sense" knowledge of culture, human nature, and the world in general is required to understand that the "it" in that sentence refers to the puppy and not to the window.

So, does RoboPlayer know how to tell a highwayman from a sheriff? That if one is looking for water, finding ice is probably also good? That when people say that something "just kills them" they usually don't mean it literally?

Sure, for every specific example I can throw out, you can respond "we'll just include that knowledge in the program." And some AI projects are based on doing just that. It turns out that the number of individual facts needed is in the millions, possibly the tens of millions -- and we're not talking about obscure facts about ancient Chinese history, we're talking about things like fire is hot.

Suppose you create a sufficiently ccomplete knowledge base, along with logic engines sufficiently powerful to access and manipulate the data, create plans, compose sentences, and so forth. At that point, "knowing RoboPlayer's program" has long since become useless as far as predicting RoboPlayer's behavior is concerned. Even a program a few lines long can exhibit behavior too complex to predict by (as far as anyone knows) any means short of executing the program step by step by hand -- or just running it.

At the extreme scenarios involving extremely complex and capable programs, your questions merge with central (and unanswered) questions of AI, psychology, and philosophy. For example, your question about railroading, applied to human brains as physical systems, turns into "does free will really exist?"

- Walt

[EDITED to correct a typo that made it sound as if Johnny might be trying to shoot the puppy (perhaps for Satan?)]

[Ron Edwards]

Hyphz,

Either I am confused, or I'm not and everyone else is. So I need your help.

If I'm reading you correctly, Roboplayer is a construct/idea you are using to represent how some people really role-play, and you are raising the issue to discuss this mode of real-person play, not some actual or presumed program/implementation.

That would mean ...

1. All discussion of whether a literal, mechanical Roboplayer is real or possible or desirable is irrelevant.

2. We should discuss whether some GMs really would prefer that their players use Roboplayer-strategy, and whether such play is perfectly fun and enjoyable for all involved. (I suspect that it often is.)

3. We should discuss types of play in which Roboplayer-strategy literally breaks the Social Contract, and why.

So, am I right, or am I the confused one? Help please.

Best,
Ron

[hyphz]

1. All discussion of whether a literal, mechanical Roboplayer is real or possible or desirable is irrelevant.
2. We should discuss whether some GMs really would prefer that their players use Roboplayer-strategy, and whether such play is perfectly fun and enjoyable for all involved. (I suspect that it often is.)
3. We should discuss types of play in which Roboplayer-strategy literally breaks the Social Contract, and why.

Ron, you are not confused at all, this is exactly the sort of thing I meant.

The other issue I was thinking of was: suppose that we accept the proposition that if RoboPlayer's program is fully known to the GM and is unambiguous then the GM is simply determining the player's actions with their descriptions.  This would imply that, for this (undesirable) determinism not to occur in a situation with a real player, it would be necessary for a) the player not to use the RoboPlayer strategy, OR b) for the program not to be known to the GM, or to be ambiguous.  

In fact, a) is harder than it seems.  It isn't enough for the player not to use the RoboPlayer strategy; they must use a strategy which cannot be modelled as an instance of the RoboPlayer strategy with a specific program.  This is difficult and I'm not sure if it's even possible.  Therefore, is b) the case - that a game *must* feature, and to some extent depend on - factors about the PLAYERS or PCs (not just the adventure or NPCs) which are unknown or ambiguous, IC or OOC, in order to offer the players true choices?  (And success/failure by die roll doesn't count?)

This actually makes sense to me based on some experiences I've had, but it's a little hard to describe for the moment (sorry)..

[xiombarg]

So, then, to rephrase the original post: Is a player like RoboPlayer desirable?

I don't see why not. A lot of players do this. Arguably, it's a very immersive player who acts in this fashion. As long as they're having fun, what's not to like? All they're doing is extrapolating conservatively from past experience.

Is this player being railroaded? Only insomuch as learning from experience and acting conservatively based on that experience can be considered railroading. The player's actions aren't being forced by the GM, any more so than it would be to use this strategy in, say, a corporate environment.

In fact, I could argue that RoboPlayer's strategy makes excellent business sense, and might even be "creative" in a way -- the action with the highest value might surprise you, and tell you a lot about your own style of play that you're not admitting. In that respect, RoboPlayer could be a sort of "conscience" for the group -- what are we really doing? Well, look how RoboPlayer acts.

Sure, in some sense the GM is "determining" the player's actions -- but a lot depends on how the "score values" are determined. If they're determined by someone else other than the GM, particularly if they're supplied and/or tweaked by other "live" players, there's no saying that GM knoledge of the strategy will detract from his enjoyment of RoboPlayer's play -- which I think is the real question here.

[Mike Holmes]

All Role-play is interactive. Railroading means that the GM is reducing the Player's options in order to get somewhere that the GM desires. Merely saying that there is oxygen and that the character can't breathe if he doesn't have any is not railroading.

IOW, you can railroad roboplayer, or you can just set him in an objective environment and let him go. I call this Open Play now, but with Roboplayer as a real robot, the phrase I formerly used, pinball play, would describe things perfectly. The world is the machine, Roboplayer is the ball. Hit the plunger and off he goes.

Consider Walt's statement about the random character in the random dungeon.

Mike

[clehrich]

Just to be a bit concrete about this, I recall John Kim describing to me a two-part tournament module (AD&D) he once designed.  In the first part, the players were a team of skilled rangers wiping out a "nest" of Kobolds, essentially a very simple tactical dungeon-crawl.

In part two, the players were the leaders of the Kobold community, and they had to stop the NPC rangers.  The thing is, the NPCs were controlled by a flowchart, not an actual brain.  They worked by simplistic SWAT-style mechanics (for example, they don't move as a group, but leapfrog tactical distance to tactical distance --- "Clear! Go go go go!").  So the issue was how to use your actual intelligence (and your map of the dungeon) to waste these rangers who are, after all, only heavily armed robots.

I think that provided we're talking about imposing pretty severe constraints on this "roboplayer" that amount to a sophisticated flowchart and priority list, this is not going to be the player everyone else looks back on as the fun guy to be around.  If you just keep loosening the constraints (no, but we could just add this other subroutine) then you're talking about something that will pass a Turing test, in which case what's the point of the discussion?