The Fundamental Question - Rationality computer game design

by Kaj_Sotala8 min read13th Feb 201368 comments

58

Gaming (videogames/tabletop)Distillation & PedagogyEpistemology
Frontpage

I sometimes go around saying that the fundamental question of rationality is Why do you believe what you believe?

-- Eliezer in Quantum Non-Realism

I was much impressed when they finally came out with a PC version of DragonBox, and I got around to testing it on some children I knew. Two kids, one of them four and the other eight years old, ended up blazing through several levels of solving first-degree equations while having a lot of fun doing so, even though they didn't know what it was that they were doing. That made me think that there has to be some way of making a computer game that would similarly teach rationality skills at the 5-second level. Some game where you would actually be forced to learn useful skills if you wanted to make progress.

After playing around with some ideas, I hit upon the notion of making a game centered around the Fundamental Question. I'm not sure whether this can be made to work, but it seems to have promise. The basic idea: you are required to figure out the solution to various mysteries by collecting various kinds of evidence. Some of the sources of evidence will be more reliable than others. In order to hit upon the correct solution, you need to consider where each piece of evidence came from, and whether you can rely on it.

Gameplay example

Now, let's go into a little more detail. Let's suppose that the game has a character called Bob. Bob tells you that tomorrow, eight o'clock, there will be an assassination attempt on Market Square. The fact that Bob has told you this is evidence for the claim being true, so the game automatically records the fact that you have such a piece of evidence, and that it came from Bob.

(Click on the pictures in case you don't see them properly.)

But how does Bob know that? You ask, and it turns out that Alice told him. So next, you go and ask Alice. Alice is confused and says that she never said anything about any assassination attempt: she just said that something big is going to be happen at the Market Square at that time, she heard it from the Mayor. The game records two new pieces of evidence: Alice's claim of something big happening at the Market Square tomorrow (which she heard from the Mayor), and her story of what she actually told Bob. Guess that Bob isn't a very reliable source of evidence: he has a tendency to come up with fancy invented details.

Or is he? After all, your sole knowledge about Bob being unreliable is that Alice claims she never said what Bob says she said. But maybe Alice has a grudge against Bob, and is intentionally out to make everyone disbelieve him. Maybe it's Alice who's unreliable. The evidence that you have is compatible with both hypotheses. At this point, you don't have enough information to decide between them, but the game lets you experiment with setting either of them as "true" and seeing the implications of this on your belief network. Or maybe they're both true - Bob is generally unreliable, and Alice is out to discredit him. That's another possibility that you might want to consider. In any case, the claim that there will be an assassination tomorrow isn't looking very likely at the moment.

Actually, having the possibility for somebody lying should probably be a pretty late-game thing, as it makes your belief network a lot more complicated, and I'm not sure whether this thing should display numerical probabilities at all. Instead of having to juggle the hypotheses of "Alice lied" and "Bob exaggerates things", the game should probably just record the fact that "Bob exaggerates things". But I spent a bunch of time making these pictures, and they do illustrate some of the general principles involved, so I'll just use them for now.

Game basics

So, to repeat the basic premise of the game, in slightly more words this time around: your task is to figure out something, and in order to do so, you need to collect different pieces of evidence. As you do so, the game generates a belief network showing the origin and history of the various pieces of evidence that you've gathered. That much is done automatically. But often, the evidence that you've gathered is compatible with many different hypotheses. In those situations, you can experiment with different ways of various hypotheses being true or false, and the game will automatically propagate the consequences of that hypothetical through your belief network, helping you decide what angle you should explore next.

Of course, people don't always remember the source of their knowledge, or they might just appeal to personal experiences. Or they might lie about the sources, though that will only happen at the more advanced levels.

As you proceed in the game, you will also be given access to more advanced tools that you can use for making hypothetical manipulations to the belief network. For example, it may happen that many different characters say that armies of vampire bats tend to move about at full moon. Since you hear that information from many different sources, it seems reliable. But then you find out that they all heard it from a nature documentary on TV that aired a few weeks back. This is reflected in your belief graph, as the game modifies it to show that all of those supposedly independent sources can actually be tracked back to a single one. That considerably reduces the reliability of the information.

But maybe you were already suspecting that the sources might not be independent? In that case, it would have been nice if the belief graph interface would let you postulate this beforehand, and see how big of an effect it would make on the plausibility of the different hypotheses if they were in fact reliant on each other. Once your character learns the right skills, it becomes possible to also add new hypothetical connections to the belief graph, and see how this would influence your beliefs. That will further help you decide what possibilities to explore and verify.

Because you can't explore every possible eventuality. There's a time limit: after a certain amount of moves, a bomb will go off, the aliens will invade, or whatever.

The various characters are also more nuanced than just "reliable" or "not reliable". As you collect information about the various characters, you'll figure out their mindware, motivations, and biases. Somebody might be really reliable most of the time, but have strong biases when it comes to politics, for example. Others are out to defame others, or invent fancy details to all the stories. If you talk to somebody you don't have any knowledge about yet, you can set a prior on the extent that you rely on their information, based on your experiences with other people.

You also have another source of evidence: your own intuitions and experience. As you get into various situations, a source of evidence that's labeled simply "your brain" will provide various gut feelings and impressions about things. The claim that Alice presented doesn't seem to make sense. Bob feels reliable. You could persuade Carol to help you if you just said this one thing. But in what situations, and for what things, can you rely on your own brain? What are your own biases and problems? If you have a strong sense of having heard something at some point, but can't remember where it was, are you any more reliable than anyone else who can't remember the source of their information? You'll need to figure all of that out.

As the game progresses to higher levels, your own efforts will prove insufficient for analyzing all the necessary information. You'll have to recruit a group of reliable allies, who you can trust to analyze some of the information on their own and report the results to you accurately. Of course, in order to make better decisions, they'll need you to tell them your conclusions as well. Be sure not to report as true things that you aren't really sure about, or they will end up drawing the wrong conclusions and focusing on the wrong possibilities. But you do need to condense your report somewhat: you can't just communicate your entire belief network to them.

Hopefully, all of this should lead to player learning on a gut level things like:

  • Consider the origin of your knowledge: Obvious.
  • Visualizing degrees of uncertainty: In addition to giving you a numerical estimate about the probability of something, the game also color-codes the various probabilities and shows the amount of probability mass associated with your various beliefs.
  • Considering whether different sources really are independent: Some sources which seem independent won't actually be that, and some which seem dependent on each other won't be.
  • Value of information: Given all the evidence you have so far, if you found out X, exactly how much would it change your currently existing beliefs? You can test this and find out, and then decide whether it's worth finding out.
  • Seek disconfirmation: A lot of things that seem true really aren't, and acting on flawed information can cost you.
  • Prefer simpler theories: Complex, detailed hypotheses are more likely to be wrong in this game as well.
  • Common biases: Ideally, the list of biases that various characters have is derived from existing psychological research on the topic. Some biases are really common, others are more rare.
  • Epistemic hygiene: Pass off wrong information to your allies, and it'll cost you.
  • Seek to update your beliefs: The game will automatically update your belief network... to some extent. But it's still possible for you to assign mutually exclusive events probabilities that sum to more than 1, or otherwise have conflicting or incoherent beliefs. The game will mark these with a warning sign, and it's up to you to decide whether this particular inconsistency needs to be resolved or not.
  • Etc etc.

Design considerations

It's not enough for the game to be educational: if somebody downloads the game because it teaches rationality skills, that's great, but we want people to also play it because it's fun. Some principles that help ensure that, as well as its general utility as an educational aid, include:

  • Provide both short- and medium-term feedback: Ideally, there should be plenty of hints for how to find out the truth about something by investigating just one more thing: then the player can find out whether your guess was correct. It's no fun if the player has to work through fifty decisions before finding out whether they made the right move: they should get constant immediate feedback. At the same time, the player's decisions should be building up to a larger goal, with uncertainty about the overall goal keeping them interested.
  • Don't overwhelm the player: In a game like this, it would be easy to throw a million contradictory pieces of evidence at the player, forcing them to go through countless of sources of evidence and possible interactions and have no clue of what they should be doing. But the game should be manageable. Even if it looks like there is a huge messy network of countless pieces of contradictory evidence, it should be possible to find the connections which reveal the network to be relatively simple after all. (This is not strictly realistic, but necessary for making the game playable.)
  • Introduce new gameplay concepts gradually: Closely related to the previous item. Don't start out with making the player deal with every single gameplay concept at once. Instead, start them out in a trusted and safe environment where everyone is basically reliable, and then begin gradually introducing new things that they need to take into account.
  • No tedium: A game is a series of interesting decisions. The game should never force the player to do anything uninteresting or tedious. Did Alice tell Bob something? No need to write that down, the game keeps automatic track of it. From the evidence that has been gathered so far, is it completely obvious what hypothesis is going to be right? Let the player mark that as something that will be taken for granted and move on.
  • No glued-on tasks: A sign of a bad educational game is that the educational component is glued on to the game (or vice versa). Answer this exam question correctly, and you'll get to play a fun action level! There should be none of that - the educational component should be an indistinguishable part of the game play.
  • Achievement, not fake achievement: Related to the previous point. It would be easy to make a game that wore the attire of rationality, and which used concepts like "probability theory", and then when your character leveled up he would get better probability attacks or whatever. And you'd feel great about your character learning cool stuff, while you yourself learned nothing. The game must genuinely require the player to actually learn new skills in order to get further.
  • Emotionally compelling: The game should not be just an abstract intellectual exercise, but have an emotionally compelling story as well. Your choices should feel like they matter, and characters should be in risk of dying if you make the wrong decisions.
  • Teach true things: Hopefully, the players should take the things that they've learned from the game and apply them to their daily lives. That means that we have a responsibility not to teach them things which aren't actually true.
  • Replayable: Practice makes perfect. At least part of the game world needs to be randomly generated, so that the game can be replayed without a risk of it becoming boring because the player has memorized the whole belief network.

What next?

What you've just read is a very high-level design, and a quite incomplete one at that: I've spoken on the need to have "an emotionally compelling story", but said nothing about the story or the setting. This should probably be something like a spy or detective story, because that's thematically appropriate for a game which is about managing information; and it might be best to have it in a fantasy setting, so that you can question the widely-accepted truths of that setting without needing to get on anyone's toes by questioning widely-accepted truths of our society.

But there's still a lot of work that remains to be done with regard to things like what exactly does the belief network look like, what kinds of evidence can there be, how does one make all of this actually be fun, and so on. I mentioned the need to have both short- and medium-term feedback, but I'm not sure of how that could be achieved, or whether this design lets you achieve it at all. And I don't even know whether the game should show explicit probabilities.

And having a design isn't enough: the whole thing needs to be implemented as well, preferably while it's still being designed in order to take advantage of agile development techniques. Make a prototype, find some unsuspecting testers, spring it on them, revise. And then there are the graphics and music, things for which I have no competence for working on.

I'll probably be working on this in my spare time - I've been playing with the idea of going to the field of educational games at some point, and want the design and programming experience. If anyone feels like they could and would want to contribute to the project, let me know.

EDIT: Great to see that there's interest! I've created a mailing list for discussing the game. It's probably easiest to have the initial discussion here, and then shift the discussion to the list.

58

68 comments, sorted by Highlighting new comments since Today at 9:29 AM
New Comment

Actually, having the possibility for somebody lying should probably be a pretty late-game thing, as it makes your belief network a lot more complicated, and I'm not sure whether this thing should display numerical probabilities at all. Instead of having to juggle the hypotheses of "Alice lied" and "Bob exaggerates things"

Perhaps the protagonist should start out being trained as an investigator on an island where the inhabitants have strong taboos against lying. Early cases are scenarios crafted for your benefit as a trainee. Eventually, you graduate to investigating real cases, and later on, you leave the island.

ETA: Possible alternative to using numerical probabilities. The player's findings are written in colored boxes, and the color of the box represents how likely the player thinks that individual belief is (ex. Blue=Practical certainty, Green=Very likely, Yellow=Somewhat likely, Orange= Unlikely, Red= Practically falsified.) Early on, players will be taught to make beliefs that are consequences of other beliefs at most as likely as the beliefs they follow from, and to assign lower likelihood to the overlap of multiple conditions, so for instance, if they have a box that follows as a consequence of three yellow boxes, they'll want to color that box orange.

I'd be happy to contribute writing to the project, if you're interested in having me.

Further suggestion: Players should learn about the distinction between accuracy and calibration. There should occasionally be scenarios where the real solution is not something the information available to you singles out as probable. Players should learn that banking on an unlikely solution is never a good bet, but highly probable solutions are still only probable rather than certain.

Players' performance would be tracked, not just in terms of their ability to get the right answers, but their ability to be right about how often they're right.

I disagree that there should be situations where the less likely situation is correct only becaus it is less likely ( as a pre-programmed result). The likelihood of an event occurring in the game should be a result of your acquired evidence and only 100% certainty can exist when there is enough concrete evidence supporting the outcome. Within the game it should be possible for the true outcome to receive a high probability. Your idea however is essential in situations where the probability of events are very close. For example in a situation with 5 outcomes where all their probabilities are 15-30% it wouldn't and shouldn't be obvious.

100% isn't a probability, and while it's often feasible to approach it in practice, it's also often not, because the evidence necessary to reach that degree of confidence simply isn't available.

If you reach 95% confidence, you should still be wrong 5% of the time.

If the player learns that they can collect all the available evidence and be right 100% of the time in the game, and then finds that they simply can't do that in real life, they may be disillusioned with the applicability of the general techniques of the game.

Players should learn that banking on an unlikely solution is never a good bet, but highly probable solutions are still only probable rather than certain.

Banking on an unlikely solution is a good bet if and only if you get odds more favorable than the solution is unlikely.

It's a question both of payoff and of likeliness.

You could even run that as a game mechanic - you have limited investigative time, and certain leads have more promising-looking rewards or more likely to end up with something that helps you out, and then after investigating the lead you update your beliefs and try something else.

A good skill for this is seeing what leads you can cheaply eliminate. It would also put into focus the costs of having too many low-value leads.

I'm concerned that separating reward from effectiveness in getting the right answers would make the game too complicated, and dilute the message.

Err, I was trying to go for a "you are rewarded for getting the right answer quickly" sort of deal.

I think it might be counterproductive to time players, because it's likely that time pressure would encourage players to use system 1 reasoning, and develop quick but sloppy heuristics for proceeding in the game.

Possibly the game could introduce some Time Attack elements late on, once the players have mastered everything else, but speaking as someone who very rarely enjoys time constraints in games, I'd prefer if it were optional.

I though that by "quickly", "with few pieces of evidence" was actually meant - the less, the better. Still, you can always get more evidence than necessary.

Yeah, "quickly" got overloaded by both "how long the player takes to make decisions" and "how many units of in-game time have elapsed". Perhaps "fewer turns used" is a better way to phrase it.

Yes, I was thinking of something similar as the island: that the early levels would be somewhere where, for whatever reason, you could trust everyone to be truthful. And I was also considering something similar as the boxes!

I'd be happy to have you on the team. :) Please, do join the mailing list that I just set up.

Worth mentioning: I haven't played it, but parts of this idea sound like they're already implemented in Phoenix Wright.

I played one of the early Phoenix Wright games, and it was fun, but also mostly a pretty ordinary point-and-click style adventure that just happened to be taking place in a detective context. But the later games in the series might be different?

I was thinking the same thing.

This sounds like the perfect game.

Perhaps look at the design of other puzzle games for guidance. One of the most successful, Professor Layton, is excellent, and introduced better thinking habits to me (Kahneman's famous €1.10 Bat and Ball problem was brought to me here). It made sure that if you wanted to get the most out of the game, you had to check your thinking - I got the bat and the ball problem right, because I'd learnt to check carefully. It wasn't, however, designed with rationality in mind. It could've been a lot better.

Please post updates about how the game develops :-)

Damn, I don't have a DS. Might have to ask some friend to lend theirs. :)

I'll do my best to post updates.

If you do a little googling + torrenting, it can be free on a computer ;)

My first thought is that one of the big departures Dragon Box has, is that it removed the central concept (math), and just kept the patterns (balancing equations).

Second, the problem with probabilities is that if the best you can do is 90% within the rules, then you'll be failing 10% of the time, and people don't like that.

The second problem seems easily solved: come up with a situation where you run the experiment N times and reap the rewards of your strategy. Prisoner's Dilemma tournaments, for example. Given we're dealing with probabilities, ANY sort of gambling, and just play out the results based on statistical odds (or, at higher levels, larger samples. You could even have the elite level be single-instances :))

The game still focuses on gathering information and learning about the system, and it probably needs to have a fairly concrete UI and mechanic for measuring and representing that. But it removes the social and cultural aspecs, and focuses it on a very simple, easy to understand system.

I'm not sure as to specifics, but a game where the goal is simply to learn how to play the game seems interesting, and then there's a meta-game that is the UI and mechanics that go in to teaching you and ensuring this isn't a ridiculously frustrating black box to deal with :)

Having multiple possible rules and permutations is important, of course. An unchanging tutorial is probably a good place to start, though. If you can write that, and it's interesting and works, you're probably on to something useful :)

DragonBox removing the central concept is both a major strength and weakness - it's fantastic in teaching people how to solve equations and avoids possible math-phobia, but it also says nothing about the reason why the rules are what they are, or even how they could be applied, so that has to be supplemented by other means. In other words, it's primarily a teaching aid, not something that would be sufficient by itself. Which is fine for a topic like math, where people are inevitably going to be taught the other stuff as well... but less so for a game about this kind of rationality, which needs to do a lot more, because people might never learn to apply the principles if the game doesn't suggest them.

If you remove the social and cultural aspects entirely, then the game becomes just that: an abstract exercise which might be fun, but which requires the player to spend a lot of independent effort in order to realize what the real-world significance of it is. For it to be of use, it has to actually put the player in an environment that is sufficiently like the real world that the notion of applying its techniques to real-world situations (e.g. social situations) will occur to the player somewhat automatically. That does make it much harder to design, but also much more useful if we do succeed.

You have a good point about the probabilities, and I like your suggestion of repeated trials. Not sure how to implement it yet, though.

I see your point, but I still think building an abstract base for use as a tutorial and demonstration is probably wise. Social situations are a fairly advanced and complex arena, since you're dealing with both complex probabilities in complex relationships, but also all sorts of culture-specific social skills. Definitely useful to teach, I just think dropping a player straight in to dealing with lying and deception in a social situation is more suited for Phoenix Wright, Ace Attorney :)

I'm not sure as to specifics, but a game where the goal is simply to learn how to play the game seems interesting, and then there's a meta-game that is the UI and mechanics that go in to teaching you and ensuring this isn't a ridiculously frustrating black box to deal with :)

That certainly reminds me of something.

I would like to contribute to this project.

I have limited programming experience as well as some GameMaker experience which I recommend for the first round of prototyping. My free time is limited though and likely to remain VERY limited until late March.

Please join the mailing list. :-)

We should discuss the language used for making the game - I saw people in the earlier game threads recommend Unity, since it makes it easy to develop something for multiple platforms independently.

Wheeeeeeeee!

That page gave me a few new things to look at, such as Haxe, which I've never seen before (despite having played some of the games built on it!) and will be taking a look at.

Past experience with Unity in the developer-friendliness department was not good, especially considering the way it was recommended to me ("Even easier to use than GameMaker, and you'll have a prototype up and running twice as fast!"). There's high chances that it's better than I thought back then or that it has improved considerably, though, considering how many people are using it. I'm quite willing to take another good look at it.

I suggest that likelihood of getting things wrong should be separated from lying. Lying is a very specific claim about motivation, and may lead to complications about why the person chose that particular lie.

Yeah, I agree. Will have to consider in detail how lying is taken into account in the game.

The obligatory copyright stuff

With any project with multiple contributors, it's best to get some clarity on the copyright status of the contributions right from the start, especially if we have hopes of this becoming popular at some point...

Since it's the clearest and safest to explicitly assign all the copyrights to a single legal person, and it would be a bit of a hassle at this point to set up a separate legal entity that one could assign them to, I'm guessing that it would be the easiest if they were assigned to me? So that people wouldn't feel threatened about how their work might be used, the assignment would be on the condition that the work be licensed under an appropriate free license. (As explained e.g. here.)

The GPL is a nice license for code, but not necessarily for fiction - if someone wanted to write a novel based on the events of the game, for example, applying the GPL to that would get a bit weird. On the other hand, something like CreativeCommons-Attribution-ShareAlike is good for fiction, but not so much for code.

So maybe combine the two. I asked about this on my Facebook and Google Plus feeds, and the comments in those places are somewhat split between dual-licensing the whole project with both GPL and CC-BY-SA, or licensing the code with GPL and the art and writing with CC-BY-SA. I don't really know which option would be better, myself, or whether it really even matters. Thoughts?

I am unlikely to contribute to this project, so this is just the bikeshed peanut gallery, but:

Please use MIT or BSD for code, rather than GPL, unless you specifically value the restrictions on reuse placed by the GPL. (And if the latter is the case, I ask you: what advantage do you gain in this scenario from people not using your code?)

Some of my readings also suggest that if there's going to be any occasional or one-off contributors, you might want to settle early all of the important "money" issues and questions.

GPL and CC do prevent quite a few problems in this regard, but won't do anything for people getting surprised or feeling wronged for some reason or another. The usual recommendation is to keep written records and archived copies of everything, including discussions and threads like this one.

I'd agree that for simplicity, just go GPL for the whole thing. I can't see why the license and rights owner couldn't change the entire licensing model at a later date anyway, but I'm no expert. This could also be one of the things people could be uncomfortable with as the project nears completion, though.

For simplicity, I agree that the copyright should be assigned to you. Also I would go for GPL for both the code and the fiction, the less time is lost on those matters the better. If the game comes out in a definite form, then the fiction copyright might become an issue, but for now it doesn't seem fundamental...

Slight tangent: My daughter's got heavily into children's MMORPGs Club Penguin and Bin Weevils and insists on playing them with me ('cos it's more fun that way). Small children, of course, are great to observe, because they show all the cognitive biases all at once, almost unmoderated. And the business model for these things is "addict the kid utterly so they nag their parents into buying a membership" (I am resisting), so they have to get it right. So I learn all sorts of things about user interactions and interface design.

Both these games have occasional quests (usually secret agent missions) in them, where the player has to solve a puzzle interactively, and what you're planning sounds like that sort of thing. Sooo, I suggest you find a child and play one of these games with them - look at the interaction, look at what the kids like, and look at the quests.

The target audience for this is probably going to be teenagers and up, so their gaming might be more useful to observe than that of younger kids, but that's a good suggestion regardless.

Wow! I like this idea a lot. If someone made this game, I'd have a lot of fun playing it. And I know several people who I would recommend this game to, and would probably like this game a lot, were it made. I really hope this game becomes a real thing.

Glad to hear that. :-)

Setting and story

So after a bit of thought, I was thinking that I could resurrect the setting for a fantasy novel that I started writing once, but never got very far in: The City of Light and Fire. The second part of that story opened as follows:

The basin had been a mold, shaping me to look roughly like a human. But I was still far from perfect.

The figures that carried me put me down on a long stone bench, and then left. There were other shapes on both sides of me, other early-stage embryos. I did not yet understand anything.

Time passed, and the lava I was made of grew more solid. My innermost parts were still hot and liquid, but I had a firm outer crust. When I had become hard enough to be worked on, the mason had me brought to him. He studied me for a long time, examining me from every direction and seeking out any imperfections. Whenever he found one, he reached for his hammer. Gradually, he shaped me into a man.

After the mason was done with me, I was taken to the clay maker. The beings carrying me were cautious, for the mason had opened holes from which my innards might spill. I don't know whether they spilled any, but when I reached the clay maker, I was still viable.

He studied my shape, and then molded a layer of lifeclay around me. It was much softer than lava was, and more sensitive to heat. The clay maker filled the pair of holes the mason had made, fashioning there eyes. Below them he made a mouth, and on their sides a pair of ears. The clay was as good in shaping heat as the walls of the towers were, if not better. It collected warmth and funnelled it deep into my core. My eyes had been made with particular care, and it now that I slowly began to see.

From the clay maker, I was carried to the edge of a great hall. They placed me next to the other infants, on a belt of heatstone close to the wall. The stone burned hot, keeping us sated. I rested there, together with the others. We watched and listened to the things happening in the hall, enjoyed the ever-shifting flows of warmth inside the stone. For a long time, we remained still.

As in the original story, the main character in the game would be a creature who had just recently been brought into existence for an unknown purpose. He (she?) comes to existence with some basic skills, such as knowing how to talk and walk, but knows little besides that. His creators clearly have a purpose in mind for him, but don't care to tell him very much about it. Other creatures that have been created in a similar fashion might have clues of just what exactly is going on, as do others who he meets on missions that he is sent on... but he has to figure out what exactly it is that can be reliably inferred from the various claims that the different creatures make.

Advantages: This would instantly set up a mystery (just what is this place? why was he created? what's going on?) to make the player curious, and gathering various clues about his origin, as well as completing different missions that the character was sent on, could serve as short-term goals for the player to pursue. An added benefit of making the main character non-human is that the whole belief network could be diegetic - perhaps connecting the different pieces of evidence would concretely build up a physical belief network inside his body, and by closing his eyes he could somehow see and manipulate that. Such a setting would also allow us to question the beliefs and attitudes that were commonly believed and taken for granted within the setting, without getting into the politics-is-the-mindkiller territory that having a game about questioning commonly-accepted real life beliefs would involve.

Here's an old description of the setting:

I earlier described that as "a New Weird-ish story about a corporation whose different departments are in semi-open war with each other. The main character is a gargoyle created by Product Development to infiltrate Marketing and kidnap some of their harpies, to be pressed into slave labor to boost Product Development's morale." Though actually sirens are a closer match to what I had in mind than harpies - I mixed up the two.

Anyway, it's this weird dark fantasy setting with a huge, black stone monolith in the middle of a city. The Corporation exists within the various halls and caverns of the monolith, and its workers produce and sell various goods to the people in the city. There are also other cities and in them other corporations, which have a bit of an influence in this city as well, but mostly they remain in the background.

I don't actually have very many details worked out yet, mostly just concepts and images. Here are some of them:

In the city around the monolith, there are sentient stone towers. They are filled with hot lava, and they communicate with each other by selectively making their outer shells more or less transparent, shining light around them. They can choose to voluntarily let out a small amount of their lava and let it cool, creating a new being separate from themselves that serves them. Or somebody might break their shell and take some lava by force to create a new being, which is how the story's main character was created. The towers have their own interests, and they often do not look kindly upon the Corporation.

The ultimate leaders of the Corporation are the Owners, who never intervene directly. The Owners are what seem to be clouds of ever-burning gas above the monolith, who communicate their desires through patterns in the fire.

On the roof of the monolith are the Prophets, monsterously shaped philosophers who sit on top of large stone pillars and watch the sky, interpreting the messages of the Owners. They keep making a constant humming sound, which communicates their interpretation of what the Owners want. This is in turn interpreted by the Scribes, who sit at the feet of the pillars and write down what they hear. A constant stream of messengers takes what the Scribes have written down and relays it to the various departments.

When the department heads receive their orders from the Scribes, they too apply their own interpretation, seeking to follow the instructions in the way that best benefits them personally. The original orders being distorted after going through several steps in the chain, the department heads are free to do almost anything they want. It is because of this personal benefit-seeking that the various departments of the Corporation are in a constant conflict with each other.

Information only seems to pass down from the Owners to the Corporation. If information does pass up to the Owners, nobody knows how.

Even the department heads do need to follow some rules, however. In the middle of the monolith there is the Timetable, a wall which shows the past and a possible future. In particular, it shows various things which the Corporation might be able to achieve in the future. If those things are indeed achieved, all is well.

If they are not, cracks begin to appear in the Timetable. This is a bad thing, for the Timetable has been built to seal a rupture to another reality, in which terrible beings live. If the Timetable ever breaks, the beings will burst out and destroy the monolith and everyone who lives in it. Because of this, the department heads cannot let their infighting get so serious that the Timetable won't hold. It is said that the Timetable, and the rupture behind it, were created by the Owners for this very purpose.

The titular Fundamental Question could then also refer to the task of figuring out just what it is that the Owners really want (which could also be the real reason that the character was created).

A disadvantage of this setting pitch is that there might be a risk of the setting becoming too weird and different to effectively relate to, or for people to very naturally transfer the lessons of that setting into real life. But that could avoided by focusing more on the surrounding city, which could be more normal, or something. Thoughts?

[This comment is no longer endorsed by its author]Reply

Hmm, on other thought a more realistic modern setting might after all feel like a better fit for this kind of game. Hmmh.

I have to agree. I really like this setting, but on a gut level it gives me similar vibes to when I played Machinarium, and that would make it difficult to connect with personal experience / outside life.

Yeah, upon consideration, it's probably not that suitable for this.

Playing around with various ideas for the setting and aesthetics, here's a rather different approach.

The game's set in our world, with elements of magical realism: the main character is a fairy-like figure of ambiguous sex and age (goes to school, but that school could conceivably be elementary school, high school or even college, letting the player project their favored interpretation, though this might be a little tricky to pull off). This approach takes the fundamental question considerably more literally: the main character knows things, but does not know why (s)he knows them. At all times, we are shown a set of maybe three beliefs that (s)he has, but cannot remember why. A large part of the game is about finding out what led her/him to come to believe those facts (what was their ultimate causal origin), and whether they should actually be believed. When ever (s)he manages to answer this question for a particular fact, it is replaced with something else.

Possible opening sequence: the game opens to a black screen with some white text.

"What is your name?"

The player enters a name. The text on the screen then disappears, to be replaced by the following:

"My name is [name]. There are many things that I know. For example, I know that X, Y, Z. But I do not know why I know these things."

(alternative, if we can find a voice actor with a clear, somewhat child-like and androgynous voice, the opening narration could be spoken: "My name is [some gender-neutral name]. There are many things that I know. For example, I know that X, Y, Z. But I do not know why I know these things.")

The text then vanishes and the blackness is replaced by a bright light which initially fills the screen with whiteness: it then fades to reveal the character being in their home, the brightness being the sunrise in the window. Icons for the three beliefs that are being investigated flash for a moment at the top of the interface, then stop flashing. Peaceful, ambient music plays in the background. A brief investigation of the house quickly reveals some of the causal history of each belief: maybe a table holds a letter from a friend, in which the friend spoke to the character of the belief. Upon reading it, the character says something like:

"Oh yes, [the name of my friend] told me this: that is why I believe it. But should I believe it?"

At the top of the screen, the icon highlighting the relevant belief flashes again: it shows a line extending back from the icon, to the icon of the friend. For a moment, the interface displays both (and they take up twice as much space as the representations of the other beliefs, which are only one icon each); then the representation of this belief shrinks back to mostly just the icon of the belief, but also showing in miniature the network behind the belief, which at this point only holds one node but will eventually expand. The icon is highlighted with a small bright glow, to indicate that it can be clicked on in order to bring up the whole history graph.

It's time to go see that friend.

A lot of the game would be about rather ordinary events, like going to school and making friends; but the question of "why does [some character] believe what they believe" comes up a lot, and considering it helps the various characters solve a lot of problems. Perhaps, after the early game, it's the player who gets to choose which would be interesting beliefs (hypotheses?) to investigate, instead of the game choosing the active beliefs for them.

This seems like a much better starting point. The first half feels like it risks evoking some unmentionable tropes not suitable at all for the target audience, but the target audience is extremely unlikely to know these tropes unless they're, like me, also the kind of audience that would find this a plus rather than a downside (the two type-of-audience traits correlate - and they're usually referred to with only one common category label).

As for the second half (i.e. once the three belief-icons pop up)... wow. I'm once again positively impressed by the thought you're putting into both the game design and player experience.

Based on this so far, if I had this game in front of me I would be very intrigued and would quite want to play it, though I can't tell how much the aforementioned trope pattern-matching (and past experience relating to those) accounts for that ;)

Really interesting concept.

Now that you've gone this far, the next step is to establish what the goal of the game is (defeat the aliens, investigate the murders, box the AI, identify the pod people, figure out the rules of science/magic in the parallel dimension, find the right sacrifices to really please the gods, etc...)

Once you've got a good goal, then the emotional compulsion, the medium-term feedback, and the story, should come naturally.

Thanks, and yes, indeed. I have an idea in mind for the setting and story already, will be writing about it as soon as I'm finished answering all the comments in this thread. :-)

I also am a writer of code, although not professionally. I have joined the mailing list group thing, even though there seem to be plenty of coders already.

Welcome. :-)

Emily Short, who has done character interaction oriented game things before just announced a new one she has co-developed.

It might be worth digging through her stuff on conversation modeling in interactive fiction.

I'm interested in helping out with programming (which I do professionally, though not games related) and have joined the mailing list.

I'll try to make some quick prototype of the mechanics over the weekend unless anyone beats me to it.

What seems to be taking shape while doing this is the idea of the player having access to a tool, mostly separated from the rest of the game, to help calculate probabilities. It would allow new (possible) facts to be entered with priors, facts split up into conjunctions, dependencies noted between facts and so on. The tool would then calculate probabilities for you while you tweak the calculation. (Possibly color coded or otherwise abstracted if we expect numbers to be seen as scary.)

I'm not even gonna try to do make this into an intuitive or full featured tool right away, but in a final release I would imagine a very polished interface with drag&drop graphs for dependencies, folding to hide irrelevant details etc.

Early on in the game, there would be heavy prompting on exactly how to use the tool while later on the player would be left increasingly on their own in figuring out how to enter statements and facts she encounters.

First of all, is this in line with what other people are envisioning?

Secondly, this seems like something which may actually be useful in real life as well (and I could see the game ending with an encouragement to do just that). Does anyone know of such a tool which already exists? If nothing else, it might be good for salvaging ideas from. Some quick Googling doesn't reveal anything beyond simple tools where you can put in the numbers for a single equation and tools special fitted for a single application.

Edit: Argument Map software seem to be what I was looking for.

Does anyone know of such a tool which already exists? If nothing else, it might be good for salvaging ideas from. Some quick Googling doesn't reveal anything beyond simple tools where you can put in the numbers for a single equation and tools special fitted for a single application.

I've already posted this link in a reply lower down the thread, but here. SMILE seems built explicitly to be used as an API, but I'm not sure I like the way it stores data nor its I/O methods. So far, OpenMarkov with the ProbModelXML are the most attractive implementations, and I think something closer to that (or perhaps use them as API / backend straight-up, depending on language considerations and interfacing issues) would be more straightforward to implement.

As for looking for apps that do this for general use, I've been toying around with this for a bit and it seems quite fun and intuitive to use (well, to someone who knows some bayes anyway).

Great! Welcome to the team. :-)

First of all, is this in line with what other people are envisioning?

Maybe. At this point, I'm not at all sure of what kind of an approach would work best and be the most fun; it's certainly worth trying a lot of different approaches.

I'm not entirely sure of what you mean by "mostly separated from the rest of the game", but I was thinking that it would be pretty strongly integrated in the sense that information that you picked up from the rest of the game would automatically update the belief network. But maybe you meant besides that?

I was supposed to create some rough prototype myself this weekend, but besides some doodling on the backs of old business cards, haven't really had the opportunity.

I meant something like the difference between:

Bob says: "There will be an assassination..." Player's notebook is automatically filled with this information. The player can assign expected probability. Bob says: "Alice told me so" Player's notebook is automatically filled with this information marked as evidence for the previous claim. The probability assigned to this being true will automatically update the assassination claim.

Or what I was considering yesterday:

Bob says: "There will be an assassination..." Player manually writes this into his notebook. Bob says: "Alice told me so" Player manually writes this into his notebook and manually marks it as evidence for the previous claim. The automatic updating would still happen after this has been done. Alternatively, the player might just go ahead and write in a conjunction for "Alice told me so" & "Alice knows what she is talking about" & "Alice tells the truth" instead.

Pro: Learning to extract facts from statements seems like a useful skill to teach. Con: Without letting the game know about the intended meaning of the facts, it would be very hard for it to find and correct faulty reasoning. It might also turn into too much bookkeeping for the player.

I'm leaning more to a middle ground now, were the game presents all facts that are part of a statement, but it is still up to you to connect them to the right place in the graph. We'd have to experiment to find what actually works of course.

I also meant that if we make it a good enough tool, maybe it would be valuable to use entirely independent from the game. If that should be a goal, it would need to be carefully designed for. This will likely introduce conflicting requirements though, so may not be worth it.

I probably won't finish up something demoable today either. I've mostly just been brainstorming on mechanics and the architecture to support them.

Some more random notes from the prototyping:

  • There are beliefs and correlations between beliefs.
  • Beliefs are entered with a prior for how likely they are without any of the given correlations (prior).
  • Correlations are entered with a belief as cause and one as effect and values for probabilityOfEffectGivenCause + probabilityOfEffectGivenNotCause
  • Conjunctions and disjunctions can be expressed as special cases of beliefs.
  • The full complexity should not be introduced all at once.
  • To guide giving probabilities they could be converted to frequencies in time or space. ("So with no evidence, you believe there would be an assasination like this every week?"; "[...]right this hour in one city in the country")
  • My biggest problem is that I have no idea how to actually score a player if he gets to come up with his own probabilities in a fictional world. Maybe the game needs to have some way of explicitly finding out the "right" values for some priors and correlations.

Sounds promising! I'll hopefully have the time to put together a design/prototype of my own tomorrow.

I meant something like the difference between:

Either of those could work, but I'm worried that the steps that the latter option would require would easily make the player feel like she was doing tedious work that could easily have been automated instead. I'm not sure about that, though: getting to enter the data could also feel rewarding. We'll just have to experiment with it.

My biggest problem is that I have no idea how to actually score a player if he gets to come up with his own probabilities in a fictional world. Maybe the game needs to have some way of explicitly finding out the "right" values for some priors and correlations.

Well, if different beliefs have different consequences in the world ("if you believe the assassin is in the bell tower, go there to stop him") and the player is scored on his ability to achieve things in the world, that also implicitly scores him on probabilities that are maximally correct / useful. But this might not be explicit enough, if the player has no clue of what the probabilities should be like and feels like they're just hopelessly flailing around.

I also meant that if we make it a good enough tool, maybe it would be valuable to use entirely independent from the game. If that should be a goal, it would need to be carefully designed for. This will likely introduce conflicting requirements though, so may not be worth it.

I'm not sure about conflicting requirements. A bayesnet backend without integrated I/O, with an I/O and GUI made specifically for the game and possibility of reusing or recoding some of the I/O and writing a new GUI for the separate tool seems like it wouldn't introduce conflicting requirements, modulo code optimization and the increase in design and coding time.

I don't think it's worth it though, unless it turns out this kind of modular system is best anyway.

Correlations are entered with a belief as cause and one as effect and values for probabilityOfEffectGivenCause + probabilityOfEffectGivenNotCause

This doesn't sound like it'll scale up easily. Correlation maintenance needs to be done manually if new causes are linked to the same effect at runtime, which means the routine that adds a new cause has to know a lot about bayesian updating to do everything properly.

For an extreme example, if the P(Z|¬A1) is .01 for A1 = Person X is Evil, and Z = Murder happens, having in mind the .01 of "someone else not being modeled kills", and then later you add into the model the 999 other people without properly maintaining each "other cause" probability, you end up with a near-certain murder given that no one is evil.

Or for a simpler example, there are two people, but you don't know about the other one. P(Z|¬A1) = .1, because P(A1) = P(A2) = .1, and thus P(Z) (base rate) = .19. If you later learn of A2 and add it to the network, you have to know that P(Z|¬A1) = .1 meant "There is still .1 that A2, but we don't know about A2 yet!", and subtract this from the (A1 -> Z) correlation, otherwise P(Z|¬A1&¬A2) = P(Z) = .19, which is clearly wrong.

Overall, I think we should let the base rates speak for themselves. If P(Z) = .1, P(A1) = .1, and P(A1|Z) = .5, we know there's enough room in the base rate for A2 at the same rate and weight. Adding a new cause should require checking on base rates and reducing it by the rate/weight of the new cause, and warn or adjust the rate upwards if there's an excess. Having to check the other correlations seems like way too much trouble.

Might be worth taking a look at how other applications have done it. (two examples).

My preferred approach, however, would be to use odds (and Bayes' Rule). Perhaps both internally and at the user level.

The "perceived base rate" vs "real base rate" issue keeps nagging me, and we may have to either force the game to upkeep background "true" rates and the player's beliefs as two separate networks, or use some hack to eliminate the "true" rates and do away with them entirely (e.g. have masked belief nodes for the base rates of other things, with hidden priors invisible to the player).

Anyway, sorry for the long stream-of-consciousness ramble. It was surprisingly hard to externalize this, given the ease I usually have working with bayesian updating.

(video-)Game design stuff in general gives me lots of fuzzies. Gaining some real experience and having contribution to a project like this on my resumé would probably give me lots of utilons (assuming I make a significant contribution). Damnit, my decision is already made for me, isn't it?

On the other hand, I've repeatedly demonstrated to myself that I'm really bad at sticking to projects like this and outputting regular effort / work.

Ah well, if you'll take the help of another amateur coder with a large mental attic full of game design theory and readings to sift through (for epistemic hygiene) who might fade into inactivity at some unknown future time, count me in. (I've already joined the google group)

Welcome, and you've already made a valuable contribution, in supporting my hunch that my original idea for the setting probably wasn't that good. :-) Every bit helps.

A suggestion for the beginning/tutorial:

[dark screen] [a girl voice crying, muttering something inaudible, only some sentence pop out of the confused buzz] "I'm sorry..." "It's all my fault..." "Please forgive me..."

You wake up, slowly focusing the room you're in. Pale azure walls, something beeping softly behing your field of visione. You are a little surprised to find yourself in this room. It doesn't feel right, but you can't seem to remember why. Your body is numb, your throat is so dried and you're hungry, very hungry. Something is clearly wrong. You remember something... There was a girl crying? You try to move your head, but your body responds extremely slowly. You turn your head to the right and see a girl, curly brown hair, standing to a chair next to your bed. She seems asleep. You concentrate to move your hand, and as soon as you do, she suddenly open her eyes, looking straight at you, eyes wide with disbelief. After a moment, she jolts up and without saying a word hurries away from you, out of your vision. Did you know her? She seemed so afraid... You are a little more comfortable in your body now, and attempt to look around: you seem to be in a hospital room, or at least in a bedroom modified as such. There are tubes and wirings connected to your left arm, and to your chest. A nice view of a garden outside a wide window to your left. A door-frame without a door, where probably the girl fled. A nice wooden cabinet, seems antique. Another girl enters, dressed like a nurse. She smilingly offers you a pen, and a pad. She points to a small label on her dress: "Ellen". "Hello Mr Rogers. I'm Ellen, your nurse. How do you feel?" Your voice tired, creaked, finally comes out: "Thirsty..." "It's a good sign, believe me. I'm going to fetch you a glass, but first I need you to write my name on the pad. You have been suffering from amnesia, and hallucinations."

This could be the setting and the tutorial: using the pad as a toned down version of the tool used later in games, and while slowly recovering the game teaches the user how to use all the controls. Added benefit: if you have to fight with amnesia and hallucinations, but you're also knee deep in a conspiracy of some kind, you better start asking yourself: "What do I know, and why?"

I like that for the intensity, but it's a pretty dark opening, which would probably turn a lot of people off.

Oh snap! Yes, all my fiction turns out pretty dark, but if the target audience are teenager, yeah, that doesn't work at all. Maybe in the future, when a library and a developer group will be estabilished, we could have a more adult "The fundamental question". Great title, anyway ;)

Around 1992 veteran game developer Chris Crawford decided he'd like to revolutionarize computer games by developing games that are based on human characters with procedurally modeled minds. Occasionally deformed monstrosities would float down the river from the jungle where he ran off into. Recently, Crawford admitted that things hadn't gone quite according to the plan.

His library might be worth poking around in.

The first thing which gives me more hope for this project than for Crawford's is that Crawford wanted the procedural modeling to drive actual plot generation instead of just model differences in the states of characters in a more pre-scripted story.

I think it's a fantastic idea, it deserves its own discussion group elsewhere if this is really going to work.

My two cents for the game.

About the ontology of the game mechanics: a player has informations and sources of informations.

Informations have as property a degree of probability and are/must/can be linked to information source(s).

Information sources have as properties a set of failure modes, linked with their own probability, and can be linked to informations or other information sources.

In your example, the information is the assassination attempt, and the source is Bob. When you dig deeper, you find out that he has a source of information, Alice. When you question Alice, though, she tells you about not having spoken about an assassination, but just a big event. In this case, we have another information, just the big event, from Alice (another source of information). But they are competing, so which is the failure mode? Is Bob lying, or just fantasizing? Or the failure mode is Alice, which is downplaying the information she gave to Bob? Another example might be a computer, with hacking as it's failure mode. Maybe at a certain point, when you have constructed your neat probability network, you find out that all the root nodes have a failure mode connected to a single source (the classic conspiration plot), etc. With this model you can play a lot with the plot...

Created a discussion group. :-)

I like your suggestion for the ontology: that sounds quite good.

Perhaps you can crib some gameplay ideas from the Phoenix Wright series, like was done with Socrates Jones: Pro Philosopher? Those games make pointing out fairly mundane logical or argumentation mis-steps pretty entertaining, and I can see how it might complement the idea of updating probabilities in a story-based scenario.

I think that lying should be possible from the beginning but, since you are a detective, you have the ability to gauge someone's reliability which is displayed as a percentage (like in your drawings). Also while reading I thought maybe it would be possible to combine 'evidence' to create new evidence. ie: Alice's shoes are wet && Bob's weather records show that there hasn't been rain in weeks +=Alice has stepped into the local lake for something today.

Great to see that there's interest! I've created a mailing list for the game, which people can join if they're interested in contributing. It's probably easiest to have the initial discussion here, and then shift to the list.

As another possible source of inspiration, Antichamber does really well at teaching players new stuff while being fun. It isn't at all about rationality, but does get players to get rid of specific preconceptions and rapidly train new habits and ways of thinking very effectively.

I think the approach used there could be very useful for introducing new concepts and getting players to start thinking in new ways (though hopefully, with a rationality game featuring Kaj Sotala as Lead Designer, those new ways would also be better ways ;) ).

My main concern is for how effective this would be for transferring the skills taught ingame into the real world (Antichamber certainly hasn't made me learn in-real-life to gel jnyxvat bire n punfz vafgrnq bs gelvat gb whzc npebff vg, whfg gb frr vs n sybbe zngrevnyvmrf. Game spoiler.)

Seems extremely cool! Will try as soon as I hit home :)

though hopefully, with a rationality game featuring Kaj Sotala as Lead Designer, those new ways would also be better ways ;)

:-)

Thanks for the pointer, Antichamber looks quite interesting. I just bought a copy, I'll have to try it out.

I think your current design will end up rewarding the 100/0 case more than warranted when there happens to be an assisination at 8 PM, and punishing it less than warranted when the 99% chance doesn't happen. If the actual odds are 99%, 99.9% and 90% should have the same score.

Or were you intending to deny the closure of rewarding someone for correctly predicting a 5% chance that the sniper will hit when the sniper rolls a natural 20?

The mechanics definitely need to be such that the dominant strategy is to give accurate predictions. I am reminded of Yvain's post on Nash Equilibria and Schelling Points, in which the optimal strategy is to attack/defend each of the cities in proportion to the values of the cities in question. One of the keys is that it is a repeated trial, which the idea of assassination at 8 does not have.

Although, it does sound as if the computer will be tracking the likelihoods by itself, and you only have to decide what to do with the information produced by the fully updated Bayes net. So maybe one of the key skills will be assessing Value of Information.

Is the objective to teach thin-slicing (getting a feel for the magnitude of the answer without doing the math)? Part of my assumption was that the player would not be given the odds that A is lying about what B said, but might be prompted to think about that possibility.