ProbDef: a game about probability and inference

by abstractapplic1 min read2nd Jan 20189 comments


Software ToolsExercises / Problem-SetsGaming (videogames/tabletop)

Hey LessWrong! Do you still like probability, inference and Bayes’ Theorem*? Well, I made a (free, open-source, HTML) game about using them to protect a spaceship from explosive mines.

I have plans to make an improved/expanded version at some point in the second half of this year. But I’ve not had much feedback on the current version; and feedback I did get focused mostly on its quality as a game instead of as an accurate and useful representation of probability concepts. So I’m posting it here, in the hopes that some risk/probability enthusiasts will share their opinions & give constructive criticism.

*Just so you know: Bayesian reasoning proper only shows up in the last third of the campaign. Yes, this is one of the things I plan on changing next iteration.

9 comments, sorted by Highlighting new comments since Today at 4:23 AM
New Comment

Simple and fun. I liked it! (Only played for a few min, though.)

Please don't take this the wrong way, but: if you liked it, why did you stop playing?

I ask because this is a recurring theme I've noticed: people saying it's fun & well-put-together (and appearing to mean it), but that they're not motivated to keep going. Or, in other words, that it's engaging but not compelling.

I'm pretty sure it comes down to the absence of story, context and overarching goals. But if the cause of this effect is something different, I really want to know.

My own experience (played for a couple minutes, felt slightly frustrated, and came back here to relate feedback).

I don't think it's anything to do with story or overarching goals. Tetris and Galaxian have no context but are great games.

The short answer is "game design is hard, game design introducing novel concepts from scratch is much harder." (i.e. to design a good game, it's not enough to have a clever premise that's reasonably well executed, there's a whole lot of fine-tuning that goes into it). I think it's an achievable goal to make this as compelling as Tetris or Galaxian (and that that's a goal worth striving for), but it's going to involve a few serious overhauls.

Some stabs at a longer answer:

  1. The entire premise of "your attacks probalistically do different things" is an extremely new premise, so you have to introduce it even more slowly than a game usually introduces things. Level 1 didn't actually mention it (and I failed to get to level 2, not sure if that'd be spelled out more later).
  2. Insofar as the game needs context, I think the most important bit is an intuitive reason why my attacks have random chances to fail or capture. (I'm tempted to say there needs to be an explicit reason, but the more important bit is just that it feels counterintuitive, and an explicit reason may not actually help).
  3. Simple UI improvements: at the very least, have the hover-text for the different weapons appear directly underneath them. (I didn't realize the hovertext existed for a long time, because human eyeballs can only actually see things near the area they're directly looking at, and confabulate what the rest of the world looks like via approximation. This means I didn't actually see the text changing at the top). Simply moving the top panel down to the bottom would probably ameliorate this a lot.
  4. I think it might be necessary to go even more explicit and have the different weapons literally spell out their percentages, so that I can see them at a glance instead of having to check each weapon over and over to remember them.
  5. (That all said, an idealized game doesn't have a formal tutorial - instead, it naturally teaches you things via gameplay and exploration. But I think erring for now on the side of explicitness is probably better)
  6. This isn't the biggest problem and shouldn't be a main concern till you're gotten the overall system in good shape, but graphical polish does play a role. (A good game doesn't need good polish, and polish doesn't have to mean AAA game. There are subtle things that can make even a minimalist Tetris or Snake clone feel good)

Also worth noting that designing a game to teach a concept is hard mode, since you don't have the luxury of changing the core concept.

Some games that I think did a good job of teaching concepts while excelling as games:

  1. Universal Paperclips teaches you what it is like to be an AI making paperclips, and a sense of how big the universe is (in about 6 hours). It introduces new concepts bit by bit, never spelling things out explicitly but dropping clues. The core mechanic is dead simple, and yet manages to continuously reinvent itself every 5-10 minutes.
  2. Factorio teaches you how to manage complexity and devise processes (as well as more common things like managing resources), and in particular, teaches a lot about how real energy production works. (i.e I learned things about how Oil Production works I didn't know before). [And unlike a similar game, Super Energy Apocalypse, it does so without feeling like an "Edutainment Game" at all]
  3. DragonBox and DragonBox Elements are most similar to your game, in that it's a) teaching a math concept, b) a fairly blatant Edutainment game instead of managing to trick you into it. But has a bit more polish and introduces the concepts more gradually.

Fun game! And the music is really nice. By the way you have a typo in there somewhere. It says you refresh to ten shields in a level where you only get three.

Good catch! I have a bunch of minor tweaks I want to make to this version of the game before moving on, I'll add that to it.

Played through the campaign. It feels too slow and arbitrary. That's why I almost quit. There's no quick feedback loop saying "you did well!" other than whether you won, and the progression in the campaign has too much time between concepts. I want to know "okay so I captured 3 mines is that a little, expected, lucky...?" I want to progress quickly to "the real game" once I understand what's going on. And only at the end do I know what mines are for, and I don't know whether I have enough or way too much or what. Arbitrary: you can do the right thing and lose and no one even says "hey you did the right thing", also there's a feeling of "am I a chump if I keep playing rather than restart when the completely unavoidable shot from the turret hits me rather than misses?".

That said, it was fun! The above is my guess at why you get fun without engagement.

I played through two levels and went through the library examples. I really enjoyed it, and I think it is useful!

Things I liked:

  • The probabilities were round enough, and there were few enough options that I didn't feel like I could only play if I did the math everytime.
  • There were just enough instances where I went "Okay this is what I should do" only to think, "Well... Is it?" and then realizing that I could totally do the math and find out.
  • Having to reevaluate what my goal was given new info info (that note on risking hits in order to capture more)
  • After playing for a bit I did genuinely feel into it.
  • When the mines started coming a lot faster, that corresponded with a though of "Oh shit, I really have to make sure I'm using the right strategy".

Things I didn't like:

  • I might have just missed this in the tutorial, but I didn't catch how exactly captured mines played into your overall score.

Basically, I feel like you hit a sweet spot of complexity. I was incentivised to actually do the math to be right, the math was easy enough to do, but still just hard enough that you're inclined to to ballpark things and build up your intuition.

##Just played the rest of the game

The decoy mines are awesome! That made me think more and try to keep track of more info in my head.

Thanks for the detailed feedback! You should be pleased to know the next iteration will make the utility of captured mines a lot more obvious, and do so a lot earlier (this is a pretty common complaint). Also, if you liked keeping track of decoys in your head, I should probably make sure you know you can turn the autocalc off in the Options menu & get the same sort of experience with the Bayesian levels.