Recommending Understand, a Game about Discerning the Rules

I likely won’t buy the game if you don’t sponsor it, but if you do, then I’ll be able to find the time to play it, and let you know what I think within 2 weeks (would be shorter, but Halloween weekend is coming up, so I’ll be busy during the largest stretch of free-time I have).

https://steamcommunity.com/profiles/76561198217407285/

Edit: here is my review [LW · GW]

I've only played Zendo once. As I understand it, there are are a small number of pieces to make rules about, while rules can be arbitrarily complex. So in this situation you could indeed get better at modeling your fellow players and become more efficient at winning.

In that perspective, playing Understand is a bit like playing Zendo against the game's designer, who has hand-crafted ~130 levels (rulesets) à 5-10 screens (koans), though you're not only trying to guess the rules but also to draw paths on each screen which satisfy them. Anyway, this means that the levels are a finite resource. So to me efficiency doesn't quite seem like the right goal for playing the game.

In a spirit of deliberate practice, I think to get the most practice of rationalist techniques out of Understand, one might:

• early on, try to understand all the subtle ways in which the game gives you feedback or evidence
• in each level, try to discover the rules while doing as little experimentation (drawing paths) as possible, as if doing each experiment was expensive
• try to notice sensations like confusion or surprise, which can give valuable information regarding fruitful hypotheses to pursue
• keep play sessions relatively short, since practice sessions presumably hit diminishing returns at some point

And so on.

Conversely, if you're sure you've correctly identified the rules of a level, and nonetheless find a few screens unsolvably difficult, from the POV of deliberately practicing rationalist techniques I see little reason not to ask for hints or outright look up solutions for these screens. I like puzzle games and mostly enjoyed that part, too; but that's not why I recommended the game here.

Oh, yall play zendo with custom rules? In auckland EA we just played it random rules generated by the game, and that was interesting enough. It's humbling to be holding one of the simplest possible rules and watch some of the brightest people you know generating epicycles and missing it for 20 minutes.

I'm interested in the general practice of integrating rationality training into puzzle games.

I've been idly wondering in the past whether the strand of the rationality community which emphasizes rationality as a practice (e.g. CFAR, Duncan, Logan), and which so far mostly focused on workshops and in-person instruction, could have found any teaching success in the form of small video games. In which case that would've been a scalable approach to teaching and knowledge transfer, and furthermore one which everyone in the world could've benefited from, rather than just those in reach of a CFAR workshop.

I'm interested in the general practice of integrating rationality training into puzzle games. 

I should mention [LW(p) · GW(p)] I'm doing one of these types of games btw. I've also got a concept brewing where a player proposes a configuration of some components (which is scored by "the aggregate" (rules of the game)), then everyone places bets on whether it can be improved. I'm fairly sure this will beat both cynicism and optimism biases out of people and turn them into very clever pragmatists.

If you'd be interested in collaborating on that I'd be up for it.

Yes, Baba is You is a great puzzle game (and as I mentioned, even the Understand dev cited it as an inspiration). I would personally have been a bit hesitant to recommend it here (see the section "Why Recommend a Game on LW?" in my post), but I did enjoy it myself.

Glad to hear you enjoyed it!

Seeing how you do math research, do you share my sense that this game requires / benefits from some basic skills that are also required in rationality and research, albeit at a necessarily much shallower level?

Thanks for the kind comment! I figured the game might work well as a cooperative experience, so I'm glad to hear that that was indeed the case :).

Thanks for taking the time to write this! Yeah, a weird puzzle game like Understand may not be a good experience when tired, and maybe the game does indeed not work for you.

Anyway, I can describe how I approached the game, to maybe give another perspective for how one can play it. In principle, the initial puzzles can be understood and solved without lots of trial & error or brute-force searches. Maybe this inspires you to have another go? (And if not, that's fine, too.)

Let me take level 1-3 as an example. (Note: spoilers follow.)

Since the main challenge early on is to figure out the rules, rather than to draw a valid path on each screen, I often try to learn as much as I can from each screen before proceeding to the next (though all screens are unlocked from the start). But that's my aesthetic preference, and there are other approaches. (For instance, trying to quickly solve the first few introductory screens in a level, then coming up with hypotheses for why the puzzle dev made these specific screens to introduce the rules.)

That said, here's 1-3-1:

6 tiles, 4 symbols, 3 rules (indicated by the circles at the bottom); and an initial suggested path that begins in a circle, ends in a square, and goes through a down-triangle.

If we just follow the suggested path, we see that it fulfills all rules on the screen, but we don't learn anything new. So instead, we want to come up with hypotheses for what each of the three rules correspond to, and then falsify these hypotheses.

Just from looking at this setup, I have the following hypotheses:

• Like in the earlier levels, the suggested path begins in a circle and ends in a square. This seems like a common theme (maybe in the entire game, or maybe just in world 1, who knows at this point). But does the direction here matter? That is, if I begin the path in the square and end in the circle, does it fulfill all rules? -> Do the experiment. -> It only fulfills rule 3, so direction matters!
• Continuing from there, a natural assumption is "we have to start in a circle and end in a square". How can we test that? Well, what if we draw only parts of the suggested path? -> Do a few experiments like that. -> If we start in the circle and don't end in the square, rules 1 and 3 are fulfilled, but rule 2 never is. If we don't start in the circle but end in the square, rules 2 and 3 are fulfilled, but rule 1 never is.

So at this point my current best guess for the rules (which I can still refine on future screens) is:

3: ?

Continuing from there:

• The suggested path goes through a down-triangle but not through an up-triangle. Can we go through an up-triangle? -> Do the experiment, e.g. draw the path circle-upper_triangle-square. -> This path does not fulfill rule 3.
• What does that suggest for rule 3? Maybe "go through one (or all) down-triangles"? Or maybe "go through no up-triangles"? How can we distinguish those hypotheses?
• Here's a meta technique that can often help answer questions like this: Ask yourself, which rules are fulfilled if I just draw 1-tile-long paths? -> Do the experiment in each of the 6 tiles. -> All 1-tile paths fulfill rule 3, except the one through the up-triangle. This is inconsistent with rule 3 being "go through one (or all) down-triangles".

So my current best guess for rule 3 is:

So here we basically managed to discern all rules of the level just based on the first screen, and using little trial & error. But if that hadn't been enough, we could've proceeded to the next screen, tried drawing a path based on our current understanding of the rules, and if that path hadn't been valid or we couldn't find any valid path at all, we could've repeated the above-mentioned process.

That's just one approach to this game, but I found it a fruitful one. In particular, many levels have 3+ rules, one or two of which are very simple stuff like "start in a circle", with the remaining rules being the actual challenge of the level; and once you've identified those easy rules, your trial & error experiments become far more purposeful and less wasteful.

How would you operationalize that?

One problem here is that the game cannot test whether you understand the rules, it can only test whether the paths you draw follow the rules. In principle most regular levels (not including the meta-level at the end of each region) could be solved with an extremely tedious brute-force approach of drawing random paths.

So what would it mean for the ML system to figure out the rules? Maybe you could give it the goal to "minimize the number of paths you need to draw in a level to solve it"; but then you separately have to train the system somehow, and what is your training data in a game with handcrafted rather than randomly generated levels?

I think the UI is following the same principle as the game itself: You have to try to understand what it does from the few cues that you get.

tl;dr: The rule circles in the last screen of each level are hidden for game design reasons. This is indicated by the oval shape over the area where the circles would usually be.

First a meta comment: Yeah, the game could be more clear about that. I suspect that the reason why the game doesn't explain this detail is that explaining details in-game requires language, which requires translation to a ton of languages, which is costly for a small developer. So indie games instead lean on existing intuitions, on nonverbal communication, and on players figuring things out by trial and error. Usually that works out fine; sometimes it doesn't, even in high-budget games. I managed to play one of the AAA Dark Souls games for 30 hours before understanding some central mechanic, which was surely not intended by the devs...

Now, on to your actual question. First, my terminology: The game consists of regions, levels, and screens. 1-1-4 would be the fourth screen of the first level in the first region. Each level (like 1-1) has a unique set of rules to figure out, but these remain the same between screens, so 1-1-4 has the same rules as 1-1-1 to 1-1-3.

Because the game's main challenge is usually to discover the rules in a level, but the game can only test this by asking for valid paths, many early levels would be easy to solve by purely brute-force drawing of paths without actually understanding the level's rules. The rule circles (= one circle per unique rule which the path must fulfill in this level) are very helpful to discover the rules but also help for brute-forcing ("just draw a path on each screen until all circles light up"). So that's presumably why the dev hid these rule circles on the final screen of each level.

Anyway, I appreciate this comment, because seeing such a different experience really makes it clear that people engage with stuff from different starting points, priors, preconceptions, etc. In this case, I could not have predicted that someone would stumble on that part, specifically. (This is not meant as a criticism; I stumble on lots of things that are obvious for others - just rarely in gaming, with which I've grown up.)

One final point: One of my prerequisites for recommending this game was that I developed so-called "trust in the developer". That is, I also had some suboptimal experiences early on (in that I managed to solve 2-ish early levels without being able to articulate the rules), which made me doubt the competence of the dev; but as I played more, my doubts were continually resolved, and by the end when I struggled with a level I had enough trust in the dev not to second-guess them. But developing trust doesn't always happen (whether it would be warranted or not), and when it doesn't, puzzle games become a chore, since it makes one second-guess everything: "maybe the dev is incompetent", "maybe my solution doesn't work because the game is bugged", etc.

Anyway, I hope you'll come to enjoy the game. But if you don't, Steam has a no-questions-asked refund policy within 2h of playtime and 14 days of purchase.

I'm a bit out of my depth here since I've never interacted with Linux in anything but a rudimentary manner. Everything that follows is stuff I've just looked up.

Anyway, the game does not officially support Linux, but according to users it seems to run fine in Proton, which makes sense as it's a lightweight 18-MB game with no demanding hardware requirements.

What is Proton?

Proton is a new tool released by Valve Software that has been integrated with Steam Play to make playing Windows games on Linux as simple as hitting the Play button within Steam. Underneath the hood, Proton comprises other popular tools like Wine and DXVK among others that a gamer would otherwise have to install and maintain themselves.

How to enable Proton / Steam Play?

It can be enabled in Steam -> Settings -> Steam Play. (A screenshot of the settings page can be found in step 3 here.)

(That "Steam Play" option is not available for me. I'm not sure if that's because I've installed Steam on a Windows 7 PC, or because it's only available in the Steam beta branch. If it's the latter, you'll first have to join it via Steam -> Settings -> Account -> Beta Participation -> Change -> Steam Beta Update, and then probably restart Steam.)

Then what?

Anyway, once you've enabled Proton, apparently you can just start Understand as if you were on Windows. ProtonDB says that the game will run just fine, but if you buy the game and it still doesn't work, Steam has a no-questions-asked refund policy within 2h of playtime and 14 days of purchase. Or I could gift you a copy for your troubles, and if you can't run it on Steam with Proton, you can still refund it and I'll apparently get the money back.

Maybe with an emulator e.g. Wine. See e.g. here: https://askubuntu.com/questions/462030/can-i-play-windows-games-using-wine-and-or-through-steam 

https://www.protondb.com/app/1299400

I don't bother with linux steam, I just boot steam within lutris. Lutris just automates config / wraps wine plus a gui, so lutris will make steam think it's within windows and then everything that steam launches will also think it's within windows. Tho admittedly I don't use steam a lot (lutris takes excellent care of me for non-steam things)

The Witness is a fantastic puzzle game, but see my section on recommending games on LW [LW · GW]: the very fact that The Witness is more engaging and addicting makes it a better experience but (according to me) a worse recommendation to a community of non-gamers who have other high-value uses of their time.

Re: completing Understand puzzles without understanding why: Yeah, I also did that at the start, and that also hampered my experience a bit. Personally I eventually returned to those levels and solved them in a way such that I understood and could write down each rule.

Anyway, for me the core experience of Understand is not so much solving any individual levels, but constantly practicing the loop of: make new hypothesis -> experiment -> falsify hypothesis -> make new hypothesis.

For example, it's one thing to learn about confirmation bias [? · GW] by reading about it on LW, and another to constantly run afoul of it during gameplay; the latter experience is a great illustration and motivation for the claim that one should try to falsify, rather than prove, one's hypotheses.

I do not currently expect there to be an Android port (e.g. some of the meta puzzles in the game play around with features of operating systems, which could not just be ported over), but maybe the dev will make one.

But I'm a bit confused as to what prerequisites you think the game requires. It requires a desktop computer (or laptop) and an operating system (Windows or macOS are natively supported, and according to a comment thread [LW(p) · GW(p)] Linux probably also works; have added this to the main post), plus the Steam game client (freely downloadable on all OSes from their website; see the final section [LW · GW] of the post).

I didn't notice that, that's pretty neat! And unexpected; my impression of such rankings is similar to trying to get the top comment on a Reddit thread, i.e. the advantages of an early comment or review ordinarily cannot be overcome by anyone who comes along later. Though in the case of this review, maybe linking from this LW essay made the difference?