problem with meta

Rationalists talk sometimes about “meta”. The definition is somewhat loose, but it usually has to do with a system referring to itself. If you write “shopping list pad” on your shopping list, that’s somewhat meta. If you write a blog post on how to write blog posts, that’s meta. Some things can be “more” meta than others, not usually due to the number of recursive levels, but more to do with how much entanglement and complication occurs when a particular light is made to shine its light on itself. If you wrote on your shopping list that you needed more shopping lists, you wouldn’t feel anything strange was happening. But other examples can be really strange.

That’s what this post is about. More specifically, this post is totally not going to be what you would have thought it was going to be about, because my problems with meta seem a little different from the typical ones people have.

Let’s take a look at what they are.

1. Meta being amusing

Some examples of meta can be super hilarious. Obviously it’s a matter of taste. But I started off in mathematics (well, undergrad. I haven’t ever done any mathematical research) and now I do software. Meta is just one of the tools of the trade. Plenty of mathematical constructions and software constructions can be used to refer to themselves in various different ways, and often really useful and interesting things happen when they do. Learning to understand and work with that is important.

As such, if something provokes giggles simply for being meta, and not for any other reason, it can be irritating. It’s how a biologist would feel if people thought that frogs were inherently funny. OK, they’re kind of cute I guess, but really every time a frog is involved it causes a smirk? It does not, I should be completely clear, come across as any kind of insult or slight to my profession. And I’m much more cool with the whole thing than I make it sound here, especially as my own humour really is pretty self-referential.

But, I just wanted to get this out of the way. This particular post is going to be completely devoid of any kind of self-referential lol, as you are at this point imagining would be so clearly appropriate for a post that starts of saying that self-reference isn’t inherently funny. Unless I change my mind and decide to throw some in anyway. The reason for this decision will become apparent in the next point.

2. Meta is confusing

Confusion can, of course, be a source of enjoyment and this probably relates to point 1, because I imagine I get less easily confused and hence more easily bored by something being meta than a typical smart person would.

But it still gets pretty confusing.

This comes in two main flavours.

2a. Confusion in communication

This is the easier one to understand. I’ll give an example. At work recently we were discussing a part of our software that we needed to implement. It was a task scheduler. There would be some things that the software needed to do, each would take some time and depend on other tasks. A pretty typical software challenge. Right?

At the same time, we were planning out our own time, deciding who was going to do what when. These work tasks also had dependencies, meaning the project management problem (common to any business, not just software) in some way mirrored the detailed structure of the software that we were going to implement. Hence, meta.

There wasn’t anything inherently confusing about this. It caused a slight giggle as mentioned in point 1, but the giggle was this time to do with the use of language: somebody got confused about which sort of “task” somebody was talking about, just momentarily.

But, we didn’t actually solve that. This is the non-entangled kind of meta, and any confusion would be easily resolved by temporarily using different vocabulary for the different kinds of task and dependency. We didn’t do that though, and I wasn’t going to try and force it because I didn’t have this blog post in hand to slap on anyone’s desk, and it didn’t seem like quite a big enough of a problem to really make an issue out of.

But while this kind of confusion is in play, each sentence that you here must, at least as a first pass, go through two entirely different interpretations. You’ll immediately reject one of them as not making sense in context (at least most of the time), but if the topic of discussion is already cognitively demanding it can add an extra layer to that.

Basically, assume that your audience is really dumb and that you’re even dumber, and you might be able to communicate effectively. <— I am not actually sure I endorse that statement but you know what I mean.

2b. Inherent confusion

This, I think, has to do with when you’re trying to use the same parts of your brain (I mean literally the same neurons, not just the same overall area) for processing the two levels of the meta sandwich.

The best example of this is probably Gödel’s incompleteness theorem. (If I can think of another example of “inherent confusion” that somehow manages to be more accessible then I will mention that after. You can skim this one a little). This has to do with whether systems of logical can prove their own consistency and so on. Even a correct statement of the theorem contains a lot of meta weirdness, and the proof (or trying to reason slightly more informally about what it is or isn’t saying) seems to be somewhat of a nightmare.

Why though? Also, isn’t all mathematics like that?

The particular thing about this one that makes it tough is that you’re dealing with two formal systems. One is the one that you’re doing your logic in, one is the system that you’re trying to prove things about.

In some sense, these are entirely different. It would be like getting Euclid’s Elements confused with a triangle.

In another sense, they are exactly the same. After all, the theorem was about whether systems of logic can prove their own consistency, so the meta system needs to have the same structure as the object system.

But wait, hang on, wait what?? We are trying to prove that a system can’t prove things about itself. We can’t prove anything about a system of logic at the same time as actually using it, mathematics just doesn’t work like that. So really there are three levels… there’s the system we’re using as our scratchpad, then there’s the formal system that we’re talking about, and then there’s the objects that it’s talking about, that themselves happen to look like formal statements too. So, what we thought was A->A is actually B->A->A because the outer system doesn’t need to resembles the others it just needs to be rigorous enough to work in, ok well fine.

But then Gödel’s second incompleteness theorem comes along, and says basically “take the proof of the first theorem, and do some weird magic with that”. What we thought was B->A->A is now A->A->A, but in order to actually do anything with it we still need our scratchpad so it becomes B->A->A->A. So what sounded like 2 levels is actually 4. Problem? Yes. But not the problem.

We can certainly handle this number of levels of indirection. Imagine two people discussing a friend, saying that he gossips a lot. This is me imagining you imagining people talking about a person talking about people. Put like that it sounds strange, but in the original formulation it sounded entirely natural. We’re used to some levels of indirection in social contexts. Our brains probably evolved for that.

Imagine a movie where half way through, it’s revealed that the first half of the movie was actually a story being told by a character, and the perspective now switches over to the framing story. You may or may not like that narrative device, but it would only take a second or two to fully understand what had happened. Some kinds of meta we’re just good at.

The math kind we’re really terrible at though. It partly has to do with dealing with kinds of object that were totally not relevant in the ancestral environment, and which take a lot of training to even comprehend the existence of let alone how they are supposed to behave. It also has to do with the similarity between the layers though. In the movie example, if the characters in the framing story and the framed story were all the same people and there was a similar kind of plot going on, it would actually become quite a bit more confusing.

Oh yeah, and now imagine that someone got kind of the wrong idea about Gödel’s incompleteness theorem and has started to say silly things about it. How on earth would you resolve that one? You are dealing with two instances of 2b, and one instance of 2a, simultaneously.

Sorry no more easily accessible examples came to mind.

3. You just went meta and I have literally no idea why

This happens sometimes. I have no idea why.

An example would be I was reading something about qualia, and it was talking about what happens when we see the colour red. The word “red” was printed in red.

Sometimes, when people feel like their wording isn’t emphasizing things enough, they will print something in bold, italic or apparently in this case, meta.

4. Going meta too soon

This is related to 3, but has some clearer explanations. Usually, because of the other problems described here, a self-referential example is not the best one to start with.

What’s a flibbertigibbet? Well… the concept of being a flibbertigibbet is itself a flibbertigibbet!

How much closer are you to understanding what that word might mean? Answer zero, or possibly even negative because you’re now just confused and not so much in the mood to learn any more.

I think we can land on the self-referential example because it’s just sitting there readily available in our minds. Try to think of words with four letters, there’s a good chance you’ll come up with “word”, “with” or “four” along with whatever else you have cached like swear words. Our minds work that way and it’s not exactly a problem, but it can become one when trying to teach a concept because you have to imagine what it’s like having literally no idea what this concept is yet, and instead build up from what people already do know.

I had another reason this might happen but I forgot it.

5. Conflict of interest

Here at least I agree with the general consensus. We don’t see ourselves the same way that we see others, and when we try to it can expose everything from outright fraud to weird cognitive biases we don’t even necessarily understand.

And since everybody knows this, even if you somehow hit on the magic formula to see yourself completely objectively, no-one’s going to believe you. So this section is really “conflict of interest, or the perception of it”.

The case study here is an organization that would evaluate the “metacharities” within effective altruism. Such a thing does not exist at present.

If you could pull it off successfully, it could be of great value. There’s a lot of confusion within EA about how money should be donated, and at least some of that comes from the difficulty of assessing the charities (they are, mostly, registered charities) that evaluate other charities and the various other activities required to get people to target their donations more effectively that we might as well call “spreading EA around”.

Evaluating a charity is hard. Evaluating a charity that does something no-one’s really done before is harder.

Evaluating the machinery of EA itself is hard enough to be currently outside our grasp. A part of this is the conflict of interest issue. The sort of people who would be best placed to evaluate metacharities are the people currently working for them. There are actually two reasons for this: having the skills and knowledge required to evaluate charities (“meta-level” experience), and understanding how the various candidate organizations work from direct experience of having worked there (“object-level” experience).

If someone like me were to jump in and try to evaluate all the metacharities without understanding what’s going on at either of these levels, it would be a disaster. So other than finding outsiders who happen to simultaneously be brilliant and actually want to do this for some reason, you’re down to forming a new organization of insiders or getting the existing metacharities to evaluate each other as a small part of their other activities (this is more or less what happens now, to be honest).

It would be nice if it wasn’t like this but I can’t think of a solution.

6. Lightweight accusations of hypocrisy: meta as an attack

Suppose someone writes something about grammar or spelling or something. You’re going to be looking over it really carefully, trying to find every last little typo aren’t you? You know that you are.

This sort of goes back to point 1, but people do genuinely seem more annoyed when something specifically doesn’t conform to itself. The likelihood is that the essay on spelling doesn’t contain more typos than anything else the person typically writes – and it’s no more hypocritical for them to be a sloppy speller in their spelling article than in anything else they write.

But we notice it. As mentioned in points 2b and 4, when something is available in our minds it’s available at both the object and meta levels, and we’re happy to let it just slop from one to the other. If it can be used to make someone feel bad then yeah we totally go for that I guess.

Oh yes this is the one I was thinking of:

If I’ve successfully convinced you that you shouldn’t form strong heuristics just by looking at a few salient examples where they seem to hold true, then shame on you.

How exactly are we to interpret that comment? I mean, clearly it’s a joke. But are both sides of the joke equally apparent to you?

The basic message is: we have a baseline level of expectation about something. Then we get a few salient examples and update too far, basically overfitting on the tiny amount of data.

The meta-level thing going on is: we have a baseline level of how much we tend to update on evidence. Too much and it’s bad, too little and it’s bad. Unless something went wrong in our minds (like me during my psychosis), this is something we’ve built up over a lifetime of experience and we shouldn’t let it be swayed by one blog post.

But understanding a cognitive bias, like you would do from reading the SSC post, is not the same as generalizing from a few examples. Cognitive biases should be somewhat obvious in retrospect: we don’t need to see lots of examples of people who generalized from too few examples, in order to tell that they’re doing something wrong because it’s just basic statistics that we already knew. The non-obvious part is, oh yeah we were probably all doing that too and didn’t notice it, maybe we should be on the lookout for it at least in the high stakes cases.

The funny version of the joke is “you were supposed to update on what I was saying so quickly that you realized that you weren’t supposed to update on what I was saying”, which is self-contradictory in a totally unexpected way and is the sort of thing I like. The less funny version is “don’t fall for the dramatic irony meta trap”.

There’s more to say about hypocrisy but it’s bedtime and I just have to accept I didn’t get anything meaningful done all weekend.


None of this is actually really a problem at all and I don’t know why it’s bothering me all of a sudden. Keep on the lookout for more weirdness that might be a psychosis early warning sign, without catastrophizing them.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s