Mo' Meta

A Powerful Force

My public thinking was never meant to follow a specific theme, but when pressed for a newsletter description I found metacognition to be the gist. The more I think about it though, metacognition is not specific at all. Everything we experience is a result of cognition. And so are most things that we do. Even if we ((much) more than) occasionally do stuff without thinking, metacognition covers that as well. The word might first make you think of psychology, or some exercise you did in a language arts class, or definitely all those note-taking power users and their second brains. But metacognition is for everyone. If you want to improve your happiness, productivity, creativity, learning ability, relationships, sense of purpose, or anything really...there might not be another activity that has such large, compounding returns on your investment of time and energy.

But why do we get these returns? Why are they compounding? And why aren't they always noticeable immediately?

Metacognition is only one example of a much larger and more powerful all encompassing that can be unleashed with naught but a prefix...a force with so much much much value to each of us as individuals...and to the entirety of humanity as a whole...

...that it is neglected by the majority of people and institutions on the planet.

The prefix, you may have guessed, is meta-.

What is Meta?

Metadata = data about data,

Metacognition = cognition about cognition,

MetaX = X about X

Meta is also used to mean doing something on a higher level of abstraction. Think metaphysics, meta-ethics, or metamathematics. In each, an academic field of study is itself studied on a level once removed from reality; which might explain their likely unfamiliarity. Sometimes prefixing with meta- can give the perception of complete and utter academic nonsense—graspable only by head-in-the-clouds savants and philosophers. Though I disagree that these studies are nonsense (their inquiries probe the most fundamental and transcendent questions of what is meant by our world and our existence but lead to very real applications) and that they can be grokked only by a few (any feeling of exclusion from their understanding is more a failure of presentation and access than ineptitude), I am not exactly referring to this kind of meta.

The meta I have become interested in is applied practically and removed only a strategically limited number of times from reality. It is about abstraction only in the sense of zooming out, noticing our past, and planning for our future. It is nearer to history, statistics, policy and innovation than philosophy. This meta is about promoting stability and growth by tackling problems from their root. It is rooted not in study but in a tight relationship between analysis and action.

The meta I speak of can be defined as working not only toward goals, but working toward better processes to achieve those goals. I call this derivative engineering.

Shifting the Derivative

Consider a function P(t)P(t) which gives you the amount of a problem that you have at any given time. It's derivative is P(t)P'(t). Say you are trying to reduce this problem as much as possible. You have some extended period of time to consider, say TT, and now is 00. And over this period of time you want to minimize 0TP(t)dt\int_{0}^{T}{P(t)}dt or P(T)P(T). Or maybe even TP(t)dt\int_{T}^{\infin}{P(t)}dt. Now say that you have limited resources of a total RR. You can spend these resources in order to decrease one of the functions. To decrease P(t)P(t) by one costs C0C_0 and to decrease P(t)P'(t) by one costs C1C_1. So the two actions that you can take are R:=RC0,P(t):=P(t)1R:=R-C_0, P(t):=P(t)-1 and R:=RC1,P(t):=P(t)1R:=R-C_1, P'(t):=P'(t)-1. The second action is equivalent to P(t):=P(t)TP(t):=P(t)-T. So it makes sense to reduce the derivative as long as TC1>1C0\frac{T}{C_1}>\frac{1}{C_0}, i.e. TC0C1>1\frac{TC_0}{C_1}>1. Call this constant k.k. It is the product of the time period and the ratio of efficiencies for the two actions [B].

As TT increases, there will almost certainly be a point where kk exceeds one. The argument for meta essentially says that the time scale of humanity is more than long enough for this to happen. Though urgent issues always exist, solving them does not require the entirety of our resources.

Notice that the strategies to this game change significantly when TT is infinite. The difference between long term thinking and full term thinking is distinctive. As James P. Carse says in Finite and Infinite Games, the more important part of playing an infinite game is continuing to play the game, rather than ending with some favorable outcome. The game humanity is playing, which is the only known game in the universe, is not necessarily infinite, but potentially infinite. It could end in existential catastrophe, but there is a non-one probability of avoiding this, a probability that the players in the game can largely control. As TT goes to infinity, kk will always be greater than one.

Whether thinking long term or full term, it seems very likely that we spend too few resources on derivative engineering.

Meta in Action

"Give me six hours to chop down a tree and I will spend the first four sharpening the axe." -Abraham Lincoln

A cliche quote, but I bet you didn't realize he was talking about meta. When put into action, meta enables better action by solving problems from their roots and preparing for more efficient change.

If change is a goal, we should engage more in political debate about political debate, education about education, coding about coding, jobs about jobs, science about science, and progress about progress.

Here are a few examples of potential ideas on how meta could be useful:

If we want a stable nation, we shouldn't be legislating about tax rates; we should be legislating about information. If our tax rates are not optimal, the root of the problem is not a public misunderstanding of taxes; it is an inability to equip voters with adequate knowledge, and rather than asking them to hit the moving target that many now believe is the truth, we need to stop powerful actors from moving the target with blatant lies and insubstantial calls to action and reassure people that truth is not relative.

If we want more innovation, we should not just start more startups, but start more VC firms to fund those startups and enable innovation at a more fundamental level.

If we want well-educated citizens, we need not to educate them more, but to educate our educators about how to educate. Many seem to have caught onto the idea that academia is about to get schooled by alternative education. If this shakes up the scene enough for a drastic improvement, the results will be monumental.

If we want to write better code, we should spend less time being strict about style requirements and standards and more time writing languages where good code comes naturally. This is an example of a meta that has already happened to a large extent. The computer science community, after all, is very good at abstraction.

If we want people to do better work, we don't need to train them better. We need to work on finding better work for people. We need to eliminate fake jobs and provide better resources to match skillsets to careers.

If we want to make more progress, march on towards oblivion in order to expand the pie at an even faster rate, we absolutely need more progress, but this is becoming harder to find. We can help ourselves by making progress on progress itself, by understanding its causes and enacting them.

If we want to achieve goals, we should spend more time selecting metrics than chasing them.

You may disagree with the specific opinions about how to improve our world, but the importance of meta remains. You may insert your own preferred policies, and as long as they shift the derivative they will be more impactful.

Finding New Fruit

At some point any enterprise runs into the problem of diminishing marginal returns. Each added employee creates less and less value, until this value is negative and hiring is counterproductive.

On an industry level, there are also diminishing marginal returns. Once an industry has matured, it shows less growth while making attempts to squeeze out profit increases by making incremental upgrades to efficiency. At this point, we choose to divert our resources elsewhere to pick lower hanging fruit.

And an even larger phenomenon is happening, as described in The Great Stagnation by Tyler Cowen. Our society as a whole is running into diminishing marginal returns and the rate of increase of economic productivity in America has begun to decline. The low-hanging fruit that our economic growth previously relied on is diminished and, as a result, we are being forced to squeeze out every little bit of marginal return we can, without the usual growth to show for it. Because low-hanging fruit is scarce, we have run out of other places to divert resources. Now is time to invent new places to divert our resources and this happens by focusing on meta, which, at its core, is all about evaluating which actions give the greatest returns and taking those actions.

When You Shouldn't Worry about Meta

While it needs to be used significantly more, there are certain situations where worrying about higher order derivatives is inefficient.

The first case is when growth is very early and you need to very quickly bootstrap yourself up to the next level. One example of this, which is well explained by Paul Graham in Do Things that Don't Scale, is in a brand new startup. Basically, when you are working with short time-scales it doesn't make sense to overthink for the long-term, because everything is do or die.

We also should perhaps avoid spending too much time on metameta. When meta isn't well defined yet, jumping straight to metameta is like guessing. Generally, we should take one meta at a time and focus hard on that first meta, unless we know enough to dig deeper.

Though these situations exist, we convince ourselves more often than not of some false urgency. Even if there is deathly deadline, sharpening the axe for a minute before hacking away at the tree will probably shorten the overall time by a lot. When things quickly move fast, be sure to start by thinking slow. Choose your direction before you start frantically swimming.

What Is Still Unclear

There is truly something to be said for dropping all else and just building. This idea of doing things that don't scale seems to contradict the entire point of meta. And to be honest I have not figured out how the two fit together or which is more important. There is clearly a need for more swift execution, but the actual process of figuring out how to fix this might fall under the label of meta. This is what Patrick Collison is doing on his "fast" page and is the type of thinking of which I feel could make a great difference.

Meta also seems a bit similar to exactly the thing I looked upon negatively: squeezing out productivity gains. I think that this contradiction at least can mostly be explained with some nuance. While productivity is about changing processes for short term gains in the productivity of what we currently do, meta is more about changing processes for long term shifts in what we actually do, not just how we do it. Of course one can always be fitted to the other linguistically, but I think there is a real line between the two; productivity converges toward a maximum efficiency limit, but meta creates diverging possibilities.

It might turn out that, like with everything else, we need to strike the right balance (between meta and raw execution). If so, this essay more makes the case that we are doing less meta than we ought to be. But it also seems to be common opinion that we aren't doing as much raw execution as before. So what are we doing? How can it be explained and accounted for? I'll explore this in the near future. For now, meta stands as an exploratory idea, one that I'm excited to dig into further.


[A] Realized the other day that my public thinking is mostly about metacognition. So thinking about metacognition = thinking about thinking about thinking = metametacognition. Cool. But then my recent post, Thinking Along Axes, was actually me thinking about thinking about how to think about thinking. So, metametametacognition? But then if this note is thinking about that...then what? But hold on a second. Metacognition is thinking about thinking. When you add the prefix meta to an action X, it means that you are Xing about X. So if you add meta in front of metacognition (X = metacognition = thinking about thinking), wouldn't you be Xing about X = thinking about thinking about thinking about thinking? Which is usually called metametametametacognition (that's 4), not metametacognition? So it seems like, when done correctly, linear increases in the prefix correspond to exponential nested actions. Or more formally, because it's kinda funny: N=2pN=2^p where N is the number of nested actions and p is the prefix count. When p=0,N=1p=0, N=1, so you are just doing that thing like a normal human being (sounds nice right about now). When p=1,N=2p=1, N=2, so you are doing something meta like a slightly above normal human being. When p>1p>1, all of our fragile knowledge of the metaverse breaks down as NN increases exponentially. What are we to do? There is only one option: fractional prefixes. An example: Say you want to address what is often incorrectly named metametacognition. Simply plug your value of N=3N=3 (thinking (1) about thinking (2) about thinking (3)) into the equation, solve for p, and insert the fraction part of that number in brackets after the integer part number of metas and before the verb like so: 3=2p3=2^pp=log23=1.58496250072p=log_{2}3=1.58496250072 → meta[0.58496250072]cognition And there it is folks: meta nomenclature. Have I reached meta transcendence yet? Also meta stuff in general is not just cool but really useful. We should do more of it. Which is what I'm about to say in the rest of this essay.

[B] Efficiency is the inverse of cost.