The Educational-Middleware problem in Crypto

Here's a series of events that has repeated itself regularly since I first got acquainted with the Crypto space. Ask yourself if it seems familiar.

1. You hear of a protocol that enables functionality that you think is very interesting. This could be in a podcast you're listening to, or a Tweet or when you're checking the price of tokens on Coinmarketcap.
2. You're intrigued and you want to find out more. You go to the protocol website and in the course of browsing, you get to the protocol whitepaper.
3. You encounter terms that you're unfamiliar with almost right away. You try reading through anyway to get a gist of wtf the protocol is about. You fail to 'get it'.
4. You go to the developer docs on the protocol website hoping that will help. It doesn't.
5. Before giving up, you Google to check if some blessed soul has some material that might help.
6. You click a few links, encounter a few blog posts that give you a layperson's view of the protocol. You listen to a podcast or two where the founders give a 30K foot view of the thing.
7. You leave deeply unsatisfied, but tell yourself you'll be back to dig through the hard stuff in the paper so you can really figure out what's going on.

If you're anything like me, you've had this sequence slap you over so many times that it isn't funny anymore. Seriously though, go through that list again. Pay special attention to points 3, 4 and 5. Together, they are symptoms of a gaping hole in the center of the crypto-eduverse: The lack of good Educational Middleware.

To understand what I mean by Educational Middleware, let's focus on the part of the definition highlighted above and try mapping some of the context to the field of crypto-education.

By making it easier to connect applications that weren't designed to connect with one another, and providing functionality to connect them in intelligent ways, middleware streamlines application development and speeds time to market.

If we abstract out the details and map the educational space of a protocol, we can think of the whitepaper as being the lowest level description - the educational equivalent of the the operating system. It has a lot of a little details. It's technical to an extent that a select few experts can read it and understand exactly what it's saying. And yet, it underpins the actual working of the protocol.

At the other end of the spectrum are the podcasts and the general interest videos and reading material. These are the equivalent of application layer software. Significantly easy to understand and work with at a high level, but you'll get no feel for what's happening under the hood.

In the middle lies Educational Middleware. Eddleware if we're going to get cute with terms. But before sketching out what I think 'good' Educational Middleware consists of, I want to take you through real world examples of this gap that I've experienced several times over.

Let's take the example of Bittensor. Before I get Bittensor stans angrily DM'ing me, let me make a general disclaimer. Any protocol that I reference here is for the sake of making a larger point about the state of crypto protocol education.

When I first came across Bittensor, it seemed to be just my kind of protocol. It took something that I was a practioner in - AI - and mixed it with blockchain-based incentives to create what the founders called a "marketplace for machine intelligence".

To imagine the initial effect this description had on me, take a look at the domain name of this site. Then take a look at my Twitter profile page below. I'm not exaggerating when I say that I was giddy with happiness when I first began reading what Bittensor was about.

My unbridled joy was, as they say, short-lived. When I sat down a few weekends ago to start reading the white paper, I experienced a mixture of frustration, inadequacy and anger (in a 2:3:4 proportion) that was heartbreakingly familiar from previous endeavors in this space. The crypto enthusiast in me felt the frustration, the Ph.D. Data Scientist in me felt the inadequacy, while the generally interested student / wannabe educator in me felt a boiling rage well-up from deep within...

So where was I? Yes, the Bittensor whitepaper.

Even after 5 attempts, I had failed miserably at understanding the core formalism in the paper. I could not understand what the authors meant by machine intelligence or how nodes were being allowed to bring their own models and data and yet mine TAO. I wasn't able to neatly figure out how to map the formalism on to how other nodes were scoring arbitrary models. Weirdly enough, I was able to understand their penalty mechanism for malicious behavior. Small mercies!

After wading through the docs and blog posts on the website, I landed on a set of video resources that cleared the "machine intelligence" fog a bit. I made a mental note to replace "machine intelligence" with embeddings whenever I encountered it in context. Funnily enough, the founders made jokes about academics publishing papers that only a few people read. Given the state of their own whitepaper, the irony seemed lost on them!

I also joined the protocol Discord group for good measure, hoping to get some insight. Not much luck there either, with the resources section turning up the same material above.

Bittensor is an archetypal example of a protocol lacking good Educational Middleware. While I might look like I'm unfairly targeting Bittensor here, this is not just a Bittensor problem. This is a Ceramic problem, this is a Fetch.ai problem, this a SingularityNET problem. At the risk of sounding dire, this is a systemic problem in crypto.

To be sure, not all protocols suck at Educational Middleware. Some protocols, especially the OGs (Bitcoin, Ethereum to name a few) have had the benefit of tremendous community participation over several years leading to really good educational material. From my vantage point of having surveyed several protocol of varying sizes, good protocol Middleware is the exception, not the rule.

Here's how would I define good Educational Middleware:

Educational material that takes the protocol-level formalism, breaks it down into a semantic tree, with the most difficult, detailed concepts on top (leaves), branching down into the most easily understandable, intuitive and fundamental concepts at the bottom (roots).

The actual vector through which the knowledge is conveyed is unimportant. It could be books, videos, long articles or even code. What is important is the process of stepping through each novel concept, casting it in terms of knowledge already acquired, while making very few assumptions about the skill level of readers/consumers.

This is a tall ask, and if you're hard pressed to find any educational content in any domain that does this, I won't blame you. Often times, with things like this, you only know it when you see it.

In that spirit, I'd like to give you a few examples of what I think are really good Educational Middleware. This is not an exhaustive list, of course, but it serves to illustrate my point.

1. Andrej Karpathy's series on creating conceptually complete models on the state-of-the-art in AI from scratch - Backprop, ChatGPT
2. Sebastian Raschka's book on Machine Learning from Scratch in Python.
3. Jimmy Song's Programming Bitcoin book.

What's common across these examples is the willingness to take an interested, willing participant from basics through to the state-of-the-art (or very close to it) via step-by-step explanations, text, diagrams, analogies and code. All this, while assuming very little in terms of technical sophistication on the part of the user.

This is very difficult to do well. It is a lot of work. It needs you to cast yourself in the mold of the receivers of this information and make it cognitively easy enough for them to follow along.

This could be why crypto protocols do a very shoddy job with Educational Middleware. Protocol builders and resource-strapped founding teams are typically engrossed in building their core product and finding product-market fit. Education gets lip-service, but is never highest on their list of priorities.

In the long-term, here's why I think this is a grave mistake: Attention is the de facto currency of the crypto economy. Not just attention from the degens wanting to flip their coins for a 100X profit. Or from interested amateurs and the occasional googlers.

A far more potent kind of attention is from developers and the people who are motivated enough to become backbone-stakeholders of the network (stakers, miners / validators).  These are the people that will write the apps that will make a protocol usable, and create the infrastructure that will enable a protocol to thrive in the long run.

The implicit bet in crypto, as best as I can tell is: incentivize them, and they will bootstrap themselves and build, stake and develop. This has worked, to a certain extent, but I think such an approach leaves out significant numbers of otherwise-willing participants.

Now, I realize that most protocols of note have open sourced their code. I also understand that anyone can simply clone a repository, make changes to the protocol and create a PR in this egalitarian, meritocratic space. But I am not convinced that this is the most efficient way to encourage long-term developer & stakeholder participation.

I think good Educational Middleware is a force-multiplier for a protocol in the long run. By making protocol mechanism transparent and easy to understand, it helps recruit tinkerers and genuine technology enthusiasts. By tying in high level concepts to low level fundamentals, it enables a clear picture of protocol level architecture and tradeoffs that might be relevant to an investment thesis.

Good Middleware also levels the informational playing field. Deep-pocketed participants with the ability to recruit teams of researchers can, in principle, have access to the same protocol-level information and intuition (not to mention data) as others.

In a space that's as crowded as crypto, an easy way to distinguish one's protocol from others' is to make it more accessible. This seems almost axiomatic to me. More intuitive explanations, more exposed wiring from protocol to foundational concepts, more step-by-step code guides from the basics to the state-of-the-art are sorely needed.

While top-down efforts by protocol teams would be very efficient (due to in-house expertise), crypto-native tools can be used very effectively to incentivize efforts at creating Educational Middleware.

Protocol teams and DAOs can incentivize such efforts through hackathon equivalents for education. Imagine a Proof-of-Work style competition in which the top-ranked contribution becomes part of the official documentation and gets a token bounty. Or a published list of resources in which the audience continuously ranks the most useful material.

The possibilities are endless, as are the avenues for creating good Educational Middleware. The only open question is: who will step-up to the plate? In raising my hands, I'm hoping that I'll find like minded folks to help fill the gap. For the long term health of the space, this is a need that desperately needs to be met.

As for Bittensor and its whitepaper, I'm currently on round 10. I've understood a lot more in attempts 6 through 9. As they say, the 10th time is the charm...

My sincere thanks for reading this far. If you liked what you read here, maybe you'll consider following me on Twitter and/or subscribing to this (completely free) letter?