Amazon Facebook

The Future of Hybrid Centralized-Decentralized Apps

Early last week, it was reported that the messaging app Telegram is working on raising capital for an audacious blockchain-based ecosystem of services and a network token called TON.  Some skeptics have mockingly called this the largest $1.2 billion seed round in history. Others claim that it is a more aggressive version of Kik’s Kin offering this summer.

I see what Telegram is attempting to do as thematically a very important next step for the internet. We are nearing the end of the road for the current development cycle of the web and internet services. The future is going to be services and companies that are able to effectively leverage centralized and decentralized paradigms in combination.

The Takeaway
Blockchain-based services are designed to be more trustworthy than centralized models, but they are also less efficient and more expensive to run. That means companies have to figure out where it makes sense to use distributed services to build trust and where it doesn’t.

That doesn’t mean that blockchain-based trusted digital memory is going to take over major functions on the internet. It means that the availability of highly distributed and trustable memory is going to enhance our ability to more deeply trust all sorts of services. That will unlock all sorts of digital interactions that were impossible before.

It Is All About Trust

There are three basic functions that allow people to coordinate their actions. Efficient communication requires trust in what is said (speech), who said it (identity) and what has been said before (memory).

Humans evolved over thousands of years to be very good at these functions in local-social contexts, which is a large part of our success as a species. Digitally, though, these three functions don’t inherently exist. We have had to recreate them from scratch.

What this means is that in the earliest days of the web, the standard advice given was to never use your real name, never post personal things and certainly not share your credit card. There was no trust in the web as a medium for speech or real identity, and so for a while it existed as a technical novelty more than serious human infrastructure.  

Much of the evolution of the internet over the last 20 years can be understood in the context of people attempting to build tools to establish trust in speech, identity and memory on top of the internet’s fundamentally stateless protocols.

Amazon did this in the commerce space, making it easy for consumers to trust it and, in time, its network of merchants.

Facebook did this in the social space. Starting with a clever social hack bootstrapped off of college networks, the company made it safe for the first time to use real names, establish relationships that mirrored the public world and share personal information.

These companies, and many others like them, succeeded in creating trust in identity and communication online by building massive centralized trust “proxies.” They solved the trust problem the same way organizations have been solving trust and communication problems for hundreds of years dating back to the British East India Company and U.S. mail service. Rather than needing to individually know and trust every node in a system, you can trust a single brand in the system and thereby everything downstream of it.

The Challenge of The Centralized-Trust Proxy Model

Centralized trusted proxies have all sorts of benefits for the world. They are amazingly efficient, and they are generally far more nimble and adaptable than alternative forms of management. But they do have some serious limitations, which are becoming more and more evident every day.

One is that your trust in the single proxy defines how much you are willing to trust all the people or businesses that you interact with through that proxy. That means that if the proxy isn’t fully trustworthy (and no proxy is), there is a ceiling on the trust you can have with anyone. That means there are certain things you might want to communicate that you can’t do on the major services (witness the rise of privacy-focused messaging apps).

The second limitation of trust proxies is simply a matter of the risk inherent in consolidation.  Things do go wrong in the world. And if or when you as an individual are too reliant on a single counterparty or proxy, the impact of error gets exponentially larger. As we have seen over and over, big proxy-services that people rely on become big targets for explicit manipulation by others. And sometimes there are just plain mistakes.

In this respect, there is diseconomy of scale for networks.

The third limitation with trust proxies is a simple matter of economics. The largest and most successful companies in the world are able to take a much longer-term time horizon than most, but they still are beholden to basic economics. Their economic interests do not always align with those of individuals using the network. That makes them unable to be fully trustworthy conduits.

Many pages of ink have been spilled on this tension, especially in the case of advertising-driven models, so I won’t go in depth here. Suffice it to say that even in the case of the best intentioned people and organizations, this affects the long-term trust in the central-proxy model for communication.

The Special Challenge of Memory in Centralized Trust-Proxy Model

On top of the general challenges with the trust proxy model, there are a heightened set of issues in the case of memory—above and beyond the challenges with identity and speech.

Of the three elements of communication, memory is by far the most challenging because it is the most expensive, complicated and highest-risk.

Think of this in normal human terms. It is one thing to trust a friend or colleague to pass a message on for you, but it is quite another to ask them to remember something for you. If they don’t pass the message on, you lose simply that last message and you know they failed you. You don’t trust them again, and you resend the message through other means.

The resource tax in running a secured decentralized service is the cost of security, not the cost of the transactions.

If, however, they forget what you told them (or if they choose not to remember it, or if a government tells them they have to forget), the damage can be much greater. The knowledge can be irrecoverable, (you can’t just resend the message). And you may not even realize they failed you.

On top of the extra structural risk associated with memory, it is of course far more expensive and less easy to directly monetize.

The Blockchain Solution for Memory and the Cost of Trust

So, given the challenge of trust proxies and the special challenge of memory, is it possible to create a trustworthy, secure, distributed solution for memory? No one had a good solution until 2009. But Satoshi Nakamoto’s paper, mixing cryptography and some smart economics, started a new path forward in the form of blockchains and open distributed ledgers with consensus mechanisms.

The idea that you could create a framework for cryptographically secured distributed trust is the fundamental breakthrough.

It is natural that the first application of the technology has been to create a series of digital assets in the form of currencies. After all, keeping a record of balance is simultaneously the simplest database, the highest ratio of value-per-bit. It also turns out that to get the distribution characteristics necessary for the system to work, value needs to be attached.

It is also natural that dozens of serious teams (and many more non-serious teams) are looking to extend the technology into all sorts of directions, including identity and verification solutions and ambitious goals like fully distributed Turing-complete compute solutions.

But, of course, all these benefits come at a cost.

Blockchains, relative to centralized services, are slow and highly inefficient. Proof of work models like Bitcoin burn massive amounts of resources and energy relative to their “centralized” alternatives, in order to achieve greater trust. Proof of stake models, which some see as the future, use less compute resources but far more bandwidth than centralized hub-and-spoke models. There are some ways to manage the distributed trust of blockchains more efficiently. But the reality is that blockchains will always be less efficient than centralized services by fundamental design.

People who uniformly look at blockchains and cry that they are “inefficient” with resources miss the point that you can’t compare them to centralized services. The resource tax in running a secured decentralized service is the cost of security, not the cost of the transactions. And there is an extent to which the cost of security absolutely should scale up with the amount of security you need given the sensitivity of what you are doing.

So, the upshot is that blockchain-based models are in many cases far more trustworthy than centralized models, but they are also fundamentally more expensive. You pay for the extra trust and always will—so the real game is to figure out exactly where the extra trust is worth the extra cost. The answer is clear that digital asset distribution via blockchains is worth the cost. I find it challenging to believe that the tradeoff works in many other cases.

The Near Future of Hybrid Proxy-Distributed Services

All of this brings us back to last week’s TON announcement from Telegram.

Thematically, it makes all the sense in the world for a messaging service like Telegram to push in the direction of decentralizing certain key components of what they do via a blockchain to enable them to build higher trust and higher risk services (like digital assets).

There is a fundamental limit to how trusted Telegram, as a traditional central-proxy run app, can ever become. So, it makes sense for it to bootstrap more trust through a distributed network, distributing “wallets,” and breathe value into “tokens” on top of that network.

In an ideal case, there are services (such as the most critical record of accounts, identity as cryptographic keys, etc.) that can be far more trustworthy once they’re distributed via a blockchain than they ever could be when run by a centralized proxy-based company.

But, all that said, a future where everything is run on top of blockchain-based computer networks in highly distributed frameworks isn’t worth the trade.

The key for companies like Telegram will be figuring out where there is value in building trust through distribution at the cost of efficiency, and where centralized services offer the right trust-efficiency tradeoff.

My belief is that in 10 to 15 years, most large companies will have a hybrid strategy of decentralizing the most critical components of what they do and centralizing where they need efficiency more than trust. But the path to that endpoint is going to be where all the value is created and lost. And likely just as many mistakes will be made over decentralizing as will be made by staying overly central for too long.


Sam is currently a General Partner at Slow Ventures and an intern at The Information. He is also the co-founder of Fin Analytics. He was formerly a vice president of product management at Facebook from 2010 to 2014. Prior to joining Facebook Sam founded drop.io, a file-sharing platform that was acquired by Facebook in 2010. Before drop.io Sam was an associate at Bain and Company. In his spare time Sam enjoys skiing and kite-surfing. He is married to Jessica Lessin, founder of The Information.