The cost of trust

The opening lines of the Bitcoin whitepaper shape a prism through which we can perceive blockchains: a network atop which users (whether humans, bots, AIs, or any entity capable of utilizing a private/public key) conduct transactions without the need for trust. Initially, Satoshi Nakamoto wrote about the removal of trust for payments, but we are now able to apply this same idea to pretty much anything on the internet: exchange of information, games, data, and much more. This concept of removing the trust, typically required in traditional systems, gave birth to one of the most used words in the crypto space: "trustless," and it was also one of the main points of criticism from the TradFi world (for example, this paragraph from Matt Levine). It has now become mostly a meme, with most of the new teams building on blockchains choosing "growth" over trustlessness (here, for example). Not that I want to refute this, but we will try to provide a more nuanced vision, one that finally goes beyond the alignment meme and provides tangible value for trustlessness.

How trust affects productivity

There are numerous examples we could draw from history to understand the cost of trust and how it can reshape the way we organize society, but I think the most striking one is the creation of limited liability companies. Before the creation of this legal structure, the cost of trust between investors and entrepreneurs was immense: entrepreneurs had to bear all the risk personally, and investors had to trust entrepreneurs with their assets. LLCs provided a legal framework that enabled entrepreneurs to take risks and innovate while providing clarity and safety for investors. They removed much of the cost of mediation on both sides, including legal costs, and thus improved the overall productivity of society. 

Now let's consider blockchains: they remove the need for trust for any transaction on the internet. There's no cost of mediation when you transact on Ethereum or Bitcoin: once your transaction is included in a block, there's no way to go back (except for reorgs - you'll actually have to wait a few blocks). There are also no intermediary fees: you can send, lend, borrow, and do thousands of other things without intermediaries. Blockchains provide trustless trust, and just as LLCs improved human productivity by reducing the cost of trust in business interactions, blockchains will improve the world's productivity (because yes, even AIs can have access to this) by reducing the cost of trust in any transaction.

What needs to be very clear here is that the trustless trust provided by blockchains is possible because blockchains operate using deterministic data. This means that there's a state A that is accepted by everyone (consensus), and it is only possible to transition to state B if everyone agrees (more than 51%, generally) on the new state—and the agreement is only possible because state transitions are verifiable: a signature is either correct or not. Thus, social consensus is straightforward: cryptographic truth exists, and anyone can verify that state transitions are valid.

Trust in oracles

As previously stated, because blockchains operate solely on their own state, anyone seeking to leverage the trustless trust provided by blockchains for something reliant on external data encounters the oracle problem. This problem can be reframed as "Is it possible to provide trustless trust for any data?" While there have been numerous attempts to address this, we must acknowledge the true bottleneck of this issue: non-deterministic data. Cryptographic truth does not apply to non-deterministic data, making it impossible to achieve the pure trustlessness attained by blockchains. 

So, how can we effectively reduce the trust associated with non-deterministic data? This is the problem we are working to solve at Pragma, and it extends far beyond crypto—it encompasses providing a general truth for questions where knowledge is not universally available, constructing trust systems for information that is not distributed or public, and overall improving the decision-making processes of organizations. 

Now, let's return to our oracle problem: providing trustless trust for non-deterministic data. Most protocols attempting to solve this problem have adopted a structure similar to blockchains, as this was the approach that succeeded with deterministic data. If you look into oracles today, while there may be different technical choices, they largely follow the same pattern: a network of nodes reporting data (often price feeds) and reaching consensus through an aggregation mechanism (such as taking the median). However, this design falls short on several fronts. Firstly, trust assumptions do not align with blockchains; oracles are permissioned (nodes must be approved by devs), slashing mechanisms are absent, and everything relies upon SLAs, introducing a cost for trust. 

The cost of trust is the sum of expenses, risks, and inefficiencies that arise when relying on intermediaries or centralized entities to facilitate transactions or verify information. These costs include legal and mediation fees, intermediary charges, the risk of malicious actors or shutdowns, regulatory burdens, and security vulnerabilities. 

Furthermore, as non-deterministic data lacks a single "truth", relying on a centralized decision-making system is nonsensical. Consider a price feed as an example. You can obtain the ETHUSD feed from various oracles, but what does it even mean? If you visit the websites of Chainlink, Pyth, and Chronicle simultaneously, you will find differing prices for the same feed at the same time. This discrepancy arises because there is no singular "price" of ETH; rather, there are multiple markets with varying prices, and one can aggregate these markets, or a subset thereof, using different aggregation methods. Given the inherent variability in non-deterministic data, it is illogical to expect different protocols with distinct purposes to rely on the same pricing, parameters, and sources. Why would a lending protocol on Ethereum require the exact same pricing, parameters, and sources as an options protocol on L2? This example only pertains to ETH, but there are countless derivatives, stablecoins, staked tokens, and other assets that cannot conform to the same model—let alone entirely different types of data, such as election results or randomness requests. 

For these reasons, solving the oracle problem necessitates departing from the blockchain approach. We require a novel approach that considers all the intricacies of the problem, which is a worthy endeavour as it will enhance the efficiency of systems and organizations by reducing the cost of trust.