IOSG Ventures: In-depth exploration of New DeFi, unlocking the potential of data

Original Author: Momir, IOSG Ventures

Smart contracts are limited because they lack the ability to interact with the environment, which limits the potential for decentralized applications (dApps). In order to achieve more and more complex functions, DeFi protocols have two options: they can adopt a flexible design, such as players can personalize various scenarios; or they can introduce external dependencies - relying on off-chain infrastructure, such as oracle , keepers, or off-chain computation — to maintain a simple user experience.

In a recent thought-provoking article titled "Why DeFi is broken and how to fix it — Part 1: Oracle-less protocols", Dan Elitzer advocates the use of DeFi primitives with zero external dependencies to minimize attack vectors . The idea is to remove the need for trust in third-party institutions. However, a zero-dependence DeFi ecosystem will have higher requirements for specialization. Most users lack the time, expertise, or resources to become market makers on Uniswap v3, or assess the quality of collateral in the protocol without external dependencies, and they have to rely on trusted intermediaries to participate.

Thus, the quest for zero dependencies may bring us back to square one, or worse, force non-expert users to trust complex entities or deposit funds into transitional smart contracts, which increases insecurity. Rather than fighting to eliminate external dependencies entirely, consider more pragmatic approaches such as putting external dependencies under stricter scrutiny and limiting potential black swan scenarios. We must recognize that some degree of dependence is unavoidable and even crucial to the development of the industry.

Among the well-known DeFi projects, the early version of Uniswap came closest to achieving zero dependencies. However, the recent introduction of Uniswap v4 demonstrates a shift towards a highly modular approach (“Hooks”) to move the field forward.

Data Primitives

Discussions about external dependencies revolve around the ability of smart contracts to interact with external data. Today, data interactions often rely on oracles to access off-chain information, albeit in a limited scope (mainly including prices of major cryptocurrencies).

As more and more activities migrate to the blockchain, a wealth of valuable on-chain data can be used to enhance mechanism design algorithmically and transparently. However, despite the transparency of on-chain data, integrating it with smart contracts is not an easy task. Reading, processing and delivering meaningful data requires a sophisticated and trusted infrastructure. As a result, developers often rely on existing tools for their data needs. However, most existing data solutions are rooted in Web 2.0 frameworks, and even more Web 3.0 native protocols cannot guarantee the accuracy of the data they provide.

Sushiswap Discussion About Inaccurate Data Sending From Polygon Sushi-Matic Subgraph

Considering that smart contracts can even manage billions of dollars in deposits, it is neither desirable nor practical for them to connect directly to a trusted API source, as this reliance would undermine the decentralized nature of the blockchain ecosystem.

Build a tamper-proof data solution

Our investment philosophy revolves around a fundamental belief that tamper-proof data will be the cornerstone of the next generation of DeFi protocols. However, achieving data tamper resistance is not a simple task and requires complex infrastructure and extensive optimizations to make it economically feasible by design.

In this context, Space and Time has become a pioneer in building a tamper-proof data infrastructure. A key part is its SQL proofs, an improvement over SNARK proofs designed specifically for querying data from relational databases. This approach provides guarantees that the query and its underlying data have not been tampered with. Additionally, it provides guarantees of data validity when retrieving data from archive nodes via RPC calls.

Some other well-known trustless data primitive projects include but are not limited to Nil Foundation, Axiom, Brevis, Herodotus, etc.

Tamper-proof data opens up new horizons for DeFi protocols, allowing them to push the boundaries of functionality, driving further growth and innovation in the industry.

Below we discuss data-driven protocol design optimization when:

1. Personalized user experience

2. Self-parameterization protocol

3. Protocol Economy

4. Qualified Access

1. Personalized user experience

In the technical business world, it is commonplace to provide users with tailor-made services. However, smart contracts (essentially strings of code representing some business logic) often unify the user experience, which often equates to poor user experience. For example, on some lending platforms, user A is a novice, user B is a long-term agreement user, and user C is a veteran transaction. This lack of differentiation fails to account for user behavior and misses opportunities to enhance user stickiness, incentivize positive behavior, and optimize capital utilization.

Protocols have a vested interest in identifying user behavior and adjusting accordingly. For example, by leveraging credit ratings, offering cheaper credit or lower mortgage rates to well-performing customers. Such a project will naturally attract users from platforms with uniform terms. Furthermore, this approach provides users with implicit incentives to perform good behavior in order to obtain more favorable terms.

Thinking in terms of fintech, where companies like SoFi gain market share by refusing to unify, DeFi dApps can learn as well. For example, SoFi found a market inefficiency in the student loan market, where Stanford graduates were charged the same loan interest rates as other borrowers, even though they were more likely to land high-paying jobs after graduation. SoFi has had notable success by adjusting rates to better reflect users' risk profiles.

Likewise, in the DeFi space, we envision an opportunity to innovate protocols that factor user risk into interest rates and collateralization. Care must be taken, however, not to undercollateralize lending based solely on existing historical data, which becomes irrelevant when game theory changes.

It’s worth mentioning that projects like Spectral and Cred Protocol are trying to build credit scoring models from on-chain data. However, these projects all run on centralized databases, so as long as the data and models they serve come from centralized data and can be easily tampered with, it is unlikely that major DeFi protocols will connect to their APIs. Instead, if these projects adopt tamper-proof solutions, they have the potential to become ubiquitous DeFi credit oracles, powering a range of innovative applications.

2. Self-parameterizing protocols (minimizing governance intervention)

Many DeFi protocols still rely on manual governance processes, often directed by off-chain consulting firms, to tune their parameters. AAVE, for example, pays heavily to external consulting firms to monitor and guide protocol risk parameters.

However, this approach creates several problems:

1. Lack of real-time support: The system lacks the ability to respond to changing market conditions or emerging risks.

2. Manual systems: reliance on human intervention introduces latency issues and potential inefficiencies when tuning protocol parameters.

3. Trust in off-chain entities: Reliance on external consulting firms raises concerns about transparency and the methodology used in making recommendations.

This static approach was exposed in an attack on AAVE, leading to bad debts that could have been avoided with suitable lending parameters that better reflect the liquidity of borrowed tokens. Additionally, the risks of using circulating tokens as collateral in lending protocols have not been adequately addressed.

To address these limitations, projects should transition to real-time, automated, transparent, and trustless designs. For example, lending protocols could leverage infrastructure like Space and Time to monitor data in real time. This will allow them to dynamically adjust collateral, borrowing parameters, and other key parameters.

Likewise, exchanges could introduce dynamic fee structures based on volatility or impermanent losses. Many liquidity pools on Uniswap v3 are difficult to achieve sustainable operation, mainly because they cannot dynamically charge LPs. With the Hook of Uniswap v4 or the module of Valantis, dynamic fees are possible.

Additionally, aggregators can be free from human labor and fixed fees to adapt to the changing risks and rewards of the underlying protocol. The collaboration between Spool and Solity is a step in this direction, with Solity using a big data approach to analyze the risk-reward of pools.

3. Protocol Economy

A data-driven approach has the potential to enhance protocol economics and token economic models in DeFi, where projects can share incentives with eligible users.

For example, a DEX aggregator looking for user stickiness and loyalty, they can allocate slippage benefits to users who meet certain conditions, such as executing a specified number of transactions and reaching a minimum transaction volume.

Such incentives heavily incentivize early adopters, build loyalty within the user base, and provide incentives directly to existing users to promote usage of the protocol within their own population.

4. Qualified Access

While blockchain has a permissionless nature, it also allows for freedom of choice. In multiple cases, permissioned access at the application layer can ensure that the protocol is not used to do evil, or effectively interact with the intended user base.

For example, privacy protocols like Tornado Cash are under scrutiny from regulators because they may be used for money laundering or other illegal activities. To prevent money laundering, protocol developers can take steps to prevent bad actors from interacting with their platforms.

Also, for market makers, knowledge of counterparties is extremely valuable, but such information is often not available to dexes. Assuming it is possible to use data to build proofs of real people, DEXs can only allow non-bot addresses to interact, then this kind of problem can also be solved.

Requirements for Verifiable Computation

What was discussed in the previous section can be fully implemented through integration with trustless data primitives. However, others will require additional resources to perform statistical computations or machine learning. For example, credit scoring programs can leverage tamper-proof data, but still require machine learning algorithms to generate credit scores.

Or in the context of a Risk Oracle, having access to data about the circulating supply, volume, transaction count, number of holders, time since TGE, etc. of a particular token is critical to determining appropriate collateral and lending factors. However, machine learning techniques need to perform precise calculations on the basis of this data.

source：

Other areas in DeFi that require more complex calculations include, but are not limited to:

• Revenue aggregator: Estimate the revenue and risks of the underlying protocol and find the optimal allocation.
• Portfolio optimization: Calculate target portfolio allocation based on pre-determined criteria, change directional exposure based on technical indicators, etc.
• Decentralized exchange of derivatives: systemic risk management, capital cost adjustment, derivatives pricing, etc.
• Advanced trade execution algorithm
• Liquidity vault market-making logic
• Clearing library

Projects like ChainML address this need by providing a verifiable off-chain computation layer, powered by a purpose-built consensus mechanism. Others that build distributed machine learning computing layers include but are not limited to GenSyn, Together.xyz, Akash, etc.

Likewise, ZKML presents an interesting opportunity where ZK proofs can compress computations into succinct proofs that can be verified on-chain, or demonstrate the use of a particular model without revealing its properties. Such as Modulus Labs, Giza and other ZK projects.

However, implementing machine learning in ZK is currently very expensive, making practical implementation challenging. While hardware acceleration and circuit optimization may improve performance in the future, the computational demands of AI are expected to grow at a faster rate, making ZKML limited to niche computing methods that cannot be adapted to state-of-the-art AI models. Therefore, approaches such as the consensus-based pessimistic approach or the fraud proof-based optimistic approach offered by projects like ChainML may be the best opportunity to integrate the latest artificial intelligence algorithms into Web 3.0.

Summarize

The fusion of tamper-proof data, advanced computing power, and data-driven decision-making has the potential to unlock new innovations, improve efficiency, and user satisfaction in the DeFi ecosystem. While this article focuses on optimizations that can be made on top of on-chain data primitives, we are equally bullish on the opportunities presented by integrating various off-chain data via zk proofs. We believe that data will enhance on-chain and off-chain interoperability and promote integration between decentralized finance and traditional financial systems.

As the industry continues to evolve, the protocol must embrace emerging technologies, cooperate with leading projects, and prioritize transparency and trustlessness, which can not only build a strong and sustainable future for DeFi, but also contribute to the impact of DeFi on the global financial landscape. This vision offers the possibility to have far-reaching impact.

Disclaimer: Space and Time, ChainML, Nil Foundation, and Solity are IOSG Portfolios.

references:

Crypto x AI:

ZKML:

eco: