It is a logical evolution for Web 3 platforms to incorporate native artificial intelligence (AI).
AI is influencing every software category so Web 3 shouldn’t be an exception. But there are fundamental, technical roadblocks about Web 3 stacks for the adoption of AI technologies.
In previous articles in CoinDesk, I discussed the relevance of AI techniques for decentralized finance (DeFi) and non-fungible tokens (NFT). Beyond understanding their clear value, it is important to see how AI can enter the Web 3 space in the near future, and what major roadblocks are currently preventing this to materialize.
Jesus Rodriguez is CTO and co-founder of blockchain data platform IntoTheBlock, as well as chief scientist of AI firm Invector Labs and an active investor, speaker and author in crypto and artificial intelligence.
“Software is eating the world” venture capital giant Marc Andreessen said in 2011, synthesizing the idea that companies operating in the physical world were transitioning to a digital one and that software would be their cornerstones.
Now, we can say that “machine learning (ML) is eating software” to pinpoint an oncoming trend in which most of the world’s software will be rewritten with AI/ML as its core building blocks. When you think about the omnipresent components of software applications, capabilities such as databases and identity come to mind. Intelligence, in the form of AI/ML models, is steadily becoming another foundational building block of modern software applications.
These days, software trends, including cloud computing, networking and cyber security are being reimagined with ML as a first-class citizen. Given that Web 3 is the next iteration of many of those software trends, ML will likely play a foundational role in the evolution of Web 3 technologies. Developing a thesis about the intersection of ML and Web 3 requires understanding both the trajectory of adoption of ML capabilities in Web 3 stacks as well as some of the fundamental challenges.
Layers of Web 3 intelligence
The addition of ML in Web 3 will not happen as an atomic trend; rather, it will be spread across different layers of the Web 3 stack. ML-driven intelligence can emerge in three key layers of Web 3.
The current generation of blockchain platforms has focused on building key distributed computing components that enable the decentralized processing of financial transactions. Consensus mechanisms, mempool structures and oracles are some of these key building blocks. Just as core components of traditional software infrastructures such as networking and storage are becoming intelligent, the next generation of layer 1 (base) and layer 2 (companion) blockchains will natively incorporate ML driven capabilities. For instance, we can think of blockchain runtime that uses an ML prediction for transactions to enable a massively scalable consensus protocol.
Smart contracts and protocols are another component of the Web 3 stack that will start incorporating ML capabilities. DeFi seems to be the prototypical example for this trend. We are not far from seeing a generation of DeFi automated market makers (AMMs) or lending protocols that incorporate more intelligent logic based on ML models. For instance, we can imagine a lending protocol that uses an intelligent score to balance the types of loans from different types of wallets.
Decentralized applications (dapps) are likely to become among the most likely Web 3 solutions to rapidly add ML-driven features. We are already seeing this trend in NFTs, but it’s going to become increasingly pervasive. The next-generation NFTs will transition from static images to artifacts that exhibit intelligent behavior. Some of these NFTs will be able to change their behavior based on the mood of their audience or the profile of new owners.
Top down, not bottom up
In considering layers of Web 3 intelligence, we might naively assume that a bottom-up adoption trend is most logical. Blockchain runtimes can become intelligent, and some of that intelligence can influence higher layers of the stack like DeFi protocols or NFTs. Yet, there are serious technological limitations that would force a top-down, instead of bottom- up, adoption of ML technologies in Web 3 stacks.
The root of these technological roadblocks trace to the architecture of the current generation of blockchain runtimes. In principle, blockchains are designed around a distributed computing paradigm that coordinates different nodes to perform computations that lead to a consensus about the processing of transactions.
Read More: Web 3 Is a Long Fight Worth Fighting
That approach contrasts to the state-of-the-art ML models that require complex, long-running computations for training and optimization which have been designed mostly for centralized architectures. This friction means that incorporating native ML capabilities in blockchain runtimes, although possible, is going to require some iterations.
DeFi protocols have fewer limitations from embracing ML features as they can rely on oracles and external intelligent agents that can fully benefit from existing ML platforms. And the limitation is almost non-existent for dapps and NFTs. From this perspective, we think the adoption of ML capabilities in Web 3 solutions is likely to follow a top-down trajectory going from dapps to protocols to blockchain runtimes instead of the opposite.
Intelligent Web3 is already here
The science fiction writer William Gibson wrote, “The future is already here – it's just not evenly distributed” to explain the trajectory of futuristic technology trends. The idea applies perfectly to the intersection of AI and Web 3.
The rapid evolution of ML research and technology in the last decade has translated into an overwhelming number of ML platforms, frameworks and APIs that can be used to add intelligent capabilities to Web 3 solutions. We are already seeing isolated examples of intelligence in Web 3 applications. so we can safely say that intelligent Web 3 is already here, just not evenly distributed.
Read More: Creating the On-Ramp for Web 3