From Digital Scarcity to Abundance: How Crypto and AI Complement Each Other

At first glance, cryptocurrencies and artificial intelligence may seem like orthogonal technologies, each built upon fundamentally distinct principles and serving divergent functionalities.

However, a deeper exploration reveals an opportunity for the two technologies to balance each other's trade-offs, where the unique strengths of each technology can complement and enhance the other.

This notion of complementary capabilities was eloquently presented by Balaji Srinivasan at the SuperAI conference, inspiring a detailed comparison of how these technologies interact.

Cryptocurrencies operate on a bottom-up approach, emerging from the decentralized efforts of anonymous cyberpunks and evolving over more than a decade through the coordinated efforts of numerous independent entities worldwide. In contrast, AI is developed through a top-down approach dominated by a handful of tech giants. These companies dictate the pace and dynamics of the industry, with barriers to entry shaped more by resource intensity than by technical sophistication.

These two technologies also have a distinct nature. In essence, cryptocurrencies are deterministic systems that generate immutable outcomes, such as the predictable nature of hash functions or zero-knowledge proofs. This sharply contrasts with the probabilistic and often unpredictable nature of AI.

Similarly, crypto technologies excel in verification, ensuring the authenticity and security of transactions and building trustless processes and systems versus AI which focuses on the generation and creating the abundance of digital content. In the process of creating digital abundance, however, lies a challenge of ensuring content provenance and preventing identity theft.

Luckily, crypto offers the antithesis to the concept of digital abundance - digital scarcity. It offers relatively mature tools that could be extrapolated to AI technologies to create guarantees of content provenance and avoid the issues of identity theft.

One notable strength of cryptocurrencies is their ability to attract substantial hardware and capital into coordinated networks serving specific objectives. This capability could be particularly beneficial for AI, which consumes vast quantities of computational power. Mobilizing underutilized resources to offer cheaper computing could significantly enhance AI’s efficiency.

By juxtaposing these two technological giants, we can appreciate not only their individual contributions but also how they might together forge new pathways in technology and economy. Each offsets the other’s trade-offs, creating a more integrated, innovative future. In this blog post, we aim to explore the nascent crypto x AI industry map, highlighting some emerging verticals at the intersection of these technologies.

Compute Networks

The industry map begins with Compute Networks which are trying to address the challenges of the constrained GPU supply side and attempt to lower the compute cost in distinct ways. Worth highlighting are the following:

• Non-uniform GPU Interoperability: very ambitious attempt that carries high technical risk and uncertainty, but if successful, it would have the potential to create something of enormous scale and impact, making all of the compute resources fungible. Essentially, the idea is to build compilers and other prerequisites such that on the supply side, you could plug in any hardware resources, and on the demand side, all of the hardware non-uniformity would be fully abstracted such that your compute request could be routed to any resource in the network. Should this vision become successful, it would lower the moats of CUDA software which is a completely dominant solution for AI developers today. Again, the technical risk is high and many experts are highly skeptical on the feasibility of this approach.
• High-Performance GPU Aggregation: integrating most in-demand GPUs across the globe into one distributed & permissionless network without worrying about interoperability across non-uniform GPU resources.
• Commodity Consumer GPU Aggregation: Points towards aggregating some of the less performant GPUs that might be available in consumer devices and that present the most underutilized resource on the supply side. It caters to those willing to sacrifice performance and speed for cheaper, longer training processes.

Training and Inference

Compute networks are being leveraged for two primary functions: training and inference. Demand for these networks comes from both Web 2.0 and Web 3.0 projects. In the realm of Web 3.0, projects like Bittensor utilize the compute to perform model fine-tuning. On the inference side, Web 3.0 initiatives emphasize the verifiability of processes. This focus has led to the emergence of verifiable inference as a market vertical, where projects are exploring ways to integrate AI inference into smart contracts while maintaining the principles of decentralization.

Agent Platforms

Moving on to Agent Platforms, the map outlines the core issues that have to be addressed by startups in this category:

• Agent interoperability and the ability to discover and communicate with each other
• The ability for agents to build collectives and manage other agents
• Ownership and marketplace for AI agents

These features emphasize the importance of flexible and modular systems that can integrate seamlessly across various blockchain and AI applications. AI agents have the potential to completely change the way we interact with the internet and we believe that agents would leverage crypto infrastructure to power its operations. We envision AI agents relying on crypto infrastructure in the following ways:

• utilizing distributed crawling networks to access real-time web data,
• using crypto payment channels for agent-to-agent payments,
• requiring economic stakes not only to enable punishments in case of misbehavior but also to improve agent discoverability (i.e. utilizing stake as an economic signal in the discoverability process),
• leverage crypto consensus to determine what events should result in slashing,
• open source interoperability standards and agent frameworks to enable building composable collectives,
• rely on immutable data history to evaluate past performance and choose the right agent collectives in real time.

Data Layer

A core component of the Crypto-AI convergence is data. Data is a strategic asset in the AI competition race and along with compute the key resource. Yet, it is often an overlooked category as much of the industry’s attention is focused on the compute layer. There are many interesting angles where crypto primitives offer value in the data acquisition processes, the two high-level directions being:

1. Access to public Internet data
2. Access to data in walled gardens

The former one is about building a network of distributed scrappers that could crawl over the internet and obtain access to massive datasets in a matter of days or provide real-time access to very specific data on the internet. However, to be able to scrape the massive datasets on the internet the network requirements are very high, about few hundred thousand nodes at least to start with some meaningful workloads. Fortunately, Grass, a distributed network of scrapping nodes, already has more than 2M nodes actively sharing internet bandwidth to the network with the objective of scrapping the whole internet. It shows the huge potential of crypto-economic incentives in attracting valuable resources.

While Grass levels the playing field when it comes to access to public data, there is still the issue of tapping into the latent data potential - proprietary datasets. Namely, there is still a ton of data that is kept in privacy-preserving ways due to its sensitive nature. Several startups are working around utilizing some encryption and cryptography tooling to enable AI developers to leverage the underlying data structure of proprietary datasets to build and fine-tune large language models while keeping sensitive information private.

Techniques like federated learning, differential privacy, trusted execution environments, fully homomorphic encryption, and multi-party computations offer varying levels of privacy and trade-offs. A great overview of these technologies is summarized in the research post by Bagel. These technologies not only protect data privacy in machine learning processes but can also be implemented at the compute level for comprehensive privacy-preserving AI solutions.

Data x Model Provenance

Data and model provenance techniques aim to establish processes that provide guarantees to the users that they are interacting with intended models and data. Moreover, these techniques provide the guarantees of authenticity and origin. Take watermarking for an example. Watermarking, one of the model provenance techniques, embeds signatures directly into the machine learning algorithms, more specifically directly to model weights, such that upon retrieval you could verify that the inference came from the indented model.

Applications

When it comes to applications, the design landscape is limitless. In the industry map above, we list some use cases we are particularly excited to see develop with the implementation of AI technology in the Web 3.0 sector. As most of these use cases are self-descriptive, we won’t provide additional commentary at this point. However, it is worth noting that the intersection of AI and Web 3.0 has the potential to restructure many verticals in the crypto space as these new primitives introduce more degrees of freedom for developers to create innovative use cases and optimize existing ones.

Conclusion

The convergence of crypto and AI presents a landscape ripe with innovation and potential. By leveraging the unique strengths of each technology, we can address their respective challenges and forge new pathways in technology. As we navigate this nascent industry, the synergies between crypto and AI will likely drive advancements that reshape our future digital experiences and the way we interact on the web.

The fusion of digital scarcity with digital abundance, the mobilization of underutilized resources for computational efficiency, and the establishment of secure, privacy-preserving data practices will define the next era of technological evolution.

However, it is crucial to recognize that this industry is still in its infancy, and there is a risk that the current industry map could become obsolete in a short period. The rapid pace of innovation means that today’s cutting-edge solutions may quickly be surpassed by new breakthroughs. Despite this, the foundational concepts explored—such as compute networks, agent platforms, and data protocols—highlight the immense possibilities at the intersection of AI and Web 3.0.