The Most Important Infra of AI Agent Personalization: Nillion

Key Takeaways

• The market's interest in the convergence of AI and cryptocurrency is skyrocketing. However, their technical limitations remain apparent, with one of the most significant challenges being "how to personalize these technologies?" For agents to become personalized, they need access to individuals' sensitive information, which poses a serious threat to personal privacy.
  

• For AI to further penetrate our daily lives, we need infrastructure that can process sensitive data while protecting individual privacy. The Nilion Network is an infrastructure designed specifically for such blind computation.
  

• Nilion's blind computation technology can enable various use cases beyond agent personalization. For instance, it is essential in scenarios that require access to sensitive information, such as medical diagnoses or inter-company data utilization. Additionally, it can serve to complement the drawbacks inherent in blockchain transparency.
  

• Of course, blind computation is by no means a simple task, and it will be interesting to see how Nilion can scale this blind computing technology more effectively in the future.

Currently, the blockchain industry is abuzz with discussions about the convergence of AI and crypto. At the center of this conversation is Marc Andreessen, Managing Partner at the prominent Silicon Valley VC firm Andreessen Horowitz (a16z). Andreessen donated approximately $50,000 worth of Bitcoin to an AI project called Terminal of Truth, ostensibly to help them build better hardware, improve algorithms, and develop their community platform. The project gained significant attention when it discovered and aggressively promoted a memecoin called '$GOAT', which at one point reached a market capitalization exceeding $1 billion, making a substantial impact on the market.

GOAT and Terminal of Truth were just the beginning. The AI agent boom triggered by GOAT spread to projects like ACT (another project backed by Andreessen) and ai16z (a decentralized AI-based fund inspired by a16z). Notably, ai16z's agent framework Eliza continues to attract attention, while in the Base ecosystem, Virtuals and its developing agents have garnered significant interest, successfully establishing 'AI x Crypto' as a new market trend.

Some skeptics question whether this AI x Crypto trend is merely riding the coattails of the rapidly growing AI industry without substantial use cases. However, the trend is moving beyond memes to create tangible applications. Infrastructure promoting decentralized AI, like TAO, is emerging, along with decentralized GPU marketplaces essential for AI computing. Therefore, rather than dismissing this trend as mere hype, we should seriously consider how blockchain technology and AI, particularly agents, can be integrated.

Currently, agents demonstrate capabilities in market analysis, trading execution, and social media user communication. However, for agents to achieve greater scalability, they need access to more extensive information. For instance, for an agent to act on behalf of an individual by learning their preferences, hobbies, and values, it would need to process personal privacy-related data. Even for non-personalized applications, agents would need access to sensitive data typically difficult to obtain in order to permeate various aspects of life.

Agents, however, are learning entities. Exposing sensitive data to them raises serious privacy concerns. Should agents be limited to learning from restricted data sets? What if these agents could learn from sensitive data without directly accessing it? In other words, what if computations could be performed on encrypted information without decryption? This would enable agents to learn from vast datasets without compromising sensitive information.

Surprisingly, there is a project attempting this very thing: Nillion. This article will explore how Nillion enables computation on encrypted data without decryption and examine potential agent use cases that could be implemented through Nillion.

1. What the Hack Is Nillion?

In essence, Nillion is "a blind computation network that enables data storage and computation without decryption." Structurally, the Nillion network consists of two layers: PETnet (ORCHESTRATION LAYER) utilizing privacy-enhancing technologies, and Nil Chain (COORDINATION LAYER), the blockchain network that PETnet uses for Coordination.

Before delving into the Nillion network, understanding blind computation is crucial. What is Blind Computation and what are its core principles?

1.1 What the Hack Is Blind Computation? - PETnet Explained

Nillion's PETnet consists of multiple nodes that can transfer, store, and compute data without directly accessing it. For example, these nodes can sign transactions on your behalf. While traditional computation requires your private key for transaction signing, blind computation distributes your private key into pieces called "shares" among nodes. Through cryptographic protocols, nodes holding different shares can collectively sign transactions without reconstructing the private key, thus preventing any information leakage. This is blind computation.

1.2 Why Do We Need Blind Computation?

Beyond agent personalization, blind computation is a critical technology. For Web3 services to match Web2 capabilities, they must securely handle private data. The key distinction between Web2 and Web3 lies in "accessible data scope." Web2 operates on private databases, limiting personal information exposure. However, Web3's blockchain transparency allows universal access to on-chain data, creating a significant challenge. While Web3 needs to address various issues including scalability, data privacy protection remains crucial for achieving Web2-level services.

This isn't to say Web2 services are superior to Web3 in all aspects. We've historically surrendered personal data to Web2 companies without protection, granting them unrestricted access to private information. The 2018 Facebook Senate hearings following the Cambridge Analytica scandal - where millions of users' personal data was collected without consent - revealed how carelessly user data was managed. Blockchain emerged as a proposed solution to these issues.

However, blockchain isn't a perfect solution either. While it can address data ownership issues, its inherent transparency paradoxically limits its use cases by preventing private data storage on-chain. This is where blind computation becomes essential.

Going beyond simple encrypted data storage, the ability to compute on data without decryption allows us to simultaneously: 1) resolve user data ownership issues through blockchain technology, and 2) overcome data onboarding limitations caused by blockchain transparency through blind computation.

1.3 How It Works: Introducing PETnet.

We've explored what blind computation is and why it's necessary. Now, let's examine how this complex blind computation actually works through Nillion's PETnet. PETnet enables blind computing using various privacy-enhancing technologies (function secret sharing, zero-knowledge proofs, fully homomorphic encryption, trusted execution environments), but this article will focus on Nillion's multi-party computation technology. What is Multi-Party Computation (MPC), how does Nillion's MPC protocol differ from traditional approaches, and how does it work?

1.3.1 Multi-Party Computation in a Nutshell

Multi-Party Computation (MPC) is a cryptographic protocol allowing multiple participants to jointly perform calculations without revealing their secret inputs. This technology ensures calculation accuracy while protecting participants' privacy, even in the presence of malicious actors. MPC operates by dividing each participant's input into 'shares', performing distributed computation using these shares, and then reconstructing the result. Due to these characteristics, MPC is utilized across various fields including privacy-preserving data analysis, secure auction systems, secret voting, biometric authentication, and blockchain, gaining attention as an innovative technology enabling complex calculations while maintaining data confidentiality.

1.3.2 Protocol for Efficient Non-linear Operations: Curl

Nillion has developed an MPC protocol, called Curl, in collaboration with Meta and the University of California Irvine to enable the efficient evaluation of complex operations. What's particularly interesting about Curl is that it extends beyond the Linear Secret Sharing Scheme (LSSS) format traditionally adopted by many MPC protocols, enabling efficient operations even when the relationship between input and output isn't linear (They will start with a linear secret sharing scheme and plan to expand from there). Traditional MPC protocols based on linear secret sharing excel at simple operations, such as adding secrets or multiplying a secret by a public constant (e.g., outputs doubling when inputs double). However, they often face challenges with more complex operations. In contrast, Nillion’s Curl MPC approach supports the efficient evaluation of complex computations (such as division, square roots, trigonometric functions, and logarithms) making it highly scalable and better suited for modeling real-world problems where outputs don’t necessarily scale linearly with inputs. This expanded capability positions Nillion’s protocol as a versatile and powerful framework for advanced data processing. For example, in privacy-preserving AI models, where user inputs (such as prompts in a large language model) need to remain confidential, Nillion’s Curl protocol accelerates the primary bottlenecks around the evaluation of non-linear functions like activation functions and normalization layers, enabling secure and efficient computations without compromising performance.

Nillion's Curl MPC protocol consists of two main phases:

1. Pre-processing to Create Shares This phase prepares randomness for future computations by dividing it into multiple shares and distributing them to participants (computation entities) before processing actual information. While this process is independent of the input values themselves, it depends on the total number of inputs and participants, as this determines how many shares need to be created before computation.
   

2. Efficient Computation of non-linearities This phase involves actual computation on secret-shared data and comprises three sub-stages:

• Input Stage. User(s) distribute(s) shares of their inputs to the computation to the participants. Each participant receives one share per input value.
  

• Evaluation Stage. Participants efficiently compute operations on input shares using Nillion’s Curl protocol.
  

• Output Stage. Participants reveal their locally computed results, which are then aggregated to produce the final outcome.

In summary, Nillion enables computation on encrypted data without decryption through MPC technology, particularly enhanced by its non-linear approach that enables more complex operations.

1.4 Nillion's Second Pillar: Nil Chain

However, what we've explored so far is just one pillar of Nillion. What's the ultimate purpose of Nillion's blind computation? It's to create various use cases utilizing this technology. These use cases will be implemented on a blockchain network, and this is where Nil Chain comes in.

Nil Chain is a Cosmos SDK-based blockchain network serving as the Coordination Layer in the Nillion network. Users pay fees through Nil Chain, and without verification of fee payment from Nil Chain, users cannot receive blind computation results from PETnet. Nil Chain coordinates all actions within the Nillion network and operates without smart contracts.

1.4.1 What About the Absence of Smart Contracts?

How can Nillion be utilized without smart contracts? There are two main approaches: 1) Pure off-chain applications using only Nillion, and 2) Applications leveraging Nillion's blind computing while using existing blockchains.

For approach 1), applications like password managers or AI inference applications can be implemented directly on Nillion without requiring smart contracts. For approach 2), where smart contracts are necessary, existing blockchains (like Arbitrum, NEAR, Avalanche, or Solana) can handle transaction settlement while leveraging Nillion for privacy-related aspects.

Let's examine specific examples of how Nillion can be utilized.

2. Nillion Use Cases - Solving Real World Problems

As consistently mentioned throughout the article, Nillion's unique capability to compute on data without decryption enables new possibilities for sensitive data operations. Nillion is expected to solve not just Web3 challenges, but broader societal issues. Let's explore specific use cases.

2.1 AI Agents

As mentioned in the introduction, for AI to function as a personal assistant handling individual daily tasks, it requires access to highly private data. Without blind computation, there's virtually no secure way to train AI on personal data. Therefore, Nillion's blind computation could become crucial infrastructure for perfectly personalizing AI assistants.

The Nillion ecosystem is rapidly expanding with significant advancements in both foundational platforms and innovative applications committing to build on the network in this area. Leading AI infrastructure platforms such as Virtuals, Capx, Ritual, and Skillful have signed agreements or committed to integrate Nillion. On the application front, end-user solutions like Pindora (personal social agent), Fulcra (agent health advisor), and Space of Mind (agent mental health advisor) are actively building on Nillion to build out their applications.

2.2 Healthcare

In the healthcare sector, data privacy presents a significant challenge. While healthcare data is protected by stringent legal and regulatory frameworks, these same protections often create barriers to effective data sharing and analysis. Medical professionals and researchers face considerable obstacles when trying to access and analyze sensitive information while maintaining patient confidentiality.

This is where Nillion steps in, offering a groundbreaking approach to healthcare privacy. Their technology enables organizations to store, process, and analyze sensitive healthcare data while keeping it fully encrypted. Through their decentralized architecture, healthcare institutions can collaborate effectively and securely, opening up new frontiers in personalized medicine, predictive healthcare analytics, and broad-scale medical studies.

Several innovative companies are already putting Nillion's technology to work in the healthcare space. MonadicDNA, for instance, has developed a secure personal genomics platform using Nillion's infrastructure to protect sensitive genetic information throughout the entire data lifecycle. This allows them to provide valuable health and ancestry insights while maintaining robust protection against data breaches and unauthorized access.

In the mental health sector, Space of Mind is leveraging Nillion's technology to provide evidence-based mental health support through group therapy sessions. Their platform incorporates AI-driven tools while addressing privacy concerns head-on - for example, enabling private analysis of journal entries' sentiment while maintaining user privacy and trust.

2.3 Data Marketplaces (e.g., Data-to-Earn)

Data marketplaces built on Nillion's technology are revolutionizing the handling and monetization of personal data. In today's digital world, individuals create massive amounts of data through their daily digital interactions, but this information is typically locked away in isolated, centralized databases. Nillion provides the underlying privacy infrastructure that enables individuals to consolidate their data and maintain true ownership while allowing potential buyers to analyze and utilize this data securely.

This innovative approach opens up new possibilities for individuals to benefit from the expanding data economy without sacrificing control over their personal information. Nillion's technology plays a crucial role in these marketplaces by facilitating the secure and active utilization of data while maintaining encryption and protecting sensitive information from unauthorized access.

Several pioneering projects are already leveraging Nillion's capabilities. ZAP stands out with their focus on analyzing user data pools to create new AI training datasets while delivering valuable user insights. Their initial product, A.I. Lingo, has already attracted nearly 40,000 users. Fulcra is another notable example, specializing in generating collective insights from smart device data. Meanwhile, Humanity works on integrating data across various platforms - social, health, and entertainment - to help users monetize their personal insights.

The ecosystem continues to grow with other innovative projects like Reclaim Protocol, HealthBlocks, and Dwinity, all of which are harnessing Nillion's blind computation technology to create robust data economies and give users new opportunities for data ownership and monetization.

2.4 Medical Diagnostics

The medical field has a high demand for blind computation due to the sensitive nature of healthcare data. Medical institutions face significant restrictions in sharing data or utilizing external AI systems due to patient privacy concerns. Public sharing of such data could lead to serious privacy violations.

Nillion's blind computation technology enables hospitals to leverage AI for personalized diagnostics while complying with strict regulations like HIPAA and GDPR. In 2024, Nillion partnered with Maya, a psychedelic therapy service provider, enabling them to securely store and process sensitive patient data in encrypted form. This allows Maya to analyze and research data without directly accessing personal information, marking a significant advancement in healthcare service quality.

2.5 Inter-Company Collaboration for Cybersecurity

IT companies often need to analyze code for vulnerabilities without exposing their proprietary code. Nillion's blind computation facilitates this. For example, when Company A has an AI model for detecting security vulnerabilities and Company B wants to analyze their proprietary codebase, blind computation allows both parties to protect their core assets (AI model and codebase) while effectively identifying security vulnerabilities. Detailed information is available on Nillion's blog.

3. Looking Ahead - The Most Critical Piece for Web3 Mass Adoption

I believe blockchain's transparency is a double-edged sword. While it ensures fair access through transparency and traceability, this very transparency deters key institutional players from adopting public blockchains. Not everything benefits from complete transparency. While data privacy could potentially be misused, I believe the main barrier to Web3 and blockchain mass adoption isn't scalability or UI/UX - it's transparency.

Most valuable data in the world is likely private. While this privacy may enhance its value, someone needs to build services using this data for Web3 to compete with Web2 services. With infrastructure enabling computation on encrypted data without decryption in a distributed environment, Web3 could create highly useful applications using sensitive or private data.

While blockchain is significant for enabling self-custody in the digital world, it's only a partial innovation if limited in the data it can handle. If private data computation becomes possible in blockchain's trustless environment, we can truly expect mass adoption of Web3 products.

However, blind computation isn't simple. As Nillion continues introducing various technologies for more complete blind computation, their progress warrants close attention. Who will emerge as the winner in the blind computation market?