How Do Blockchain and AI Agents Work Together?

TL;DR

Blockchain can give AI agents accounts, payments, traceability, and access to marketplaces, allowing them to coordinate work as economic participants. NEAR aims to hide Web3 infrastructure behind an operating-system-like experience, while Polosukhin argues that AI alignment also requires addressing persistent human problems such as misinformation and conflicting behavior.

Transcript

a blockchain operating system might just be the key to a democratized web3 in fact more than 25 million users are already getting a taste of this thanks to near this week ilad and I are joined by Ilya palosuken the co-founder of near and co-author of the landmark Transformers paper to discuss the interaction of blockchain and AI Technologies what w... Read More

Key Insights

• Transformers emerged from efforts to replace slow recurrent neural networks in production-scale language tasks. Polosukhin and his collaborators explored whether attention alone could support encoder-decoder systems, learn word order, and perform translation without relying on the sequential processing used by earlier architectures.
• NEAR originally pursued machine-generated code rather than blockchain infrastructure. Its team assembled a community, largely composed of students, who completed crowdsourced tasks such as describing code, writing code from descriptions, and creating tests for training data.
• Cross-border contributor payments pushed NEAR toward blockchain. Many participants were located in China, Eastern Europe, and other places where banking access or monetary controls created practical payment difficulties, making decentralized payment infrastructure relevant to the company’s own operations.
• NEAR describes its platform as a blockchain operating system that abstracts technical complexity. The goal is to let people discover and use Web3 applications without thinking about underlying networks, payments, or implementation details, much as a phone operating system hides low-level infrastructure.
• Blockchain and AI intersect most directly through marketplaces for compute, models, and data. Blockchain can provide traceability and a shared environment for participation, while marketplace structures can connect people and machines that supply or consume specialized resources.
• AI agents become economic participants when equipped with blockchain accounts. Such agents could pay people or other AI systems for work, exchange context directly, and coordinate tasks without requiring a human intermediary to translate machine output into instructions.
• AI-managed organizations could coordinate specialized human work around defined missions and performance indicators. Polosukhin proposes that an AI agent could distribute funding, select suitable contributors or laboratories, monitor delivery, combine results, and replace underperforming participants without personal affiliations.
• Decentralized autonomous organizations are a likely early setting for AI coordination. Bots with access to organizational context could answer recurring questions, onboard members, assign initial tasks, and reduce repetitive administrative work in communities where payments are already integrated.

Questions & Answers

Q: How did Illia Polosukhin contribute to Transformer research?

Illia Polosukhin worked on natural-language understanding and question answering when recurrent neural networks, particularly LSTMs, represented the state of the art. Because those systems were too slow for production as document sizes increased, he and his collaborators explored attention-based encoder-decoder designs. Their experiments tested whether attention could learn word order and perform translation, contributing to the landmark Transformer work.

Q: Why did NEAR change from an AI company into a blockchain project?

NEAR initially sought to teach machines to write code and built a crowdsourcing community around language-to-code data. Contributors wrote code descriptions, produced code from descriptions, and created tests. Paying participants became difficult because many lived in China, Eastern Europe, or other regions affected by limited banking access and monetary controls. The team investigated blockchain to solve that payment problem and discovered a broader infrastructure opportunity.

Q: What does NEAR mean by a blockchain operating system?

A blockchain operating system is an environment that hides low-level Web3 complexity from users and developers. NEAR’s stated goal is to let people discover and use decentralized experiences without needing to understand the networks, payments, or underlying implementations that support them. The comparison is to a phone operating system, which handles infrastructure while developers build applications and consumers simply use them.

Q: How can blockchain support AI marketplaces?

Blockchain can provide infrastructure for marketplaces involving compute, models, and data. Polosukhin says Web3 is effective at creating marketplaces, establishing traceability, and offering an equitable place where participants can contribute. In this model, people or machines could supply resources, receive payments, and track transactions through a shared system rather than depending entirely on a conventional centralized intermediary.

Q: How can an AI agent become an economic agent?

An AI agent can become an economic participant when it receives a blockchain account and the ability to make payments. It could then pay people or other AI systems to complete work, communicate the required context directly, and collect results. Polosukhin considers natural-language communication important because machines can now interact with people through the same communication medium, reducing the need for an intermediary human.

Q: How could AI run or coordinate an organization?

An AI-run organization could give an agent a defined mission, measurable performance indicators, and oversight from a community or board. The agent could divide work, identify suitable contributors, provide context, allocate funding, monitor delivery, and combine outputs. Human overseers could correct unwanted behavior by supplying feedback and training data, while blockchain accounts could support transparent payments and economic activity.

Q: Why might decentralized autonomous organizations adopt AI managers first?

Decentralized autonomous organizations contain repetitive coordination tasks that can already be automated. Running a DAO may involve onboarding members, repeating background information, answering recurring questions, and assigning initial work. A bot with access to the organization’s interaction history could perform these functions. DAOs also have payment systems readily available and fewer traditional workplace constraints around introducing an AI coordinator.

Q: Why does Polosukhin describe AI alignment as a human problem?

Polosukhin argues for thinking about human alignment alongside AI alignment because harmful behavior and misinformation existed before current AI systems. He points to Byzantine fault tolerance, a foundational blockchain problem with historical roots in managing unreliable or misleading information. His framing suggests that aligning machines with human values cannot be separated from disagreements, deceptive behavior, and coordination failures among humans themselves.

Summary & Key Takeaways

• Illia Polosukhin helped develop the Transformer architecture while working on natural-language understanding and question answering. His team explored attention-based encoder-decoder systems because recurrent neural networks were too slow for production as documents scaled, eventually demonstrating that translation and word-order understanding could be handled through attention.

• NEAR began as an effort to teach machines to code, supported by a distributed community that created descriptions, code, and tests. Difficulties paying contributors in countries with banking and monetary restrictions led the founders toward blockchain, where they saw an opportunity to build infrastructure addressing their own coordination problem.

• Polosukhin identifies marketplaces, autonomous agents, and AI-coordinated organizations as major intersections between blockchain and AI. Blockchain can support payments and traceability, while language models can communicate instructions and context directly. Early applications may appear in decentralized autonomous organizations, where repetitive onboarding and coordination work can already be automated.