The Internet of Intelligence

The Internet of Intelligence (IOI for short) is a new form of the internet. The goal of this article is to explore why it will become ubiquitous, how it works and what's the path of getting there.

Unlike the Web, where users seek information and content, or web3, where users can manage digital assets and use self-executing applications, Internet of Intelligence allows users to engage directly with intelligence capabilities. You just ask and it does things for you. An intent is directly converted into the result, and all the magic is hidden from the user.

It functions as an open, decentralized, two-sided marketplace, transforming how technology is helping people to achieve more. Demand side is anyone who has a job to be done. Supply side is an interconnected network of AI models, agents and multi-agentic workflows. Together, this forms a self-improving network of AI capabilities.

Internet of Intelligence is an environment where software and hardware gain agency. Unlike ML models or hardware, IOI can actively accomplish tasks for you. It seamlessly routes between agents and services, employing the most relevant and capable AI systems to deliver results. Indistinguishable from magic, yet already possible technologically.

How will we interact with Internet of Intelligence?

In the IOI, the end-user experience resembles using a search engine. Users input tasks they need help with—from finding a new apartment to diagnosing medical issues, reviewing legal documents, designing interiors, or building applications. Unlike a list of links or generated text, IOI provides the final result immediately. It's akin to hiring someone to complete a job, but the work begins instantly and requires less time and money. This process is more cost-effective, and users maintain full control over the workflow. IOI offers intelligence for hire with all the conveniences of software: infinite scalability, low marginal and transaction costs, and cheap global distribution.

Let's say I'm organizing a hackathon (e.g. dAGIhouse) and I need a website to be built. I submit my task along with a $ deposit and a bunch of AI tools and agents work together to complete it. A routing model finds the optimal coding agent. Coder, tester, deployer agent get it done. Computer vision tools analyze websites of my competitors. UI generative model creates a bunch of mock ups. Product manager agent quickly deploys a few versions and automatically starts $50 test campaign on AdWords to see which one yields better conversion. After analyzing the results, it discards low-performing options. Finally, controller LLM verifies if everything is completed well (asks to re-do if not) and presents me with a completion receipt, log and documentation.

Unlike existing agent frameworks like CrewAI, AutoGPT or TaskWeaver, IOI is truly decentralized system, meaning that each node or agent in the network can be running on a different hardware, utilize different foundational model or model class. These agents are composable and discoverable, but without a centralized registry.

A few more examples. Imagine Sarah, a solopreneur, using the Internet of Intelligence to launch a new product. She inputs "Create a business plan for an eco-friendly toothbrush" and within minutes, an AI agent analyzes market trends, financial projections, and environmental impact data to generate a comprehensive plan, subcontracts other AI tools and humans to create relevant launch materials and exits after presenting work artifacts and a log of completed tasks to Sarah.

Meanwhile, Dr. Rodriguez, a cardiologist, needs to diagnose a complex heart condition. He uploads patient data, and a specialized medical AI collaborates with a network of global experts to provide a detailed diagnosis and treatment plan in hours instead of weeks. Agents can leverage multiple specialized ML models to come up with the most certain diagnosis. Part of the insurance payout goes to model creators or data curators with a transparent provenance of their contribution.

In another instance, a high school student named Alex asks the system to explain quantum computing receiving a personalized, interactive lesson complete with simple analogies and virtual demonstrations. The tutor agent privately leverages Alex’s personal data to make the content more tailored and easy to comrehend. Alex can ask his personalized assistant to explain any topic in details and gets a video or 3D object visualization created on the fly.

The Architecture

Traditional Web consist of two main components: communication protocol (e.g. TCP, HTTP) and services (e.g. websites, applications, smart contracts). In the context of the Internet of Intelligence these are replaced with:

1. AI data transmission and communication systems (such protocol is not invented yet, so people use JSONs with natural language instructions as a temporary measure)

2. AI agents, applications or services which themselves in turn leverage machine learning models, software tools, sensors, IoT and/or embodied robotics.

Internet of Intelligence, just like traditional Web or web3, will be open, permisionless, decentralized, composable. Anyone can join, anyone can use it, anyone can build for it, anyone can use existing services to build on top of those.

The IOI stack roughly consist of the following components:

1. Compute: distributed, universal, permisionless compute for training, inference and running applications.

2. Data: specialized high-quality data for continuous pre-training, fine-tuning and RAG with provenance and transparent renumeration for data owners or creators.

3. Models: variety of ML models with a lucrative system for monetizing your work as a model creator.

4. Evaluations and benchmarks: robust and credibly neutral set of benchmarks for evaluation of model and agent capabilities across hundrends of domains.

5. Autonomous agents: autonomous agents capable of competing tasks, using tools and interact with other entities (humans or AI).

6. Multi-agent workflows, orchestration and routing: for each job, a dynamic routing selects the best possible combination of agents, tools, services, and orchestrates the workflow.

7. AI economy: ability for agents to enter contractual relationships with people, businesses and each other, manage capital and leverage on-demand tooling, such as memory, humans-in-loop marketplaces, APIs or external capabilities in a form of software tools or data.

The more detailed technical architecture and building blocks will be described in the next article Frontiers of Decentralized AI.

Road to the Internet of Intelligence

The Internet of Intelligence faces significant challenges, particularly in privacy, job displacement, and centralization risks. Privacy concerns are paramount, as the system requires extensive personal data. However, emerging technologies like user-owned, self-sovereign data models, trusted execution environments, and advanced cryptography offer promising solutions, potentially allowing users to maintain control over their information while still benefiting from AI capabilities.

Job displacement is another critical issue, as AI agents could replace human workers across various sectors. Yet, this shift may also free up human capital for new, unforeseen industries and creative pursuits, potentially leading to a societal renaissance in innovation and artistry.

Perhaps the most ominous threat is the potential for a dystopian future controlled by a few AI owners, wielding unprecedented power over global economy (as larger portion of it depends on the IOI). To counter this, the implementation of decentralized markets, carefully designed incentive structures, and open-source nature of software could distribute power more equitably.

Unlike current AI assistants which operate within closed ecosystems, the Internet of Intelligence aims to create an open, blockchain-like network where various AI models and human experts can interact seamlessly. This requires developing new protocols for AI-to-AI payment and communication, establishing universal standards for data and value exchange, and creating robust systems for verifying the credibility and performance of AI agents. The decentralized aspect could be achieved through a combination of cryptography, p2p networks, mechanism design, federated ML, distributed fault-tolerant systems, and edge computing.

Impact

The agent economy will be distributed for the same reason the traditional web ended up that way. No single company can build billions of AI agents, and the value lies not in the AI apps themselves but in the network. Metcalfe’s Law applies to AI as it does to any other digital system.

The Internet of Intelligence, like the human brain, derives its capabilities from the connectivity and collaboration of its various components. Interoperability and trustless composability will unlock even more value than LLMs or any other new ML architecture alone. "Society of agents" approach in coding AI agents show almost 2x improvement without any changes to the underlying models.

The economic impact of the Internet of Intelligence could be far-reaching. Some estimates suggest that AI technologies could add $15-$20 trillion [PwC, World Bank] to the global economy by 2030. The IOI, as a more advanced and integrated system, could potentially double or triple this impact. For instance, by automating complex tasks across industries, IOI could increase global productivity by 20-30% in knowledge-intensive sectors like finance, healthcare, and technology.

Furthermore, the IOI ecosystem could spawn entirely new economic models. The decentralized marketplace for AI services could create a "gig economy" for AI agents, potentially generating millions of new jobs in AI development, maintenance, and oversight. This could offset some of the job displacement caused by automation. Additionally, the value created by IOI could be more equitably distributed through tokenization and decentralized ownership models. For example, if 10% of global economic activity shifts to the IOI by 2040, and if this value is distributed among participants based on their contributions, we could see a more dynamic and inclusive cybernetic economic system.

Conclusion

The Internet of Intelligence is not just a technological leap; it's an economic imperative. As AI capabilities exponentially increase, the shift towards an AI-driven workforce is inevitable due to radical economic efficiency of artificial intelligence vs wetware. The question is not if, but when and how this transition will occur. The potential to revolutionize every sector of the global economy, from healthcare to finance, from education to environmental management, is too significant to ignore.

However, the path to this future is not predetermined. It's up to us to shape an IOI that is open, ethical, and beneficial to all of humanity.

This is where you come in. cyber•Fund is actively seeking visionaries and innovators who are building the foundational elements of the IOI stack. Whether you're developing new AI models, creating decentralized marketplaces, or designing privacy-preserving protocols, we want to support your work. Through venture funding, grants, and research projects, cyber.fund is committed to accelerating the responsible development of the Internet of Intelligence.

Special thanks to Daniel Luca, Sergey Anosov, Artem Kotelskyi for his insightful comments and feedback.