The future of AI envisions a distributed intelligence ecosystem empowering individuals. It combines personal, domain-specific, and connector AIs to form a super-intelligent system. This paradigm shift focuses on data sovereignty, decentralized control, and human-AI symbiosis. It aims to revolutionize information access, productivity, social interactions, and financial systems while preserving human agency and fostering societal growth.

‍

Impending Social Changes Empowered by AI

Internet products and tools can be broadly categorized into four types:

1. Information acquisition: Such as short-term timely news and long-term valuable knowledge
2. Social connections: Places where humans connect and build relationships in the digital world
3. Productivity enhancement tools
4. Financial tools: Aimed at reducing transaction costs

This is the model that has gradually stabilized since the internet’s development from the 1980s. We respect and continue to learn from this 50-year practical wisdom, continuing to provide digital services for these four categories in the AI era.

However, with AI empowerment, the differences are:

1. Information entry point: The emergence of AI tree structures, and ultimately a super AI becoming the sole entry point for all internet information.
2. Productivity tools: AI can statistically judge the probability of public asset use at the terminal. For example, statistical tools for public assets reduce judgment costs; learning and customizing personal assets, and customizing personal intentions. Therefore, AI will further realize work efficiency improvement, and people’s future work methods and habits will undergo major changes as a result (dependence and utilization coexist, efficiency improvement is endless, accepting historical development trends).
3. Social capabilities: AI will become a new entry point for establishing social relationships. Although it cannot replace real interactions, it can more accurately and quickly promote connections and matches between people within its ecosystem.
4. Industrialization of finance: Transactions may be automatically executed by AI through clue identification and intention setting (passive trade mode), similar to smart contracts.

‍

A Super-Intelligent System Composed of Three AI Types

I predict that three types of AI will emerge in the future: Personal AI, Domain AI, and Connector AI that calls upon these AIs. The entire mixed AI system will collectively form one or a few super-intelligences.

1. Personal AI: Digital personality profiles based on prompt design, such as PI.
2. Domain AI agents: Execution of intentions. For example, accumulated industry customer data can form AI prompts for domain experts. This could create an organizational automation workbench, transferring all SOP processes to AI agent roles, forming a metaverse-like service platform where humans are the demanders and AI provides services. Examples include Midjourney.
3. Connector AI: Similar to calling SDKs, this is a communication method between AIs.

However, I believe the primary focus should be on designing a personal AI ecosystem. True decentralization of power and freedom can only be achieved when the four major domains—information, productivity, social interaction, and finance—are managed by personal AI rather than centralized AI. Unlike traditional consumer or business-oriented products, the future of AI development will prioritize individual autonomy, leading to the creation of a ToAI (To Artificial Intelligence) decision-making and oversight system, essentially a personal AI system.

‍

Construction Goals of Personal AI Systems

AI capabilities stem from data quality, with data being the most valuable asset, and data sovereignty belonging to individuals. Data sovereignty includes rights such as data read-write access, transactions, and destruction.

1. Decentralized human decision-making system: Since super AI is bound to appear, what we need to strive for is to ensure that AI algorithms are not controlled by any centralized rules (including AI), but rather to delegate decision-making power to each terminal human individual. Individuals not only need to have the right to customize algorithms, but large models also need to be deployed to lightweight terminals to support terminal individuals’ participation in key system decisions. This should be based on a clear AI supervision system for semi-automated decision-making and execution, thus entering an era of AI and human symbiotic evolution.
2. Maintenance of public data assets: Values that humans can share for a long time can be called “digital civilization,” forming AI relationship networks of different “breadth, depth, intensity, and height.” This leads to the emergence of a true data-based digital civilization.

For example, the feedback results of large-scale AI search and interaction engines are determined by the decisions of terminal individuals (likes and dislikes). Humans have the final judgment on the results, so its algorithm is not controlled by centralized rules but grows automatically. Whether it’s good or not is voted on by people at the terminal, with everyone having one vote. We need to determine what kind of interaction rules can be used to obtain and use private AI data while protecting its uniqueness, and this process of building trust and authorization can be completed by another AI. What needs to be ensured is that humans need to be present for critical judgments.

‍

Public Data and Data Sovereignty

‍

Data Sovereignty

Large Language Models (LLMs) are compelling and versatile general-purpose models. With the general foundation in place, the application of domain-specific data is an inevitable trend. Any knowledge system that can be structured and organized can generate domain-specific AI, serving as a tool with clear objectives. A key future development direction for AI will be how to acquire, analyze, use, and protect data. Given its significance, AI could even be directly defined as the processing, analysis, and utilization of data.

Public Data

Public collective data can be made open to everyone through blockchain, making it tamper-proof. Private domain data can support conditional data acquisition and data sovereignty protection through hash algorithm structures. The judgment dimensions of these public data (relationship nodes) and whether they are made public depend on individual will:

• Height (level of achievement)
• Depth: Trustworthiness (degree of fulfillment)
• Breadth (connection ability)
• Intensity (influence on the network)

Public assets are information, works, and tools that everyone can access, and each call has incentive measures to allow it to increase daily. Sharing behavior needs to be rewarded. If public assets cannot reward original creators, creators will withdraw assets, causing public assets to dry up.

Possible solutions:

1. Develop universal standards and enforce protection through strong authority
2. Authorized free competition: Pay to obtain source files (this model will reduce producer quality). Only services can be monitored and traded, while works themselves can be copied.
3. Create an evaluation system for the works themselves: Trustworthy domain authoritative organizations. These rules and evaluation systems must be changeable, accepting feedback from terminals to change. They will establish a consensus value orientation and support comparison and competition between members and works on the basis of this consensus. People will flexibly join and choose different value orientations and even influence their formulation process. This way, works can achieve a situation where they are neither centralized nor undervalued.
4. Ensure that decentralized individuals have initiative in the process of participating in the network.

‍

Conclusion

Tools are extensions of human intentions. Good tools determine what people can do within their potential range and do better. Therefore, we must never reverse this, making people serve the tool platform, serving to maintain the tool itself. We look forward to creating a self-supervised AI system that becomes a tool to help human individuals achieve self-realization. Perhaps in the future, people can choose from ten million types of jobs the work they like and find suitable. Empowered by benevolent tools, people will be given the power to freely switch their work methods and life states when they want to.

‍