AI Governance DAO

From DAO Governance Wiki
Jump to navigation Jump to search

The aiGovDAO is a DAO devoted to democratic regulation of the development of artificial intelligence (AI). It uses the DAO Governance Framework to coordinate decentralized collaboration toward the development of artificial intelligence on a global scale. The goal of aiGovDAO is to use AI to promote human flourishing, and to develop wise practices in the deployment of this technology. aiGovDAO has an open membership policy to any person from any background who can contribute meaningfully to the goals of the group in accordance to our shared values.

aiGovDAO practically achieves effective regulation by participating in the non-profit and for-profit development of AI tools. The aiGovDAO is capable of instituting incentives which promote benevolent outcomes because it uses the decentralized governance procedures of DGF. So the distribution of power over the governance decisions of how to regulate the development of AI is not solely concentrated amongst those who develop tools which immediately gain the DAO monetary profit. In the aiGovDAO, power is consciously distributed to those who serve the higher humanitarian goals of the project.

Background[edit | edit source]

Every tool is used to increase or diminish humanity. When hominids first used sticks to extend their reach, they discovered a new opportunity to defend their band from predators and increase their power to acquire food. The first spear was also used to oppress the band of hominids violently. Every tool can be used to support, defend, and promote human flourishing. And every tool can be used to undermine, attack, and oppress humanity.

AI is a new tool that is disrupting human society on a scale previously unimagined. Machine learning has transformed the way people interact with their world. The Google search engine has given people powers of understanding their world that would embarrass the gods of Greece. Humans are now faster than Hermes was ever described. Google Maps has made average humans more knowledgeable of local geography and economics than Athena.

Machine learning and neural networks are different than other tools in their global effect. The sweeping interconnection of web communication aggregates global information more effectively than any national intelligence agency has ever before aspired. These tools effectively exploit this information to address any problem posed in ways that surprise us with their effectiveness and creativity. Contemporary AI tools such as ChatGPT and DALL-E 2 have put this power in the hands of anyone with a cell phone. The use of new and future AI innovations anticipates tools which can solve complex problems with minimal human prompting, mimicking the most sophisticated behavior of humans. The promise and threat that these tools pose to humanity cannot be overstated.

The current state of academic understanding of AI is disappointing. Previous technologies, such as electricity and nuclear power, had been anticipated by science in a manner that allowed greater time for deliberation about its deployment. Nevertheless the application of those technologies have led to negative outcomes. The globally networked application of AI technology has the potential to be even more powerful and dangerous to society than any previous technology, and so demands careful deliberation on the wise deployment of the tools that are arising.

Like any tool, AI cannot be simply suppressed. We don't have the governmental structures to achieve that in our current global societies. Perhaps that is for the best, since as it allows us the freedom to develop openly. However, for such powerful tools, it is essential that development happens in a wisely regulated manner. Wise regulation is a matter of carefully monitoring its development and guiding it in directions which serve our values.

aiGovDAO will seek to govern the development of AI tools broadly, and will have specific protocols for each type of AI research. The most effective way to guide the development of a technology is to be the primary developer. To attract the best developers, the aiGovDAO should pay the the best wages available. To do that, the aiGovDAO should lead the market for AI tools. The plan for controlling the market is to provide free services and tools that are beneficial to humanity broadly (Phase 1, described below), and to charge fees for specialized applications which align with the DAO's values (Phase 2). In the inevitable cases where AI tools emerge which conflict with our values, we will develop freely available tools which police the use of those malign tools (see Phase 1).

Decentralization[edit | edit source]

Many groups are aware of the threat to humanity of technological disruption. And every major center for development of AI has instituted some process for analyzing the ethics of their deployment of the technology, to a greater or lesser degree. However, almost all of these major initiatives are centralized, since the major power behind AI development is for-profit exploitation of the technology.

Therefore, unfortunately, at the moment, the ultimate values and goals that are guiding the development of AI are largely limited to monetary profit. Because of the centralized corporate deployment of the leading AI tools, a primary application of AI today is to improve the effectiveness of advertising. This is having negative effects on society as social media is being engineered to increase engagement at the expense of social harmony.

Further, we cannot establish wise regulation for such a complex problem as AI governance without improving our governance mechanisms, without improving democracy and human communication.

aiGovDAO uses the structure of DAOs to foster collaboration to develop AI tools in the service of the common good. Decentralized organization is used to empower as many people as possible who are conscious of the problem and wise with respect the the consequences of governing the progress of AI. We use the architecture of a DAO to filter information at the edge, to promote the best ideas from the globe to guide the development of AI.

A decentralized AI stack includes research and tooling conversations around data storage, compute, hardware, AI models, and governance. Therefore, in alignment with the goal of decentralizing the governance of AI, we center our focus on the use of small processors[1], P2P networks, open source principles, the use of transparent training data, and democratic decision making.

Value statement[edit | edit source]

We seek to develop AI to be aligned with human values.

We seek to develop wisdom in the deployment of AI technology, for the purpose of increasing individual human agency, balanced by the promotion of human social harmony. Any tool can serve human agency or become a social necessity which humans rely on to the degree that humans live to serve the tool. We wish to develop AI tech in a manner which centers human values over social reliance on technology.

At this moment in history we believe the current state of AI tools and research demands greater transparency and decentralized ownership. Therefore, we are creating an open decentralized organization with the developmental goal of making control of AI as decentralized as possible, so that the excesses of its power are always countered by balancing forces. Yet this decentralized control should be in universal harmony with a will toward the good.

Comments[edit | edit source]

Individual agency is sometimes referred to as hyperagency, which means that the individual participating in the system (or game) is capable of making choices with respect to values that are not obvious from within the confines of the system.


Several polarizing questions arise in the process of governance.

Polarizing questions[edit | edit source]

1. Can anonymous AGI agents participate in aiGovDAO?
Our initial answer is, "No." The purpose of aiGovDAO is to promote human flourishing, as a group and individually. Ceding our authority to machines is counter to that goal.
On the other hand, a natural function of aiGovDAO is to develop the use of AI, especially in the creation of defenses against malign uses of AI. AI is an important tool for helping each individual and community filter information. Both to promote human flourishing in general, and specifically in order to make aiGovDAO more effective, we encourage the use of AI.


2. Does aiGovDAO promote universal control of the development and deployment of AI? If so, is that centralization of power?
We wish to promote the healthy development of individual and collective human power. The healthy development of power must be tempered with wisdom. The power of the individual must be balanced with the wise application of that power in the service of communal harmony. The individual member's agency is always in tension with the group's harmony. Each side supports and erodes the other side. So this question is impossible to answer. We wish for the group to be healthy and powerful, but we also require there to be openness to leaving the group and creating alternative organizations. It is natural and inevitable that people will come to different answers for how to best develop a new technology. So it is natural, and perhaps healthy, that there be different regulatory organizations developing this technology.
In aiGovDAO, we seek to create a community which wisely guides the development of AI. We wish to create a powerful organization that controls that development inasmuch as its goal is to prevent negative outcomes. The power over the decisions of what is good and bad can lead to oppression when control is too strident, and can lead to chaos and dissolution when there is too little control. While we wish to give individuals greater freedom and power, we also wish to support groups which limit the damage of the consequences of greater individual freedom and power. The solution to these problems of too much or too little control, is to nurture the development of healthy applications of AI which promote human flourishing. Healthy applications also include tools which police malign uses of AI.

Business mechanism[edit | edit source]

From an abstract level the business development of the aiGovDAO involves the following basic 2-phase process:

Phase 1. Network of developers make free AI tools.

   cREP-weighted references develop the DAO's Forum with posts on 
      i.   Governance proposals & protocols 
      ii.  Culture and values
      iii. academic research articles
      iv.  AI development Projects


Phase 2. Devs do work for hire, modifying the existing free software for customer fees.

   a) $  --> wREP  
   b) Citations --> DAO power & values


This process collects and curates a network of like-minded developers who make useful and good AI tools which follow the values of the group. In service to the group's value for decentralization, AI tools are developed with a focus on 1. open source principles and transparent training data, and 2. small processor & P2P network focus. Once the objective value of the network's tools and talent is demonstrated after Stage 1, market demand from customers for the group's personalized work will lead to $ fees. Once $ fees are entering the DAO for work, the reputation mechanism functions according to DGF workflow, which makes the DAO's power structure explicit and expresses their values concretely.

Phase 1[edit | edit source]

In more detail, during Phase 1 the aiGovDAO is an academic society, which develops its governance protocols, as well as developing the field of AI research in general. Citations between researchers and administrators for approved published articles, proposals, and projects will provide the basis of future power (i.e., reputation) in the DAO.

During Phase 1, the aiGovDAO develops free tools as a collection of approved Projects. A Project is a collection of Proposals for how to develop AI tools (including how to limit their development). A Proposal can be added to the Forum by anyone. For instance, a Project may be an individual's recommended tool, or it may be a crowdsourced dev project (similar to the Polymath Project) consisting of multiple Proposals linked by citing each other, or it may be a statement intended to clarify the values of the group.

Before the Projects and Proposals lead to citations from profitable uses in Phase 2, a generalized PageRank algorithm weights the value of such Proposals by their WDAG context in the Forum. This citation system allows the DAO to select some proposals to promote for member attention over others. The weights of value are specified by chat reputation tokens, called cREP.

Phase 2[edit | edit source]

At Phase 2, the aiGovDAO has become more mature, and has implemented protocols determining rewards and punishments for member activity. Several approved Work Smart Contracts (WSCs) are available for customers and workers, created during Phase 1 with the protocols for validating the WSCs. Protocols for proposing legislative changes to the DAO are also in use, including how to recommend improvements to WSCs or validation scripts, or review procedures. At this stage, when $ fees from customers enter the system, work reputation tokens (wREP) are minted and distributed to participants. The WSCs reference the proposals that helped the members complete their for-profit work, so the previously free tools cited reward their authors with wREP. The customers' fees are shared will all members proportional to their wREP holdings via the REP salary. This function of power and monetary reward assignment is basic DGF workflow.

Governance[edit | edit source]

Main page: aiGovDAO governance process

The aiGovDAO follows the general DGF governance process. We use the REP-weighted Forum to aggregate and filter ideas communicated through member Proposals. wREP tokens then determine the power to validate Proposals for regulating future AI tech development, since each Proposal used by members is validated using wREP voting.

This page gives the details of some sample protocols.

AI itself can be used in the process of governance, as proposals may be generated by NNs, but selected and promoted by human discernment.

Domains of research[edit | edit source]

The general fields of AI research fall into three categories with many sub-categories.

  1. Machine Perception
    • Image classification; robot vision
    • speech recognition; robot hearing
    • knowledge representation
  2. Information Processing
    • Learning
    • Planning and scheduling
    • Natural Language Processing
    • Social Intelligence
  3. Machine Behavior
    • Robotics


Further, every field of information technology has its own specific AI research issues, e.g., hardware for information processing or transmission.

Code[edit | edit source]

See Also[edit | edit source]

Notes & References[edit | edit source]

  1. Large NNs dominate public attention. They currently are superior to small NNs, yet are inaccessible to the average researcher due to limitations on shared compute time on the few large networks. This perspective argues against the success of a decentralized AI governance group. But it is not insurmountable. Small NNs are surprisingly powerful, even today with the limited attention they have been given. Small NNs can be competitive with--even superior to--large models, since they allow more experimentation from a greater number of researchers. See, e.g., this discussion. And the processing power available to P2P networks exceeds that held in centrally owned corporations.