As a human beings and therefore part of humanity, having recognized the dangers which Artificial Intelligence (AI) poses to our lives, privacy, livelihood, health, security, and happiness, we propose the following Treaty of AI.
Article I: The Right to Digital Sovereignty (Privacy & Exploitation)
Issue: We must move from the view of data is oil to be mined to data is an extension of self.
Clause: We establish that an individual’s digital footprint, biometric data, and behavioral patterns are an extension of their physical person and as such are their personal property.
Proposed Solution: We mandate an Opt-Out by Default policy for all biological training sets. We propose the creation of Personal Data Vaults: encrypted, user-owned spaces where AI companies must request permission and potentially pay a fee in order to access specific data fragments belonging to individuals.
Article II: Creative Remuneration (Intellectual Property)
Issue: We must end this era of uncompensated extraction, which is threatening the entire ecosystem of the open web.
Clause: Any model trained on copyrighted material or high-skilled creative resources must either expunge stolen data, contribute to a Global Creative Dividend, or make its model weights freely available to the public.
Proposed Solution: A micro-royalty system, using technologies like blockchain or transparent ledgering to track when a piece of training data contributes to a model’s weight, ensuring that when a model generates value, a fraction of that value flows back to the original creators. We recognize this may not be agreeable to all parties, but feel it would be a step in the right direction.
Future Protections: AI corporations and trained AI models must respect the intellectual and artistic property of artists, authors and other creators. Recently, internal Microsoft documentation described their massive scraping of online data in order to train large language models (LLMs) as the “largest theft of labor in human history,” noting that it made a “complete mockery of the idea of fair use.”
In the future, open web sources should either be protected by default, or have the right to shield their data from such scrapes with a general robots.txt rule, as opposed to a specific entry for each crawler such as the one below:
# OpenAI (ChatGPT, GPT-4, Sora)
User-agent: GPTBot
User-agent: ChatGPT-User
Disallow: /
Currently the standard robots.txt protocol does not have a single, universal user-agent or wildcard tag that targets only AI crawlers while allowing search engines. Because new agents will be introduced over time, this puts an unnecessarily heavy burden on the web sites to opt out of. A new universal AI user-agent rule needs to be created to address this.
Article III: Ecological Equilibrium (Sustainability)
Issue: We must decouple Artificial Intelligence from its environmental destruction.
Clause: We must prohibit AI’s infinite scaling model which relies on energy-intensive, centralized data centers without a 1:1 carbon-offset or renewable-energy mandate. We must also focus on performing AI inference operations on the edge (on personal devices).
Proposed Solutions: We suggest subsidizing the development of Small Language Models (SLMs) and high-efficiency hardware. We also propose setting a carbon cap on model training, where the size of the training run is limited by the available renewable energy capacity of the provider. We promote AI inference be primarily performed on the edge.
Just as the PC moved computing from mainframes to desktops, AI must move from the Cloud (data centers) to the Edge (personal phones, laptops, local devices). The goal is a local-first AI, which reduces the need for massive data centers, which minimizes environmental issues, and which also ensures privacy, since user data never leaves a user’s device.
Article IV: The Doctrine of Human Augmentation (Purpose)
Issue: AI’s north star or highest purpose should be to improve, not replace, humanity.
Clause: We should commit to a Human-in-the-loop (HITL) design in for all AI operations.
Proposed Solutions: We propose tax incentives be created for companies whose AI tools are demonstrably proven to increase human productivity and/or well-being, rather than those that simply automate roles out of existence.
Additionally, we propose that AI should be prohibited from performing high-stakes autonomous decisions (legal, medical, or lethal) without the supervision of a highly trained human arbiter. Humans must be held accountable whenever any AI decisions or misinterpreted data enters a patient’s records and adversely affects their health.
Article V: Decentralization of Truth
Issue: Industrialized AI has positioned itself as a singular source of truth. The issue with this is that often there are factual issues with production material.
Clause: We have witnessed the ultimate outcome of corporations who have positioned themselves as a singular source of truth. Eventually they bypass the truth and bypass human diversity.
Proposed Solutions: If AI is to be integrated into commercial search engines, a new, decentralized marketplace of ideas could be given a voice through a distributed system of search engines which present multiple points of view. The people could take back the Internet by creating a mesh system of intelligence systems.
Alternatively, search engines could revert back to their originally intended design which was surfacing the best source of publicly available material on specific topics and presenting curated lists of web links. If AI summarization/correlation is requested, the user could either perform this on their own devices with multiple links as source material, or trust that the search engine’s output is accurate.
Article VI: Algorithmic Transparency & Auditing
Issue: A lack of transparency in regard to use of AI generated resources undermines public trust.
Clause: Standardized transparency mechanisms and attribution systems should be developed which inform users as to which resources have been AI generated. Additionally corporations who replace employees with AI systems completely (as opposed to augmenting these positions) should be required to publicly report this so that the general public can decide whether they want to continue to do business with said entities.
Proposed Solutions: Standardized human and machine readable markdown templates could be developed (similar to robots.txt) which would allow both users and agents to easily audit and aggregate web site information in service to transparency.
Additionally, if prompts are used to create artistic works, these prompts could be shared in order to benefit other users or to provide attribution for an artist’s works/style.
Final Words
The current situation with AI is evolving at a mind-bending pace with little to no oversight. Without public pushback, it has already caused severe harm, and will continue to do so. This is untenable . We simply must begin work to guide AI back into service of humanity. This treaty is only a start.