We can no longer afford to regard AI as a free market invention, left to be determined by the natural interplay between corporation and consumer. The trajectory of AI development is rapidly approaching critical intersections with human rights and prosperity, and, in its growing capability, increasingly threatens their dissolution. 

The immense degree of funding poured into autonomous weaponry, policing, surveillance, and already existing media algorithms has wrought novel capacities for phenotype-based profiling, targeting, misinformation, intimidation, and—in short—control. With these inevitably comes the potential for gross power imbalances, abuse, and subjugation to the private entities and nations able to fund the development of such advanced AI technology.

Without clear laws to protect and guarantee our rights within this new reality, there is a significant chance that we will lose the ability to advocate for ourselves altogether. The advent of AGI* and its trajectory toward superhuman intelligence is poised to displace public input from the world narrative entirely; we will lose the ability to predict—let alone control—the design and outcomes of future AI models. Many experts have suggested that a point of singularity is near—the fabric of societies worldwide is expected to change, fundamentally. And the AI arms race that has taken hold between various governments has served only to reinforce, expedite, and incentivize the overriding of safeguards in the AGI development that is anticipated to enact such change. 

The potential dangers of advancing AI have thus come to demand immediate, public, and international concern. 

We, the more than 99.99% of people not consisting of big-tech CEOs and their billionaire investors, can no longer afford to passively regard AI development as a corporate or private endeavor of which we—the people of the world—are at the mercy; we cannot afford for decisions that will affect humanity worldwide to remain solely in the hands of a few individuals or to occur under the secrecy of the world’s wealthiest governments. Continued indifference can and will rapidly become catastrophic. THESE ARE OUR LIVES, and big-tech’s private means of dictating them without our representation, consent, or input undermines every democratic constitution that has ever been put forth for the betterment of human equality and life. 

THIS IS OUR FUTURE—WE ARE ENTITLED TO A SAY IN IT!

We are the majority of the world. And we are entitled to a voice in it. We demand a say in the impacts that AI development and decision-making are inflicting upon our futures, our jobs, our lives, our shared reality. WE, more than anyone, are disposed to suffer the effects of reckless AI progress. An invention with the potential to wield world-altering power requires a tremendous deal of foresight, oversight and responsibility. We demand safety, accountability, transparency, and consent take precedence over material greed, competition, and myopic notions of “progress”.

 

THE SOLUTION

Given the great, international dimensions of this issue, We, the People of the World, call upon the United Nations, entrusted to preserve the sanctity of human rights and spirit worldwide, to create and, in the form of an international treaty, consent to the following AI safety measures, with the overarching goal of ensuring safe and transparent AGI development and the well being of humanity:

I. We propose the development of an international regulatory body*—which, for the purposes of this document, we will henceforth term the “UN AGI Regulatory Body”, or “AGIRB”, for short—consisting of academic experts across multiple disciplines and entirely distinct from corporate, private or national interest, to secure transparent AI and AGI development internationally. Consenting UN member states will be bound by the body’s regulations and oversight, consisting of the following AI and AGI-specific regulatory measures:

1. The UN AGIRB is to create and maintain a public database of all large-scale AI and AGI models and software both existing and in development. The aim of this database is to foster transparency, enable oversight, and enforce accountability in large-scale AI and AGI models.
2. The UN AGIRB is to mandate the registration of all existing AI, AGI and related models and systems otherwise in progress to this database. Registration will include a unique model number and pertinent model information, including (though not limited to) “system cards” that list the data sets on which the model was trained and include substantive risk evaluations, including the model’s projected ability for “rogueness”, self-replication, and resistance to deletion or shutdown.
3. The UN AGIRB is to conduct monthly audits to confirm the registration of all AI and AGI models, projects, and development otherwise in progress, as well as to confirm approval status for each model’s HRSP (described in greater detail in section II. below). The audits may involve a combination of in-person and remote informational review measures, designed and executed to effectively account for all AI and AGI developmental activities undertaken by any given agency, company, organization, or other agent and to ensure both transparency and safety in these developmental activities.
4. The UN AGIRB is to enact punitive measures against AGI developmental agents that fail to register all completed or developing AGI systems. Such measures should non-exclusively include: (1) public declaration of the entity’s failure to register and of the associated risks to public safety. This declaration should occur principally in the form of a “Monthly AGI Compliance Report”, whereby noncompliant entities are announced as potential human-safety threats; (2) the issuance of a deadline for noncompliant entities to register all AGI models and return within compliance; (3) pressure on national governments to withhold funding and enact mandatory fines against entities that fail to register all technologies within 14 days of notification of audit failure; (4) sanctions and other legal action against entities that fail to register all technologies within 14 days of notification of audit failure.

II. We propose that the UN AGIRB require each entity engaged in AGI development to publish a “UN Human Rights Safety Plan”—henceforth termed, “HRSP”, for brevity—for each registered AI and AGI model. The goal of each HRSP must be to demonstrate clear and effective controls for the preservation of human rights and safety. The HRSP will require approval by the UN AGIRB and should be published along with pertinent model information (described above) in the UN AGIRB database. While specific to each AGI, each HRSP must explicitly list and define the following general safety requisites: 

1. The prevention of autonomous updating and/or generation that could result in recursive self improvement and exceed the UN AGIRB’s ability to review, authorize, and oversee the safety of each model prior to its release. Specific controls—such as the inability to actualize system updates without explicit human developer and UN AGIRB review and authorization, and the inability to resist or engage in “self-preserving behaviors” against human-authorized shut down—should be established and clearly explicated in each AGI technology’s HRSP.
2. Parameters that will prevent human-directed overriding, corrupting or “hacking” of advanced AI systems for terroristic purposes—including by developers, investors, and other stakeholders. Specific training and safeguards to prevent such usage must be reviewed and approved by the UN AGIRB.
3. Specific training that will guarantee the upholding and protection of each of the 30 universal human rights declared in the UN’s Universal Declaration of Human Rights (UDHR). Each model’s HRSP must explicate detailed data sets, training, and preventive controls dedicated to upholding each Right and must demonstrate hardlined incapacities for behavioral outputs that transgress against these (e.g., actions or outputs that would inhibit freedom of speech or expression; the undermining of reasonable expectations of privacy; race, gender or ability-based profiling, etc.). While all rights should be accounted for, particular concentration should fall on those established in Articles 7, 12, 19 and 22.
4. Specific training and controls established to prevent both human and AGI-lead misuse of autonomous weaponry, soldiers, policing, and other means of intimidation and force, each of which hold the potential to threaten our Universal Rights and general safety on both national and international scales. At a minimum, each HRSP must sufficiently demonstrate that the associated AGI model is programmed with the inability to autonomously kill individuals. Furthermore, each model should be programmed with the inability to target individuals by discriminatory means (including but not limited to gender, race, ethnicity, religious practices, and political affiliation). Such preventions and controls should generally overlap with section II.3, cited directly above, to list specific programming that will prevent autonomous soldiers, police or weaponry of any kind from infringing upon any of the 30 Universal Human Rights.
5. Specific training and controls to effectively prevent misinformation and biased informational narratives. All entities engaged in the development of AGI must demonstrate sufficient safeguards against the technology’s potential to spread misinformation, enable political propaganda, partake in political bias, or otherwise compromise freedom of press via audio, visual, written or other media. Such safeguards may non-exclusively include (1) auto-declaration of all AI and AGI-generated content’s having been AI-generated (and not human-produced), thereby allowing consumers to be informed of its non-human source and enabling them to interpret the information accordingly; (2) specific training that will definitively prevent AGI’s intent for (a) autonomous deception and (b) obedience of untruthful orders or bias intentions that could lead to false or misleading information; relevant AGI models should be unable to publish information that is non-factual or biased; (3) programmed safeguards that require an individual’s consent before his/her likeness (visual, auditory, or otherwise) is able to be replicated, portrayed, or otherwise reproduced in AI and AGI-generated media; (4) training and programming controls that will prevent the use of AI to eliminate, stifle, censor or reduce the circulation of certain news events in the service of politically, economically or personally motivated agendas.
6. Specific training and controls to prevent both human-led and autonomous financial abuse (including but not limited to theft, hacking, and other, unlawful interference). The controls should include training for already established national and international currencies (e.g. the Euro, Rupee, U.S. Dollar, Yen, etc.), rather than striving to replace these with a universal or digital currency that may be more susceptible to corporate and AI-related control, manipulation, and abuse.

Each HRSP must provide detailed planning that specifically addresses points II.1-6 as outlined above. Furthermore, each HRSP must pre-date the actual implementation of AGI models in business, military, government, public, and other societal settings; release may not ensue until a HRSP specifically addressing points 1-6 as outlined above is submitted to the UN AGIRB database, and explicit review and approval of the HRSP and its projected efficacy is granted and determined by the UN AGIRB. The continued development of AGI models already existing or in progress should be suspended until the UN AGIRB has granted full HRSP approval.

The UN AGIRB must continually monitor the database and deem any entity that fails to register and actualize the safeguards outlined in I. and II. above as noncompliant and subject to the punitive measures outlined in I.4 above.

III. The UN should mandate monthly collaborative meetings between national leaders, leading AI developers and CEOs, and AGIRB constituents. The aim of these meetings will be to foster a spirit of transparent and collaborative progress, with the common goal of furthering the prosperity, safety and well-being of humanity worldwide.

IN CONCLUSION

 

We demand the members of the UN take decisive, large-scale action to ameliorate the current, profit-based and unilaterally beneficial trajectory of AGI development. Clear, unequivocal and universal regulations must be imposed to ensure the hands and minds holding our futures move and think with safety, transparency and accountability—with our rights and prosperity, and not personal gain—at the forefront of this great endeavor. AGI decision making is not something we can afford to entrust exclusively to big tech CEOs, with uncertainty or indolence as our excuse. We DEMAND accountability, and more importantly, we DEMAND security in knowing that the ability to determine the courses of our own lives will remain unimpeded; we DEMAND a framework enabling our futures to remain clear and open—and we DEMAND to see its proof, NOW, before emerging AI renders us unable to demand at all!

Sign in advocacy for your rights—GIVE US A FIGHTING CHANCE FOR A FAIR AND FREE FUTURE!!!

*For the purposes of this petition, AGI will stand for “artificial general intelligence”, defined as any technology—whether in progress or already actualized—that is engineered with human-level or better intelligence in a majority of domains. This definition is intended to include ASI, or “artificial superintelligence”.

Many have argued that the achievement of AGI will  inevitably trigger a process of “recursive self-improvement”, whereby AGI will engineer itself and future models independent of human oversight, rapidly accelerating in intelligence and capabilities and increasingly exceeding human comprehension. This would essentially be the “point of no return”; humans would lose all ability to predict and control the outcomes and trajectory of AI and, more than likely, our future.

*While the UN currently has at least two governing bodies intended to facilitate research, international dialogue, and establish guidelines for safe AI development, the proposed power and scope of these are purely advisory. As a result, they are ill-equipped to effectively monitor the rapid and deeply funded AI and AGI development efforts already underway, and because of this, will likely fail to curtail many of the associated risks that have already begun to materialize.
While we recognize that UN endeavors must be largely dialectical and prescriptive in nature, we believe an exception must be made for the arena of AI development. A body invested with more authority and oversight is needed, as international peace, security and the enjoyment of human rights will not ensue in the wake of emerging AGI unless developers are held accountable by a clear and pervasive regulatory framework.