Can a global version of the 1786 Annapolis Convention lead to the governance we need for AI?
Can the intergovernmental constituent processes that led to the U.S. Constitutional Convention of 1787 be replicated globally to create the federal global governance we need to manage the immense risks and opportunities of AI, and create a model for other dangerous technologies and global challenges?
by Rufo Guerreschi and Roberto Savio
(For comments, likes and reshares, refer to this Linkedin post)
On their own, nearly all states stand powerless in the face of AI, unable to avoid its immense risks for safety, for concentration of power and wealth, and unable to realize its astounding opportunities. Even large states, like Brazil, India and Germany.
On their own or together, AI superpowers stand unable to ensure the non-proliferation and safety of AI, as its innovation and accessibility increase at a shocking pace, much faster than nuclear did at its outset in 1945.
As the risks of nuclear power and the promise of atomic energy became evident to all after Hiroshima, the US and the Soviet Union formally proposed in 1946 a truly multilateral intergovernmental agency to control all research and arsenals via their Baruch and Gromyko plans - not subject to the veto of the five winners of World War II.
They failed to agree and chose instead to manage the risks via informal coordination of security agencies and pressures on states, while keeping nuclear energy science mostly to themselves.
Only after all five of them achieved solid nuclear weapons capability - and nuclear energy was revealed to be less disruptive than planned - that the IAEA was established in 1957 and weak treaties were created, to discourage others from doing the same.
Not much has changed since 1946 in the global governance of dangerous technologies.
Almost eighty years later, the architecture of world power is much the same. Our de-facto global governance is essentially one of a global anarchical oligarchy.
It is partly a global oligarchy of the five veto-holding WW2 winners, with their power to block any binding resolution or changes to the UN charter, as enshrined in the charter itself. It is partly a global anarchy, whereby those with the most guns and money rule over others, with very weak international law and institutions.
This state of affairs has worked, more or less, for most of humanity with relative material abundance and peace. It only partly worked in managing hugely disruptive technologies, like nuclear, bioweapons and encryption, with minor accidents, but their risks are higher today than ever.
Yet, as we race to an AI Age, today's global governance is utterly inadequate. It will likely lead to enormous catastrophes or concentrations of power because AI innovation and investment and its proliferation are likely to be much harder to control than nuclear, and because of their magnifying effect on those other risks.
As in 1946, we are again at a historical crossroads. The immense risks posed by disruptive technology, this time AI, create a historic opportunity to leverage those risks to create truly multilateral global organizations that can reliably steer it toward the global public good - and even establish a governance model applicable to other dangerous technologies and global challenges, like climate change, peace, and social equity.
But how can small and medium states foster the creation of such organizations?
In 1946, the debate was mainly between Russia and the US, which eventually settled on a solution that served them well economically and geopolitically, entrenching their dual hegemony over the world.
This time around, while states, on their own, lack the strategic autonomy and geopolitical weight to advance alternative constituent processes to build such a multilateral organization, they can engage in open constituent processes to progressively aggregate more and more states around democratic and participatory processes to create such organizations.
It has been done before with great success.
Nine years after the U.S. Articles of Confederation were enacted in 1781, many states realized it was far from enough to safeguard their economy and security.
Five of them met in the Annapolis Convention in 1786. They decided to design and convene a two-month U.S. Constitutional Convention in 1787 to build a true federation of states, each with its form of governance and idea of human rights.
There, state delegations agreed by simple majority on a federal U.S. Constitution that was bound to come into force when and if 9 out of 13 state legislatures ratified it.
In hindsight, it was an astounding success overall. Over time, it turned out to be an economic powerhouse, and vote and civil rights were extended from only 1 out of 8 adults to all, regardless of gender and race, setting an example for the rest of the world.
What can we learn from the Annapolis Convention of 1786 for a future beneficial governance of AI?
A similar federal intergovernmental constituent process, for the same reasons, can and should be replicated at the global level for AI - a history-defining technology with immense implications for the economy and security - providing a model for a later expansion to other domains.
A small number of globally diverse states, even as small as 7, could gather to design and promote such a constituent process, gradually attract dozens of other states to reach a critical mass like that of the Annapolis Convention, and then convene for an Open Transnational Constituent Assembly for AI and Digital Communications.
The convening of such an assembly should include a clear mandate and rules for its election, conduct and approval, which should prescribe that the resulting charter will be binding if ratified by 65% of the participant states, as for the US Constitutional Convention. At any time in the process, the two AI superpowers of the US and China would be welcome to join on equal terms, or even with some moderately higher decision-making power, for geopolitical realism and for the kind of assets and expertise they contribute.
As foreseen by the Baruch Plan for nuclear, the mandate of such an assembly will include the creation of a public-private consortium to create the most capable, safe AI - and shared digital communications infrastructure. Even if it fails to attract superpowers to join, it'll still increase the leverage of participant states vis-a-vis superpowers in trade relations and safety regulations and remain an open alternative to all major states together, eventually.
We hope our Harnessing AI Risk Initiative and its 1st Harnessing AI Risk Summit, to be held in Geneva this June 12-13th, 2024 - or its 2nd edition - could have the same role for global governance of AI as the Annapolis Convention had for the federal democratic government of the United States - and create a model that can be extended to other dangerous technologies and global challenges.
For good reasons, many are concerned about a strong global governance of AI.
Not all worked well with the U.S. federal constitution. While it successfully promoted democracy and civil rights in nations, the US has actively countered the application of those same principles and democratic federalism at the global level. So far.
In recent decades, its political economics of media, powerful new media technologies and immense concentration of wealth have enabled a few to exert enormous influence over its campaign and media systems to further narrow interests, resulting in distortion and control of informed public consent and civic debate.
Yet, the science of constituent assemblies and constitutions has advanced dramatically since 1787. Many safeguards have proven to substantially mitigate risks of degeneration of a global federal governance organization and the constituent processes leading up to them. These include periodic or supermajority-based constitutional revisions to update to changes in technology and other factors; better models of political economics of media; citizen's assemblies; use of digital technologies to foster transparency, accountability, fair and effective deliberative discourse, and translation among languages.
Since over three billion humans are unconnected or illiterate, a new global organization should commit to ensuring universal literacy, primary education, and connectivity for all in a few years. Since nearly all wealth and political power has accrued to older people, policymakers and academics proposed doubling the voting weight of younger adults, who comprise a wide majority of adults.
It is hard, but we've done it before. The risks are significant, but the risks of not trying are much greater.
And there may be some unforeseen allies as Sam Altman suggested last March the US Constitutional Convention as a “platonic ideal” of what a constituent process to build a global governance of should look like.
Join or support us in this historical venture!