|Pix Credit here|
My Administration places the highest urgency on governing the development and use of AI safely and responsibly, and is therefore advancing a coordinated, Federal Government-wide approach to doing so. The rapid speed at which AI capabilities are advancing compels the United States to lead in this moment for the sake of our security, economy, and society. (Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence §1)
|Pix credit here|
The AI Executive Order is quite lengthy and follows below. One of the most striking elements of the Executive Order is in its form. It is not written as a precise set of directions; rather it provides a complex, interlinked but general set of objectives directed to the administrative apparatus of the state. Those objectives, then, are left to the discretionary choices of that apparatus--subject to overarching rules and modalities of coordination and requests for legislative action constituting additional authority for administrative supervision--to flesh out and implement. But not just to flesh out and implement but also to administer. That administration, in turn, will be fueled by compliance measures. In this case, compliance measures requires the governmentalization of targeted private actors (i.e., those in the business of creating and using AI and AI related systems, however these are defined)) onto whom the responsibility for implementation, reporting, and policing, will be delegated in the first instance.
Perhaps the image above best sums up the conundrums that remain to be resolved: one has an image of a President hand signing a piece of paper the text of which, as a physical object, is to be transformed into action by a series of human organizations. That moment of symbolic signature is attended by other humans who preserve the moment for posterity--or attest to its reality--by recording the moment of signature on their mobile phones. It is not the event itself but the abstraction of the event and its imagery that becomes the reality propelling consequential action. Those images will then by uploaded into various platforms the effects of which (singular and cumulative) will be curated by a set of generative programs that are themselves the object of the Executive order and the regulations that will flow from them. And yet, that image is in a sense theater for humans. The text of the measures have been uploaded in word searchable data bases the protection against the corruption of which is automated and their use powered by systems that are also the subject of the Executive Order. Data that will drive the shaping of the regulations, the scope and breadth of delegation to regulatory subjects, and that will shape the exercise of administrative discretion in shaping the expectations of administrative supervision will all be generated by and with the assistance of big data and automated systems tat are themselves the objects of this effort. And it is the biases that will, in the aggregate, shape both the text of regulation the expectations of compliance, and the exercise of administrative supervision, that will both reflect and in its iterative reproduction likely also potentially corrupt these measures (discussed in its theoretical aspects here). The domestication of A.I., like that of dogs millennia ago, will likely be a partially successful project as log as canine sentience and cognition (like that of generative intelligence) cannot be wholly subsumed within that of those seeking mastery. And yet maybe, just maybe, that is for the best. The canine that sometimes bites its human, and the human that can exploit their canine produces the sort of uncertainty that reminds each that neither is the other.
|Pix credit here|