Open Call for Online Residency: Deadline February 4th -- First meeting Feb 18th
Organizers: Sepideh Majidi and Maure Coise
Deep Objekt [0]: Agency at the Computational Turn | Sepideh Majidi and Reza Negarestani (Framework For AI Research Group Open Call)
In exploring AI’s impact on social, political, and economic domains, we find that responses to current and future AI tend to fall within established orthodoxies or forms of doxastic conservatism not about AI per se but rather about the nature of agency, specifically, what constitutes rational agency. The enigma of rational agency in its relation to itself and the world, as a tightly interwoven lattice of veridical statements, decisions, and actions, appears to be an unexplored or unresolved issue. To a great extent, extreme positions surrounding the phenomenon of AI stem from unresolved problems or misapprehensions concerning the nature of agency, including its inherent attributes, abilities, and revisable values, goals, and attitudes.
The central premise of this open call posits that understanding AI and its discourse necessitates a thorough reevaluation of agency. Traditional approaches to the question of agency are insufficient for reshaping the landscape of AI; we must instead rearticulate and delve into the deep structures in which it is embedded. This involves providing perspectives on the question of agency from both below and above, acknowledging agency's connections with itself, other agents, and the world. Philosophically, these perspectives, delineated as views from below (bottom-up) and from above (top-down), traditionally stem from the confrontations between empiricism and rationalism. Reconciliation between the two, crucial for a comprehensive understanding of agency, must be an integration without fusion, preserving the bounds of empiricist and rationalist views. The evolution of AI underscores the need to negotiate established boundaries, prompting the genuine question of revising these boundaries without reliance on preconceptions or predefined methods.
The proposed reconciliation between empiricist and rational views, exemplified by philosophers like Gaston Bachelard and more recently by Nathan Brown under the term Rationalist Empiricism goes beyond policing each other's claims and instead presents them as reinforcing wings of a shared set of ambitions. Despite their commendable insights, these propositions alone are inadequate for addressing the elaborate issues of agency revealed in current research fields like artificial intelligence and computational psychology. What is required is a research field akin to what philosopher Peter Wolfendale has dubbed Computational Kantianism, or transcendental computationalism–a faithful merger between the concerns of today’s theories of computation and logic and the problems raised by transcendental psychology understood as an investigation into the conditions of possibilities for that which thinks. This approach directs perennial philosophical concerns about the concept of agency to be mindful of ‘the computational turn that began at the beginning of the 20th century, and whose consequences we are still working out; consequences which land blow after blow on our intuitive conception of what thinking is, breaking our ways of rationalizing what we are, and shattering our illusions regarding what it’s reasonable to believe.’
What does it entail to redefine the inquiry into agency in light of the computational turn, enabling the integration of views from below and above? This integration aims to offer a comprehensive account of agency and, by extension, artificial intelligence—remaining revisable and sufficiently critical to counter unjustified hopes and fears associated with future agency or AI.
This open call advocates for conceptual and methodological demarcations post the computational turn. It urges the creation of an open-source armamentarium—comprising tools, techniques, critical and cognitive resources—to be shared transparently. These resources should be shaped by philosophical considerations surrounding the nature of agency and the imminent phenomenon of AI in the era of the computational turn.
A lattice is presented in terms of a tentative number of Cs and Vs. Here Cs roughly correspond to the main issues raised by the view from below, such as the question of Cost (computational cost, space, runtime, computational power or metabolism), Complexity (the issues around the question of structure and model as, for example, defined by algorithmic information theory, logical depth, and more recently, by theories around what counts as a complex system as proposed by the likes of James Crutchfield, James Ladyman, and Karoline Wiesner), and Concurrency. What is meant by concurrency here, in the vein of theoretical computer science (see the works of Edsger Dijkstra, Samson Abramsky, and Giorgi Japaridze, among others) is a generalized and refined concept of computation capturing processes and events that could not be otherwise accounted for by classic sequential theories of computation. These include long-range, multimodal, synthetic behaviors and abilities which require forms of computation or interactions between agents and their worlds that are not merely one-to-one responses between a predetermined state of the machine and a predetermined state of the environment, one agent and others. Theories of true concurrency have advanced a whole new horizon on the undergirding mechanisms responsible for rich behaviors, and in that, they have presented the concept of computation beyond a solipsistic account of the machine or the agency which only reacts to the environment, and the environment is only defined in terms of responding to specific actions of the agent or the machine.
While the wefts of the mentioned lattice are determined by a set of Cs—still under negotiation—its warps are currently referred to as Vs. In the context of Big Data discourse, Vs relate to computational powers for data handling, specifically velocity, volume, and variation. In contrast to the conventional approach to data and its management, we propose additional parameters or Vs, including Value and Veridicality (distinct from mere veracity). This expansion aims to account for the conceptual behaviors of the agent, particularly the role of judgments in actions and attitudes, in computational and logical terms. We contend that without considering both the wefts and warps of this lattice of Cs and Vs, examining the question of agency, AI, and intelligence across cognitive, social, economic, and political dimensions may lead to responses laden with negative biases. Neglecting these considerations inhibits a satisfactory response to the phenomenon of AI and the acknowledgment that we are currently in the pre-history of intelligence, leaving reactions entangled between unjustified hopes and unconfirmed fears.
This invitation encourages participation in a shared repository of questions, concerns, and methods centered on contemporary manifestations of agency in its myriad forms (AI research, political organization, economic paradigms, etc.). It encompasses research on the artificialization of intelligence, theories of social organization, and economic configurations, emphasizing their implications for decision-making and action. The demarcation of this communal pool or open-source tool-set is guided not only by philosophical insights from the legacies of empiricism and rationalism but also by scientific constraints defining structure, organization, behavior, and appropriate responses to environmental dynamics. These considerations aim to navigate the increasing complexification of behaviors and attitudes among interacting agents.
The AI Research Reading Group: The complexities of velocities and the velocities of computation
Oscillating between unfounded suspicion and fervent faith, encompassing fears, and aspirations that fuel human hubris and uncertainties, a considerable portion of the popular discourse on Artificial Intelligence revolves around its ascent and prospective impacts. Yet, as a direct consequence of this prevailing trend in discourse, which not only shapes public opinion but also steers AI-related policymaking, a myriad of significant issues and themes regarding concepts, methodologies, and implicit assumptions about the nature of intelligence, both general and artificial, remain overlooked or dismissed.
This research group is dedicated to exploring, uncovering, and critically examining precisely these underlying assumptions deeply rooted in computer science, logic, philosophy, natural sciences, engineering, and even economics and politics. Through an investigation of these theoretical and practical orientations and premises, the aim is not only to dispel self-imposed dogmas concerning human cognitive abilities, creativity, and intelligence but also to appropriately contemplate and respond to the intricate relationships among humans, AI, and the social, technical, and political frameworks. These relationships serve as both a crucible and a cradle for future intelligence.
The reading materials for this research group will encompass introductory resources on the ongoing advancements in Artificial Intelligence, classic texts anticipating the rise of Artificial Intelligence, and seminal essays by researchers, developers, proponents, and critics, spanning both historical and contemporary perspectives. No prerequisite background knowledge or experience is assumed, as the group strives to make accessible the arguments and dilemmas while fostering an environment of learning and collaboration.
Suggested readings: AI Research group repository (updated frequently)
John McCarthy, Marvin Minsky, Nathaniel Rochester, Claude Shannon. 1955. A Proposal for the Dartmouth Summer Research Project
Marvin Minsky. 1985. Communication with Alien Intelligence
David Kirsh. 1991. Today the Earwig, Tomorrow Man
James L. McClelland and Matthew Botvinick. 2020. Deep Learning