February update: Eliade again.
Like a dog returns to its vomit, I return to Mircea Eliade's historiography as a way of exploring genevan neoliberalism and californian singularities.
Fragmentary update this time, for reasons that will become apparent.
What I’ve been reading this year:
Richard Bourke’s Hegel’s World Revolutions
Daniel Velleman’s How to Prove It: A Structured Approach
Peter Wake’s Tragedy in Hegel’s Early Theological Writings
Peter Singer’s Practical Ethics
Amartya Sen’s Idea of Justice
Martha Nussbaum’s Monarchy of Fear
John Rawls’ A Theory of Justice
Dimitris Xygalatas’ Rituals
Rachel Chrastil’s Bismarck’s War
Robin Wall Kimmerer’s Braiding Sweetgrass
Leo Strauss’s Studies in Platonic Political Philosophy
Jonathan Sumption’s Trial by Battle (Hundred Years War Vol. 1).
Singer’s book was probably one of the most evil things i’ve ever read. Nussbaum’s work wasn’t as good as her previous work on cosmopolitanism; this was mostly ‘Trump bad’, which yes we know he is, but that doesn’t really explain anything.
I’ve inevitably also been on a Hegel kick, Wake’s book was so good I wanted to shout about it from the rooftops, ditto Bourke.
What I’ve been writing:
I started by drafting something about how Hayek’s negative theology, which Slobodian identifies as characteristic of the heretofore under-acknowledged Genevan School of neoliberalism, had similarities to the epistemology of machine learning and, in general, AGI.
Concurrently, I was hoping to show the embededness of AI in terms of ethics, worldview, history etc with the wider historical and philosophical tradition. This is a wider project of mine as I hope to do some postdoctoral work on the matter.
Half a dozen mediocre drafts later, I’ve come to see these two things from a different angle: Mircea Eliade’s mid-century concept of the “terror of history”. How, in other words, these two fields’ reactions to historical crises had similar intellectual journeys and outcomes.
Slobodian identifies the roots of the Geneva Schools’ distinctive epistemology in the wake of World War One. From a post-imperial Vienna(!), Mises and Hayek wondered how to rebuild a continent of liberal empire and free trade from a shattered continent of new nation states.
At first, their answer lay in mathematical models and nation states. Business cycle research institutes would identify the world economy’s position in these cycles of boom and bust and translate this data into charts for nations’ politician and businessman alike. In so doing, a nation’s public and private economies could work together for the best economic outcome and, in turn, each nation would be interlinked to one another via obedience to the same information. The ecumenical world of pre WW1 Europe could be rebuilt, with one eye on the doubled world identified by Carl Schmitt a few decades later, by forging dominium (the world of business, markets, trade, private relations) and imperium (the general power of government and its administration). To this end, all information had to be freely distributed with no concept of private ownership of accounts. Such lack of privacy would, in turn, reward the capitalist with a global freedom of labour and capital that would sweep away economic sovereignty, tariff walls, and surely history itself. At the end of the 1910s and through the 1920s, as international statistical associations and free trade flourished, it appeared that this mix of national coordination and economic science would triumph in its return to the status quo ante.
This is not, of course, what happened. The great depression, and in particular the suspension of the gold standard, meant that there was no longer a common measure of value to underwrite international trade. Dominium was shattered. Tariff walls and imperial preferences rose.
For the Genevans, the shock was profound. They were quick to identify the main problem: democratic interference caused by moves towards universal male suffrage after the Great War. Rather than being the focus of international statistical research then, nations would have to be encased in an international legal and federal framework in order to mitigate the ability of democracies to assert economic sovereignty.
The democratic shock caused a more pessimistic epistemology. Indeed, the Great Depression and turmoil of the 30s and 40s led the Genevans to reject the very possibility of knowledge, particularly as a source of economic or political prediction. Since, Hayek inter alia contended, knowledge was diffused among a population who act without reason in response to unknown stimuli, perfect markets were impossible. We cannot say what influences an economic decision we ourselves make, let alone the sum of those decisions at a national or global level. In turn, talk of cycles, prediction, or economic knowledge was worse than nonsensical: by making economic life legible as a series of exchanges and allocations, it gave dangerous ground to the idea that planned central economy was possible. Economics was no longer a science but a legal and political project to build international institutions capable of protecting this ‘order’.
The development of a sublime world-order, beyond both categorisation and representation, in the wake of a world crisis and its aftershocks therefore leads us to wonder about how the Genevan’s historical context contributed to their world view. Slobodian highlights how their position in Vienna underscored their thought, particularly in their nostalgia for late Victorian Habsburg Europe. To build on Slobodian’s work, I’m going to bring in - who else - Mircea Eliade.
Eliade argued that the default world view of most human groups is cyclical. Events are meaningful only insofar as they repeat some ur-myth, some golden age, which - with one eye to the lunar and agricultural cycles - goes through apocalypse and rebirth. Typical narratives include some original fault which caused humans to fall into profane time, a sinful populace that leads to a flood, and the survivors forced to rebuild. The cruelties of life are therefore meaningful because they partake in and commune with that “actual” reality, and in so doing promise a final victory even after the gravest defeat. Counter to this, of course, is Mircea’s observation that this points to a fundamental terror of profane time and therefore history; of events as such, which have no meaning and therefore no possible consolation.
Judaism ushered in a new and profoundly unsettling alternate model of linear history. The prophets challenged the people to find God in profane history. In other words, they challenged the Israelites to root themselves in the unfolding bare facts of life, and there to encounter the divine Will, good or bad. As Eliade notes, such a world view was profoundly unsettling. It require almost superhuman will. People were called to face the terror of history without certainty, and without any sense that their acts reflected and partook in some universal narrative. There was only the bare event, in which one might perhaps triangulate a fickle divinity’s pleasure or displeasure - but the event had no meaning in and of itself.
The desire to escape history, to rejoin that golden age, the sense of the relentless march of profane time and uncertainty, continued to be found in Jewish and Christian life. Bare materialism was difficult enough in the 19th century, Eliade observed. How much more difficult was it from his position in the mid 20th century? The impossibility of facing history in itself was always an elite occupation; most movements, even the most ostensibly materialist Eliade noted, offered some sort of an escape through transcendence or rebirth. Christian baptism, the lives of the saints, the day of judgement, the great flood; all reflect the fundamental cyclical way in which humans tend to think about the divine in terms of lunar myth.
Genevan neoliberalism’s distinctive epistemology, where Hayek rejected first universal predictive statistics and then turned to a transcendent sublime order, were therefore reactions to the problem of history. The one-two blow of Great War and Great Depression and their apparent contingency revealed the contingency of historical order to Mises and his circle. It was one matter for the Great War to shatter a secularising continent. It was quite another for the resulting embrace of cyclical ontology to cause not unification - indeed, a return to an Eliadian golden age from which the Genevans found themselves ejected - but to drive further chaos. Indeed statistics and models encouraged people to believe in a rational cyclic order. This led not to self-knowledge, but to a belief in the ability of humans to perfect themselves, their countries, and the world through planning. The Genevans worried then that explicit knowledge of historical cycles risks encouraging history rather than merely comforting its victims.
Encasement was therefore a desperate attempt to roll back the clock; to restrain nations from delusional beliefs of agency, reason, and perfectibility and instead link daily life to the prelapsarian Golden Age that had ended in 1914. What Slobodian calls a “negative theology” of world-order transcendence, maintained and protected through international federal entities and bodies, in turn reveals the Genevans’ obsession with returning to the eden of the last century - of escaping, in other words, the knowledge of cycles and markets in favour of sacred time and emergent order. In much the same way they had argued reason and prediction and openness might lead to transcendence, they now inverted this in the hope of forgetting. Slobodian’s point that the Genevans became concerned with extra-economic life at the same time they came to understand the ignorance or unreason of our daily choices therefore suggests that the Genevans were tormented by self-knowledge. What Schmitt called the doubled world, then, might also be for Eliade the world of the sacred and profane. Federal institutions, by creating and maintaining the sublime world order through encasement, sought to bind all facets of daily life to the sublime order. By getting rid of the distinction between economic and personal life, they hoped to ignore profane time and the terror of history.
Machine Learning went through a similar intellectual reaction to world crisis. Whereas in the Genevan’s case this crisis was the First and Second World Wars, for the the Californians it was that of 9/11, the GFC of '08, and the Eurobonds crisis. As Tooze has documented, these ushered in the turbulence of the 2010s and founded the world in which we find ourselves in the 2020s. Like the Genevans, the first reaction was a turn to predictive statistics - big data, granular analytics, social networks - that would explain and model an American quasi-Empire in crisis. At first, Tahrir Square and the Arab Spring’s ‘Facebook Revolutions’ promised a digital dominium. National and religious difference would be elided in digital citizenship of the ecumenical world. When the Arab spring failed, wars intensified, ISIS rose, and populism conjured up economic sovereignty, the Californians turned to the opaque and sublime: neural networks, reinforcement learning and generative AI. These technologies were, like the Genevans’ late ideas of world order, almost beyond representation; operating at a scale and complexity impossible for humans to fully map, let alone comprehend. Like the Genevans’ models of human behaviour, they rested partly on tuning reactions to stimuli in particular environments. But even the link between input and output was difficult due to the inscrutability of the dense networks themselves. The belief that input a straightforwardly mapped to output b was false; models could be ‘overfitted’, their behaviour over-optimised to their reward function.
Big Data, predictive forms of artificial intelligence therefore rose - like business cycle research - in the wake of world crisis. They promised to create an ecumenical world in which power was flat, information was free, and where nationalist fervour faded into the background of the global community powered by knowledge and processed in CPU cycles. When financial crisis spurred populism, data sovereignty (both GDPR, the location of data centres, and more recently copyright of LLM training sets), tariff walls, and whispers of economic sovereignty, the Californians in turn rejected the possibility of total knowledge or global digital community in favour of an epistemology of opaque transcendence. In other words, open source data and legible algorithms were replaced by dazzlingly dense networks, legible only by stimuli, and for which pattern matching was a better metaphor than prediction.
It is in this historical context that we should place the alignment problem. Alignment is ensuring that human and machine goals align both in terms of goal and process; in other words, ensuring we don’t ask an Artificial General Intelligence to maximise happiness, only for it do so by giving us all an overdose of opioids. This is particularly important in terms of reinforcement learning, in which programs’ behaviour in environments is formed via that behaviour’s reaction to a measure of reward. The worry about alignment - via talk of safety, killswitches, and the transcendence of the singularity - along with the looming threat of unemployment from generative AI, has led to the formation in the mid 2020s of federal bodies devoted to AI safety. These spaces, I contend, have sought to “solve” the alignment problem by removing the possibility of democratic control. Just like the Genevans saw economic life as something that needed protection from democratic accountability, from trade unions and conservatives alike, so too did the Californians come to view AI in terms of transcendent, sublime order that needed protection from the misguided optimists whose life nevertheless contributes to and is rooted in the utopic “cloud”. In both cases, the perceived problem was common belief in a rational, predictable world-order. Initial promises of “transparency” led not to acquiescence but demands for data legibility, economic sovereignty, and subordination to national and religious identities.
In both cases, the answer was to remove the human element.
Both Genevans and Californians therefore reacted to double world-crises and the threat of nations by adopting predictive statistics, only to abandon it in favour of encasement and transcendence. Both of their epistemologies were reactions to the march of history and the problem of suffering and the visceral shock of unanticipated mass violence. Both involved Schmitt’s doubled world and the nostalgia for the certainties of liberal empire, austro hungarian and american alike. Both, ultimately, sought to find a way to link daily life back to the sublime and prelapsarian state.
AI safety, then, as an international and federal project, cannot be taken as a de facto space in which there remains a possibility of democratic control. Rather, as Slobodian argues
Geneva School neoliberals prescribed neither an obliteration of politics by economics nor the dissolution of states into a global marketplace but a carefully structured and regulated settlement between the two.
(Slobodian, Globalists, 12)
WTO, the GATT, and the EU were the spaces in which 20th century global order was negotiated and national demands suppressed. The emergent AI policy spaces in California, London, and Beijing risk fulfilling the same function in the 21st.