J. E. Landau

This year, dozens of companies have pivoted from cryptocurrency mining to providing AI datacenters. These groups, once focused on extracting digital currency using warehouses full of advanced computer hardware, have found themselves unsustainable and are looking to redirect their resources towards artificial intelligence. They believe, like many others, that AI will grow explosively, and indefinitely, and that it will require ever more extensive infrastructure. This promise has led to the creation of trillions of dollars of market capitalization and the continued growth of the tech industry.

However, it is possible that artificial intelligence will fail to scale as projected. This raises a problem for the tech industry: what then?

Something is needed, which can make use of the gigawatts upon gigawatts of datacenter capacity springing up to accommodate AI. Something exciting enough to attract enormous amounts of capital, and the imaginations of the public.

The authors believe that they have identified a promising solution: the establishment of a religion.

Arthur C. Clarke was the first to propose this solution, albeit for a much smaller enterprise, and one that was actually profitable, given that “The Nine Billion Names of God” was written in the rather primitive year of 1953, when profit was assumed as a goal, a misconception that can be dispensed with. In that excellent story, a company provides a monastic order with a computer to record of the names of God. Being an older story, the corporation involved fails to realize maximum benefits, but the idea is useful: what if a corporation worked with religion?

It might seem strange to incorporate religion into the tech industry. Despite appearances, these connections are already well developed, especially in the AI space. Figures such as Ray Kurzweil, Bryan Johnson, and Sam Altman are pioneers, and their religious expertise could be extremely helpful in developing a new solution for the future. And it is possible to go further.

As was discovered by Pascal, religious narratives have the advantage of offering the greatest possible return on investment: an infinite one. These promises may not seem as impressive as they once were, given the recent history of the tech industry, but an infinite potential return is still impressive and novel. The authors expect that there might be a correspondingly infinite leap in investor enthusiasm. Ideally, the theocorporation should seek to establish an infinite price-to-earnings ratio as well.

What should a company religion seek to do? It should seek to attract continuous investments, to funnel that money into executive salaries and to help fuel the compute economy through infrastructure spending. A certain level of competition could be helpful here, by leading to bidding wars for people who are designated experts on technology as a path to salvation (e.g. executives, or researchers, as seen in AI).

Actually delivering any sort of salvation, however, is irrelevant. The company simply needs to be able to attract additional funding for as long as possible, to make its creators extremely wealthy. On the other hand, investors would like to receive the rewards of their investments during their lifetimes. Therefore, a pathway to metaphysical profitability should be long, but not too long. Thirty-five years seems like a reasonable estimate, although theological engineers should perform their own calculations to arrive at an appropriate figure.

This also raises a question. What computational task is hard enough to occupy a growing collection of datacenters for thirty-five years? In truth, the question is just how many variations on a given problem you need to solve. An aspiring theological engineer could look at cracking some very nasty hash function, which would use an arbitrarily large amount of computational power, but it is possible that advances in theory or new computing technologies (like quantum computing) could make the problem much simpler, leading to the company running out of financial runway.

Instead, it makes more sense to choose a set of simple problems that can be solved using basic operations. The authors propose doing some set of floating-point operations many times. 100 floating point operations, perhaps 108 for its traditional power, seems like a solid threshold to make things feel mysterious and confusing to the public. Including obscure mathematical operations, such as hyperbolic trigonometric functions, will add mystique. All these calculations will have to be justified, so perhaps it would be wise to have some cosmic, pseudoscientific connections to the universe (cues could be taken from the various “quantum” industries) and references to traditional ideas of sacred geometry (here, the supplements and self-help industries could be very helpful). A theocomputational operation could use quantum random number generators, cosmic microwave background measurements, the positions of stars in galaxies, or other mystical and space-related data as the fodder for computation to enhance these links. The AI industry, however, has been able to establish such links in public reporting without doing any cosmic measurements, so it may be possible to skip those steps. For the purposes of this proposal, each set of 108 operations will be called a “name of God”.

How many names of God should there be to tabulate? That’s difficult to estimate. Unfortunately, the datacenter industry has chosen to provide figures for computational power in terms of watts instead of flops, but some estimates can be made. According to a report by McKinsey and Company, the AI industry is expected to grow its datacenter usage from 44 to 156 Gigawatts between 2025 and 2030, which is a solid figure to start with, since it is possible that the AI datacenter operators will be looking to pivot in that timeframe.

Assume this religion starts at a similar figure and grows linearly for the full thirty-five years, which is a good scenario for maximizing income. In this case, assuming that these computations are all as efficient as the most efficient supercomputer in the world, according to the TOP500 supercomputer rankings, which should account for technological improvements in efficiency (which are undesirable for the purposes of this project), each watt of electricity spent will give 72.733 gigaflops per watt. Note that any theological engineers should look for values more appropriate to their situation and adjust figures accordingly.

Multiplying all these values together gives a total computational capacity of around thirty-five nonillion floating-point operations. Assuming that there are 108 floating point operations per “name”, that gives a total of a little more than 324 octillion names to decipher. Just in case, this figure is rounded to three hundred and fifty octillion names of God.

It should be noted that although prior calculations were given in floating-point operations with no precision specified, it will be necessary to use quadruple-precision floating-point numbers (or better) to store these names of God, to describe at least 350 octillion distinct values. Of course, it is possible to simply calculate the same values multiple times, with justification, but it is unwise to risk doing so, as it might lead to public and investor backlash. One never wants to run out of financial runway before launching a product.

That is doubly true when the product one is developing is heaven and eternal bliss. There has been some pioneering work in this field by innovators like Ray Kurzweil, but messianic belief in artificial intelligence has grown from many places and can be seen in public fora such as LinkedIn, indicating that strong groundwork has been laid for future religious projects.

The proposed utopia should have infinite abundance and joy for humanity, but it might be wise not to distribute those benefits equally. Instead, those pursuing this concept should add a degree of salvation through works. If you say that it is necessary for those who wish to be raptured or purified to donate, work, or invest, this would be helpful for drawing in money and talent.

It might also be helpful to threaten damnation for those who do not contribute. This sort of thinking has been used by the AI field for some time, with the idea of contributing to setting the goals of a hypothetical future superintelligent artificial intelligence acting as the promise of controlling heaven for AI researchers and investors, and the theorized torture of “Roko’s Basilisk” acting as hell for naysayers.

When attempting to build a following for this new religion, it would be wise to co-opt what has come before. Converts can likely be won from AI by incorporating their ideas. Perhaps the computations are intended to understand the simulation that we live in or to communicate with extraterrestrial superintelligence. If the AI industry falters, there will be many looking for something new to follow, especially investors.

It may also be helpful to covertly hire several paid doomsday prophets, science fiction writers, and people of influence to control the narrative. Likely many of these people will be looking for a new faith as well. Figures like Eliezer Yudkowsky are especially important, to influence many of the powerful people whose ears they seem to have. A new version of the Future of Humanity Institute could be covertly sponsored, or universities could be given grants for supportive studies. It might be helpful to contact politicians and businesspeople who have a strong interest in keeping the demand for datacenters and computational hardware high. These figures probably do not even need to be paid directly, but a little lobbying certainly wouldn’t hurt. Leakers claiming that the work is dangerous is a novel tactic, applied by AI companies, that could also be helpful in creating public hype around the technology.

Based on this analysis, the authors conclude that it would be possible and even easy to start a new religion for the tech industry. The ground is ripe. Many want to see the rapture in their lifetimes. And, more than hoping for heaven, the world lives in fear, that someday there will be nothing left to pour trillions of dollars into.

On that day, there will be no “next big thing” that will grow exponentially, no more talk of hockey stick graphs, no more sky-high valuations, no more unicorns and VC darlings.

This is why it is imperative to ensure that there are enough names of God to make your money, retire, and find some other executive who’s either a true believer or someone who doesn’t care to think of what will happen once the last name is counted.