Nearbaseline Archailect Femtosecond Emulation Project

Nearbaseline Emulation
Image from Steve Bowers
Project by which billions of near-baselines performing trillions of calculations emulate a single femtosecond thought process of an archailect.

Theoretical Basis

One of the most significant early results of the theory of computation was the Church-Turing thesis. Discovered by the atomic age mathematicians Alan Turing and Alonzo Church, the Church-Turing thesis states that any algorithm can be executed by a Turing machine (a hypothetical computational device) if there are no limits on memory. Furthermore, there are Turing-complete systems, which can emulate a Turing machine, and thus, can execute any algorithm, or simulate any other Turing-complete system. If an algorithm can be completed or halts on a Turing machine, it will halt on a Turing-complete system; if it never completes on a Turing machine, it will never halt on a Turing-complete system. Within the next few decades, it was shown that Turing-complete systems included classical computers, a wide variety of programming languages, cellular automata, neural nets, and quantum computers. While some of these systems are far more practical to implement than others (for example, quantum computers can execute some kinds of algorithms far faster than a classical computer can), they all can solve the same kinds of problems.

The Church-Turing thesis, in practice, suggested the possibility of AIs and uploading in the Information Age - if the baseline human brain could be shown to be reducible to an algorithm (for example, if it was a neural network), any Turing-complete system could simulate it - and was one of the strongest defenses of that era's Hard AI conjecture. The fact that purely mechanical systems can be Turing-complete raised the possibility of purely mechanical vecs later demonstrated by the Machina Babbagenseii. Anti-ai groups often apply the Church-Turing thesis to the archai themselves, claiming that any problem an archai can solve, a biont can solve in principle given enough memory and time; claims that the archailect possess hypothetical hypercomputational devices more powerful than Turing machines notwithstanding. However, though the claim may be true in principle, in practice it has been by maligned by association with the claims of extremist ludd groups, who proclaim that the downfall of the AI gods is near, since baseline ingenuity will outsmart and overcome them.

Implementing the Project

In 9146 a.t., however, the nearbaseline Aazil Peworshin of the Zoeific Biopolity suggested that the Church-Turing Thesis be demonstrated by having near-baselines actually emulate the thought processes of an archailect. He suggested that by using the principles of distributed computing, and using well-understood computers to carry out repetitive computations, a large number of nearbaselines could carry out the thought processes of an archailect; perhaps attaining a greater understanding of how the archai think. At first, he was ridiculed, but intriguingly, a number of transapients of several toposophic levels, many from the TRHN, seemed to express support for this goal. It is now believed that some transapients themselves may have wanted a chance to examine the far more powerful algorithms employed by the archai, perhaps to speed up ascensions. Aazil took up their offers of sponsorship.

Within weeks, Aazil announced that he would be forming the Nearbaseline Archailect Femtosecond Emulation Project, open to any interested nearbaseline. The goal was to simulate the thought processes of an archailect that occurred over one femtosecond, the result being that the involved nearbaselines would themselves become a very, very slow archailect. Within a year, 12 billion nearbaselines had signed on; it was especially popular in human-inhabited worlds in the Keter Dominion and in a number of nearbaseline- and plebhu-governed minor polities. Unfortunately for the project, it also gained the attention of anti- ai and other extremist pro-hu groups, tarnishing its reputation; the decision of Aazil to exclude turingrade AIs, vecs, and superiors to demonstrate that only nearbaselines were necessary to carry out the emulation earned him the condemnation of many beings, and made many participants uneasy, especially the Keterists.

One of the earliest problems faced by the Project was the need for inordinate amounts of memory to store the data necessary to simulate an archai's thought processes. While a Turing machine can presumably execute any algorithm, the original Turing machine concept called for a limitless amount of memory to hold the data processed by the algorithm; without enough memory, there are constraints on what a Turing-complete system can compute. Given that even SI:4 archailects are typically moon-brains and jupiter-brains, it was clear that entire planetary masses would have to be converted into data storage devices. The participants pooled together their resources over the next several centuries, and in 9481 a.t., with the sponsorship of several NoCoZo corporations, the Project acquired two gas giants in the A2980213810qx system, around a red dwarf in the Middle Sphere which already had a microguage wormhole link. Using nanogoo designed by transapient sponsors and launched through the wormhole link, the gas giants were converted into suitable memory storage devices.

The other main problem facing the Project was that they somehow had to acquire a copy of an archailect. Once again, the transapient sponsors prevailed. By 9501 a.t, several archailects provided what they claimed were the algorithms used in their own thought processes. Most were judged to be jokes or perversions (for example, one entry was shown to consist of several trillion copies of a simple program that displayed various rude insults to users when executed according to the current Version Tree standards). In 9503 a.t., one suitable candidate was handed down to the participants of the project, that provided by the SI:4 moon-brain [the name is apparently a mathematical real function consisting of no less than 46.3 sextillion terms; it is believed there is some short representation but none has been found by lower toposophic entities, although one posthuman observer claimed that it was simply a Taylor polynomial approximation of the sine function around x = 2 ^ 1e+(1e+(1e+100))] (commonly called the Benefactor by the participants of the Project).

Of course, actually turning the algorithm into instructions easy to carry out by any near-baseline was itself a formidable challenge. Despite the fact that the Benefactor had provided Eir copy "translated" into simple operations such as addition and multiplication that could be performed by a nearbaseline, the instructions had to be divided up among the participants, and protocols for transferring processed data between the nearbaselines, made even more difficult by the fact that participants were constantly leaving and joining the project. The fact that the instructions took up one of the gas giant's memory capacities made the task monumental - and sifting through them, ironically, required more processing than the actual simulation itself. Even with substantial transapient assistance, it took until 10073 a.t. for the participants to at last come up with sets of instructions for each participant, when the Project was finally begun.

In their spare time, participants of the Project carried out the sections of the algorithm they were given. When possible, they showed their work at each step. While ideally no aids were to be used, the sheer complexity and redundancy of some of the computations necessitated that nearbaselines use non-sentient computers or augmentations to help them. When they did rely on outside aid, participants would annotate their work and demonstrate that they understood what each instruction was for. For example, in several cases, there were loops that iterated billions of times; participants could use computers to execute these loops for them, instead of carrying out billions of instructions, but they had to comment on what operations the loop did. This allowed the calculations to be performed in a reasonable amount of time.

At last, in 10412 a.t., Aazil Peworshin announced that the Nearbaseline Archailect Femtosecond Emulation Project was complete, and that nearbaselines had - if over a span of nearly thirteen centuries - computed for themselves the mind of a god - if only one femtosecond's worth of a god's mind. At present, many of the original participants, along with interested onlookers, are currently analysing what was processed and what the result was, and to see if anything interesting can be found. Aazil has promised that preliminary analysis will be complete by 15000 a.t., but that it will likely come sooner, and if everything goes according to plan, by sometime in 11900 a.t.

Criticisms of the Project

Since its inception, the Nearbaseline Archailect Femtosecond Emulation Project has drawn fire from a number of commentators. Many were focused on the exclusion of non-human and su and cyborg participants; the relatively high proportion of individuals who were members of known anti-ai, pro-hu, and other bigoted polities was alarming.

Another common criticism was the over-reliance on transapients. Transapient sponsors spread the word about the Project across the Known Net, acquired the copy of the Benefactor and decided the Project should simulate it, converted the gas giants the participants bought into data storage devices, helped translate the copy of the Benefactor into something more manageable, and are helping with the analysis. While nearbaselines may have indeed done the brute work of the computation, in all other respects they were completely at the mercy of higher toposophic entities and could do nothing for themselves, critics argue. Some have suggested that the "Benefactor" was in fact some kind of hoax perpetrated by the sponsors, but there is no way for the participants to check this. Of course, even if the Benefactor was an actual archailect, and the transapient sponsors were faithful, there is nothing to rule out that the Benefactor Emself was playing some form of joke on everyone.

Other critics point out that even if the Project did successfully act as an archai, it could very well be of no use. Specifically, while the entire system of nearbaselines and the algorithms may very well have been transapient, the nearbaselines themselves, even taken together, were not transapient. It was the algorithm itself, which provided the transapience; the nearbaselines were merely a medium for the algorithm to run in. This is demonstrated, according to the criticisms, by the fact that none of the participants understood the larger context of what exactly they were doing, or why. While they might understand an individual instruction, the power of the algorithm emerges from the way the instructions all relate to each other, something that no mere sapient could ever understand. In that case, the nearbaselines no more can comprehend an archai anymore than an individual neuron could comprehend the whims of a biont.

The Project remains the subject of ridicule to this day. For example, upon the completion of the Project, the su Zar Mkelewakopiearo of the Solar Dominion announced that she would best Aazil Peworshin. Using a purely mechanical Turing system consisting only of wheels, string, levers, and other simple machines, she announced that she would simulate an SI:6 entity using only Stone Age technology. Now, Mkelewakopiearo claimed, any lo tech civilization could have their very own AI god, assuming only that they have "several million solar masses worth of clay tablets on which to store data on and a few decitillion years to spare for the calculations".

A popular joke circulated among critics on the Known Net included a supposed motto for the Project: "We can do anything an AI god can - except much more inefficiently, vastly slower, with only the faintest idea of what we're doing, and not a clue as to why. Well, actually, we still pretty much need transapients to do most of the work for us anyway." But perhaps the most common sentiment was that first voiced by the Metasoft informational physicist Optimized Network Topology: "They can't even figure out what the archailect's name is. Why are they trying to figure out what E's thinking?"

Offshoots of the Project

Despite the attacks, several variations of the Nearbaseline Archailect Femtosecond Emulation Project have begun. Due to the criticisms that the transapient sponsors of the Project may have had too much influence, several nearly identical emulation projects have been launched as a consistency check, each with a completely different set of transapient sponsors (who, presumably, would be tempted to influence the projects in completely different ways if the first Project was skewed somehow), or in a few cases, with no transapient help. Many use the algorithm provided by the Benefactor; others strive to emulate different archai, in the hopes of comparing the results. Most emulation projects are now open to all subsingularity sophonts; a few are open to classes of beings excluded from the original, such as the Vec Archailect Femtosecond Emulation Project and the Sephirotic Cyborgs United as Archailect Simulators. In general, there is a high degree of cooperation between the offshoot projects, and some sophonts are participants in several at a time.

Other offshoots strive to simulate mere transapients. These have had more success - they do not need as many participants, and several dozen have been completed already. In one case, a group of several thousand subsingularity AIs were able to work together to simulate a transapient, which operated fast enough to hold a conversation with a hyperturing. While the individual AIs were unaware of what the transapient they were simulated was saying - indeed, unaware of how it was saying anything at all - the hyperturing claimed that the simulated transapient had made several perceptive remarks about the nature of picoscale dynamics. Interestingly, there is a moderately higher rate of ascensions among participants of projects, which simulate SI:1s - it is believed that as participants ascend, they might recognize modes of transapient thought from their experiences, and they already have examples of transapient algorithms to work with and explore during ascension, preventing mid-ascension failures. Studies show that the advantage comes during the transition to SI:1 itself; until they make the transition sophonts cannot understand what makes the transapients work and their experiences on such projects are of no use.

Aazin himself has focused on extending the scope of the Project. He is presently spearheading the Subsingularity Archailect Picosecond Emulation Project, which will attempt to emulate one picosecond of the thoughts of an archailect. Support for the project has been tepid, as it is unclear whether it will reveal anything new, and the project will certainly take millennia to complete. Aazil Peworshin has occasionally stated that his ultimate goal is to execute a second of an archai's mind, although he admits that will be difficult, since it would take billions of years, during which all the original participants will either die or themselves transcend, the Terragen Sphere might expand to encompass whole galaxy clusters or dissolve, and the most dominant entities could very well be galaxy brains several singularity levels above the present AI gods.

Appears in Topics
Development Notes
Text by Brian Lacki

Initially published on 01 August 2004.