Just as long as the results are never used in actual worship.
The very idea of using soulless deterninalistic models to compose prayers in place of seeking inspiration from the Holy Spirit give me serious pause.
Naturally, we want our worship to be inspired by the Holy Spirit, particularly when it comes to something like an anaphora. And ChatGPT and other large language models, which are not only soulless but literally incapable of consciousness, since all these programs do is function in, as you say, a deterministic model, albeit one so complex that unlike a conventional computer program, the neural network that drives an AI is effectively a black box, and indeed it is trained more than it is programmed, with extreme amounts of computing power and specialized hardware used for the training process, but ultimately, it is still a machine created by humans, and since it can only output text, it could not reproduce itself, although multiple Ai networks can be used to produce each other - this is still basically a glorified computer program, relying on elaborate software running on high end hardware. But essentially, it is only an infinitely more advanced abacus.
However, computers are a valid tool for purposes of liturgics; computer software can be used for example to automatically generate the texts for the services with the appropriate propers based on the Eastern Orthodox typikon, and avoid potentially embarassing mistakes from incorrect use of the typikon, which is known to happen, and consequently the use of software in this way could silence the liturgical reformists in the fringes of the Orthodox Church such as at New Skete Monastery who propose a radical simplification of the liturgy and its rubrics, and enable instead the liturgical maximalism historically cultivated by ROCOR to be expanded to other jurisdiction. Indeed, in different Orthodox churches, software like this is already in use, and there is also such software available in the Roman Rite (santamissa.org , which generates the propers for the Latin Mass and the Divine Office according to a variety of versions of the service books). There is a superb app for the Coptic Orthodox liturgy called CopticReader which contains the entire Coptic Rite liturgy and can generate entire services with the correct propers for the day, without having to pull out a copy of the Psalmody, or the Khiak Psalmody, or the Euchologion, the Lectionary, the Synaxarion, etc, and additional sources which historically were documented only in relatively rare and poorly translated books, for example, the Paschalion, the English, Coptic and Arabic parallel of which I have, but which is very poorly translated. All of this is of course done using conventional software.
The legitimate utility of AI is to simplify the writing and maintenance of such software, as AI can efficiently translate human guidance into machine code, but one still needs to be a programmer qualified in the language in question in order to be able to communicate with the AI, debug the software, and also make sure it functions correctly, so essentially the AI is used to make programming less tedious, and to do things like check software for security errors, and perform changes to the code that could not easily be done using the existing basic UNIX or GNU text processing tools at a Linux, BSD, Solaris, Cygwin, MacOS or other UNIX-like command line, programs in use since the 1970s like grep, awk, sed and so on, which are immeasurably useful for text processing. The AI can also help with the use of these tools, provided it has been programmed in their use, and it can also in the case of ChatGPT be integrated with them, as there exist multiple programs which access chatGPT from a UNIx-like command line, so one can talk to it via the standard input and standard output, allowing its incorporation in a typical pipeline of UNIX like utilities.
In like manner chatGPT can be used for language translations, where it is infinitely superior to existing software, but herein liturgical applications would still require someone competent in both the input language and the output language in order to ensure that the output translation is of sufficient fidelity. It can also be used to translate between ecclesiastical English, which ROCOR has wisely retained, and contemporary language which is demanded elsewhere, to my frustration; I myself never like prayers in contemporary language, and to a certain extent one could already use tools like grep and sed, or the functionally similiar but less powerful Find and Replace feature in Microsoft Word, or a macro expansion program for that matter, to take an incorrectly translated phrase such as the dreaded “And also with you” popularized by the dreadful 1969 English translation of the Novus Ordo Missae, which Pope Benedict XVI to his credit insisted be replaced by the more correct “And with your spirit” and in this way greatly improve works like the 1979 BCP, or the 2006 Lutheran Service Book, by applying the late Pope Benedict’s change to them automatically. However, using ChatGPT, a complete change can be made to the entire text, albeit not all at once, since chatGPT can only handle about five or six paragraphs of text at a time (this post would be approaching its limit). This is also extremely beneficial because it promotes for the conscientious use of the AI system in performing such a conversion, by providing a convenient window of time in which to double check the reformatted text’s outputs, which is necessary as it is likely to make mistakes, for example, it is not intelligent enough to know the response to “and with your spirit” is a single person, but it is generally intelligent enough to figure out if a sentence is a prayer addressed to God, so converting the former sentence to “and with thy spirit” is a case where manual intervention might be required, although it is unlikely because the AI software’s vast training data fortunately included enough liturgical texts so that it was, for instance, able to generate a convincing West Syriac anaphora with only one obvious mistake, as seen above, from one line of input, and then convert it to the traditional ecclesiastical English I posted with another.
However, in life in general, and especially in all matters to do with worship and prayer, complacency is to be avoided as slothful, and one must instead strive to be attentive, reverent and conscientious. If chatGPT or any future AIs are used in a complacent and slothful way, for example, if one just assumed the text it provided was accurate without checking, the results would be abhorrent for liturgical purposes, and disastrous if used in, for example, engineering, but these are valuable new tools if used correctly.
Another interesting aspect is liturgiology; if one trained GPT on enough liturgical texts it might be able to provide new insights on the origin of specific parts of liturgical texts, because most of them are a synthesis of influences from different places. For example, we know the Eastern Orthodox liturgy is a synthesis of Antiochene and Jerusalemite Eucharistic prayers, and the typikon and the Divine Office it specifies are the synthesis of materials compiled, but not neccessarily written by, the monks at Studios Monastery in Constantinople and Mar Sabbas Monastery in Palestine. However, what exactly is of Studite versus Sabaite origin, and how the balance shifted, for example, between the older typikon still used in the Russian Old Rite, and the newer typikon used on Mount Athos, and those churches that use it such as most of the Russian churches and the Georgian, Serbian and Jerusalemite patriarchates as well as by Greek and Romanian Old Calendarists. And also it can be used to quickly identify subtle differences that have formed over the centuries in the liturgical texts used by the different Eastern Orthodox churches, so that these can be preserved as part of the distinctive heritage of Holy Tradition as received in those countries, and such differences do exist. For example, even among the Russian Orthodox, some advocate using red vestments and paraments on Pascha, whereas others insist on white, and indeed this example is not reflected in the typikon which only specifies light or dark vestments, but it is an example of an emergent variation in liturgical usage that is obvious, but there are other subtler variations that an AI could detect that most human scholars would miss, and which are the sort of thing that could be inadvertantly or intentionally removed by ill-advised attempts at liturgical standardization.
But indeed, to be clear, I am not advocating that we go and tell chatGPT to write our hymns and services for us and then pray whatever it comes up with. Although I daresay it could do a better job than some late 20th century composers of liturgical books, who I need not name, for I think at this point most people in Traditional Theology would agree with me that their output has been a disaster. I also expect it could do a substantially better job than the rather creepy bishop who led the Rennovationist heretics who the Soviet government tried to give the Orthodox Church over to. I have a copy of his Rennovated form of the Divine Liturgy, and it is pretty bad, being an eclectic mix of Byzantine influences, some random stuff borrowed from the Ambrosian Rite in Milan, inexplicably, and other peculiarities. Although I have seen worse in recent years, from some extremely liberal churches, but the Rennovationist liturgy is probably the worst liturgical composition of its era.
Although prior to the communist takeover, there opened in Romania this weird Spiritualist-influenced heretical church which has an iconostasis and uses vaguely Byzantine liturgics, which I believe is still in operation, and I suspect its material would be worse, in that as incompetent as the liturgy of the Rennovationists was, it did not include anything blasphemous, which unfortunately cannot be said about some liturgies composed by the far left extremists in some mainline Protestant churches starting around 1990 or so, for example, the notorious output of the Womanist theologians that happened at the Reimagine! Conference around 1995, which featured a liturgy celebrated using milk and honey, which is a direct and explicit and probably deliberate violation of one of the ancient canon laws of the early church, which explictly prohibited that (whereas two other canons prohibited celebrating the Eucharist with water, or with fish in addition to the bread and wine, and the Mormons do the former, like the Hydroparastae of antiquity, while I am sure a survey would find some random duffers celebrating the Eucharist with fish (what would we call them, Icthyparastae?).
