Can we compare the language models of AI with our own deeply-held faith models?
Road maps and musical scores are examples of models. Each is a simplified representation of a complex system and is used to make understanding, communication and decision making easier. A key feature of a model is that it strips out some of the details of the thing that is being represented. So, a road map indicates a road’s location and direction, but not the kinds of things one would see when traveling along it. Similarly, a musical score provides information about the notes, tempo and so on. But each musical performance will be slightly different, as the score on its own cannot capture the subtle nuances introduced by the instrumentalists.
In the field of Artificial Intelligence we have become used to “Large Language Models” (LLMs) of the kind used by OpenAI’s ChatGPT. As the names suggests, these too are models. As the AI system is trained, it is fed an extraordinarily vast quantity of text, and from this it creates a more compact and manageable model of how language works. (The model is still very large, hence the name.) Just as with road maps, musical scores and other models, some information is stripped out. The key to an effective AI learning process is to ensure that the right things are kept while the unimportant details are thrown away. However, inevitably some elements are lost.
It is because of this loss, that the text, images and music that are produced by generative AI have a degree of predictable “sameness” about them. The songs produced by AI music systems have a certain appeal, but they don’t linger in the memory. Moreover, they lack the genuine feel and emotion that a talented flautist breathes into a Mozart concerto – evoking something far deeper than can be grasped from the score alone.
It may come as a surprise to some, but models also play an important – if underappreciated – part in our faith and theology. Neither the rich variety of human experience, nor the inconceivable depth of God, can ever be fully captured and expressed. So we distil our faith into the models of Scripture, creeds, doctrines, canons, hymns, and the pronouncements of church leaders. Yet these models are all partial and provisional. Even the twelve hundred pages of the Bible can only tell us so much about our infinite God. Some of the mysterious and elusive aspects of both creation and Creator must unavoidably be omitted. There is always more that could be said.
We wouldn’t mistake a map for the place itself. Yet, the danger with both the constructions of AI, and with the models on which our faith is built, is to confuse the model with reality. A face created with AI looks like an actual person. The AI-generated sermon resembles wisdom. A proof text from the Bible sounds as if it is the final word. However, in each case, they are only ever an approximation to what is genuine. The antidote to the model is real life.
It is for this reason that God’s ultimate revelation to the world was not a text – a model – but the person of Jesus Christ, living a real life alongside us. It is also for this reason, that Jesus came to bring not knowledge or information – another model – but real and eternal life, more and better life than we ever dreamed of.[1] And it is with this attitude that we should embrace AI – not as real intelligence, insight or originality, but as a useful model – just as a road map is a useful model which aids our own understanding.
Our calling is not to exist in a stripped-down model or digital abstraction, but rather to enjoy real life with God.
The Revd Canon Dr Tim Bull
Mon, 1st April 2024
[1] John 10.10 in The Message.