A voice answers your question. It sounds patient. It uses your name, follows what you said and responds in familiar language. The interaction can feel personal very quickly. What does that experience actually establish?
At PTLT in La Joya on 10 October, I asked Million, my AI assistant, to send a thank-you email to the organiser while I continued my presentation. The email was sent, and the recipient acknowledged it in the room.
That was a useful demonstration of an authorised task. The discussion also raised questions about faith and trust. As I begin this Sunday conversation about AI, faith and religion, I want to start with the response itself.
How a response is generated
I'm focusing here on conversational AI, such as chatbots and voice assistants. AI includes many other kinds of systems. These tools run on computers, often in data centres, and those computers require electricity.
During training, a model processes large amounts of material. Its numerical settings are adjusted as it learns relationships in that material. OpenAI describes public information, partner material and examples provided or generated by users, trainers and researchers. Sources and practices vary by provider.
Source: OpenAI's explanation of model development
A typical text-generating language model builds its reply in small pieces called tokens. A token can be a word, part of a word or punctuation. At each step, it predicts possible next pieces from the context and generates one, then continues. The reply is shaped by learned relationships, instructions and the available context, which can include earlier messages or information from tools. It isn't necessarily looking up a ready-made answer. A request for a warmer response can influence the wording it generates.
Source: Google's introduction to language models
Voice systems can generate speech as well. Some use separate stages to handle speech and text; others process audio more directly. The design varies. A tool-enabled assistant can also search, calculate or act in another service. That is how an interaction can go beyond conversation, as it did in my email demonstration.
Sources: OpenAI on speech model designs in 2024; OpenAI's 2026 voice model release
What the voice can tell us
A natural voice tells us something about the system's ability to produce speech. By itself, it does not establish consciousness or spiritual identity. Those are much larger claims than the observation that a machine answered in a convincing voice.
We can explain how a reply is produced and still have serious questions about souls, spiritual discernment and the place of technology in Christian life.
Trustworthiness needs its own check. A fluent answer can contain a made-up reference or a confident mistake. The warmth or certainty of the delivery doesn't verify the claim.
Source: NIST's Generative AI Profile
A simple check you can make
If you already use a chatbot, choose a short notice that is already public. Ask it to rewrite the notice in a warmer tone while keeping the facts unchanged. Leave out private correspondence and personal details.
Compare the result with the original. Check every date, time and instruction. Notice which changes affect the tone, and whether any change affects the meaning. A friendlier version still needs to be accurate.
This is an optional exercise, not a test of consciousness or a reason to adopt AI. It is a way to practise separating how a response sounds from what it says. You can make the same distinction when reading someone else's AI-generated material.
For a Bible-related answer, go a step further. Open the cited passage, read the surrounding verses and consider how the interpretation was reached. A warm delivery cannot establish that an interpretation is faithful to the text.
Making room for faith and judgement
My successful email demonstration doesn't remove my responsibility for the task. The recipient, wording and permissions matter. So do the consequences if the system gets something wrong. Software providers have responsibilities too, including clear limits and safeguards.
I'm not anti-tech, I'm pro-human. I want technology to improve people's lives and protect humanity and creation. That leaves room to use a tool carefully, limit it or decline it.
Understanding how a response is produced gives us a better starting point for the conversation. We can take questions of faith seriously while being precise about what the evidence supports.
When an AI answer feels personal, what would you check before giving it your trust?