members of Faith PRC in Jenison, Michigan

* A version of this speech was given to the Reformed Free Publishing Association (RFPA), which gathered for its annual meeting at Southwest PRC on the evening of September 25, 2025. The version printed here has been condensed for clarity, but the full speech can be viewed via the RFPA YouTube channel at https://www.youtube. com/watch?v=pbtJO—b_zo.

Digital readers can can access a reading list prepared by the authors by using the link here; print readers will have to wait for this issue to be posted on the RFPA-SB website.

INTRODUCTION

If you have been keeping track of the world of technology over the past several years, you know that artificial intelligence (AI) is everywhere around us. Whether it is through media coverage, day-to-day online work, or conversations with other people, you cannot avoid running into AI. Given the increasingly large impact it is having on our lives, AI is a topic that must be considered by the church and its individual members.

Like many transformative technologies, AI has generated strong opinions both for and against its implementation. Some will tell you that we are approaching an “existential crisis” in which AI will bring an end to our world. Others are lauding it as one of the greatest inventions of all time, a technology as significant as fire to human history. What we need to understand, of course, is that the truth is somewhere in the middle. AI is everywhere—so how should you and I respond?

As believers, our first impulse should always be to ask what God’s Word says about an issue, even something as recent as AI. Although we will not find direct statements about modern technology in the Bible, what we can discern are principles that transcend time and culture. Throughout Scripture, and especially in the book of Proverbs, we find recurring references to the themes of knowledge, understanding, and wisdom. These three terms are not just synonyms. Rather, they form a continuum in the Christian life that can be applied to any technology with which we might engage.

The first thing that we need to do with any new technology is to gain knowledge about it. We should try to get at least a basic idea of how it works, what it does, and some of the concerns that specialists who know the technology well have expressed about it. The calling here is not necessarily to become a specialist, but to know enough to be able to apply principles from God’s Word.

Once we have an informed knowledge of the technology, we can seek to apply biblical understanding to it. Understanding is distinct from knowledge because it seeks to answer the deeper question, “What is God’s perspective?” This means that we are looking for biblical principles to guide our own perspective on the uses—and potential misuses—of the technology.

Knowledge and understanding naturally flow into wisdom, which asks the question, “How should I conduct my life knowing the principles God has laid out in His Word?” Wisdom involves the application of spiritual principles to concrete, real-world questions. One who is wise directs his or her life in such a way that God’s good gifts, including technology, are directed for good rather than evil. In an ultimate sense, wise living based on true knowledge and understanding directs all things to God’s glory.

Our interest here is to apply this threefold biblical approach to a discussion about artificial intelligence. This article and the next will seek to accomplish this through the following goals: (1) provide a framework for thinking about AI as an emerging technology; (2) identify key biblical principles than can deepen our understanding of AI; and (3) explore some of the issues that might be associated with use of AI in different spheres of life.

FOUR KEY QUESTIONS

When considering any new technology, four basic questions need to be asked. For the sake of making these questions easier to remember, we can frame each of them with a single word starting with the letter “P.”

• Purpose: Why was this technology created and what was it intended to do?
• Provenance: Where did this technology come from, and what are the underlying foundations upon which it is built?
• Potential: How can this technology be used, and what is the magnitude of its impact?
• Pitfalls: What are the consequences (both intended and unintended) that this technology will have on us as individuals and as a society?

As you read this article, keep the four P’s in mind. Ask yourself, “What is the purpose behind artificial intelligence? What is its provenance? What kind of potential does it have to shape our society? And finally, what are some of the pitfalls that we need to be aware of?”

KNOWLEDGE: THE EMERGING TECHNOLOGY OF AI

If we want to understand AI, it is important to begin with definitions, clear terminology, and a sense of history regarding how this technology emerged.

A broad, working definition of AI is technology where a machine exhibits characteristics that we typically would attribute to a human. For example, if you see a device that seems to be able to reason, to learn, or to give guidance, you are effectively seeing AI in action. With this definition, it becomes clear that AI is somewhat of an umbrella term which describes a broad category of things. When you hear that AI is part of a certain technology, follow up by diving deeper and seeking to understand the kind of AI in use and the purpose it is serving.

Why does AI seem to be everywhere right now? First, there has been incredible technological progress in the last three years, with great advancements in capabilities. A second reason would be its potential. There is a strong degree of potential impact as AI continues to roll out and advance in education, the workplace, and other areas of life. Third, investment of both time and energy, as well as media coverage, has increased as firms dedicate financial resources to winning the AI race. These are only a few of the reasons that you are hearing a lot about AI in the world right now. It is not just smoke and mirrors and not just a parlor trick, but a substantial technological change.

Let’s develop our knowledge a little further by distinguishing between two types of AI: traditional and generative.

Traditional AI

Traditional AI shows up in technologies like Google Translate, where we now see the ability for translation to be done by a machine when in the past this would have been a human job. AI language recognition powers digital voice assistants like Alexa and Siri, to the point where they understand what you are saying and can respond appropriately. Traditional AI also enables product recommendations, visible when you are shopping online and are offered a list of recommended products for comparison or consideration. In this context, traditional AI looks at your browsing history and any other demographic information it can ascertain about you and then uses it to give you recommendations. Finally, and perhaps the best example of traditional AI, is your social media feed, where the content that is pushed to you is driven by a personalized algorithm that knows your browsing habits.

The word “algorithm” is important as it is the best way to conceptualize how traditional AI works. Think of it as a complex set of mathematical concepts that have been trained with a specific logic to optimize success in a very narrow job. For example, readers may remember the chess-playing computer called “Deep Blue” in the 80s and 90s, which IBM trained to play against world grandmaster Gary Kasparov. It was specifically optimized for playing chess by analyzing years of championship chess matches. A good analogy for traditional AI would be driving through a field hundreds of times and wearing deep grooves into the ground, which eventually become a rough road. Great time, preparation, cost, and knowledge are required to train traditional AI algorithms, but they can be used in highly effective ways.

When we consider these examples of traditional AI, it is clear that this technology did not emerge in just the last three to five years. These technologies came into existence over decades, whether we have realized it or not. As the internet, cell phones, and other digital technologies advanced, AI has been there. It has been taking on new forms and growing in different ways but doing so in a less obtrusive way than in recent years.

Generative AI

If someone has mentioned AI in the last three years, however, they were probably talking about a newer form called generative AI. This includes platforms like ChatGPT, Gemini, Claude, Grok, and many more. All of these are chat-based AI tools with which you can hold conversations, ask questions, or seek specific guidance.

On recent Apple iPhones, you now see Apple Intelligence or text message summaries: both features are examples of a generative AI model integrated into your phone. Or, if you have used Google search in the last few months, you have started to see AI summaries at the top of the search results. This is Google saying, “Yes, I could search, and I could give you results, and you’ll get to those if you keep scrolling, but here’s what you really need to know,” with the summary coming from their AI model, Gemini. Even if you are not trying to use generative AI, it is being integrated into the devices and technologies you likely already use.

So, what is different about generative AI? A key concept for generative AI is the use of a large language model (LLM). While traditional AI is trained to optimize its capabilities for very specific purposes, think of LLMs as broadly capable generalists. These models are trained on massive data sets that have been curated from the internet: and not just text from websites, but also images and videos. Generative AI models can generate an amazing breadth of content because of the huge data sets that have been used to train them. In some cases, the makers of these LLMs are now being sued for the things their models have incorporated into their set of knowledge, such as copyrighted news and magazine articles.

The advent of LLMs is why there have been such significant advances in the power of AI over the last few years. If you have experimented with one of these tools, you have probably been surprised to see it give a pretty good response to a question you asked. It may seem to know things that you would not have expected an AI tool to know. Generative AI platforms are now able to pull things off the internet, make predictions, and be right more often than wrong. And these models have advanced to the point now where they may include the ability to appear to reason and think. Reasoning models do not just immediately try to answer complex questions or requests, but “think” about how to approach and then “reason” its way through to pulling together the right information.

The distinction between traditional and generative AI matters because many current AI conversations focus on the latter category. That is where recent advancements have been, and that is where a lot of additional investment will be made as we look ahead to coming years.

History

Looking back at the history of AI accentuates just how rapid AI progress has been. The term “artificial intelligence” first came to be in the 1950s. This was a time when futurists were looking ahead and thinking, “What if I could interface with a computer? What if that computer could talk back?” Computers were a fairly new development, however, and the power of computers was limited, so this vision was restricted to futurists’ dreams. This early era, sometimes called the “AI winter,” lasted all the way up to the 1990s.

As the internet launched and digital technology advanced in the first decade of the twenty-first century, a period of ubiquitous high-speed internet began. Smartphones emerged, and the online ecosystem began to develop without the need to be hardwired into a physical network. Together, these technologies paved the way for the traditional AI applications that were mentioned earlier. Further development of “cloud technologies,” which store massive volumes of information online, began to power the internet and make broad training datasets available for the next phase of AI evolution.

Then, in 2017, Google pioneered a foundational technology approach called the “Transformer,” which revolutionized how AI models were built and trained. This paved the way for others to use this approach— OpenAI for example—in their products. The 2020– 2025 timeframe is when the foundations for current AI tools were built, rolled out, and have since rapidly advanced. Since November 2022, when ChatGPT first launched, there have been incredible leaps in technology, with many new companies and models introducing and expanding upon AI capabilities.

There is obviously a wide variety of projections for the future. There are some who say, “This is the future. Soon, we may not even need to work; AI will handle everything for us.” Others view it as all hype. The truth, most likely, falls in the middle. What is certain is that we are quickly arriving at a new phase after the period of rapid advancement and adoption in which things like regulation, legislation, accountability, and governance will start to apply more to AI tools.

How Generative AI Works

Interacting with AI starts with input from a user. Typically, as with ChatGPT, that is through typing. If you are using a tool like Siri or Alexa, your input might come through voice, but it all translates to text in terms of how the input reaches the AI model.

At bottom, AI’s job is prediction. Predicting the best response to your specific question. In some cases, the AI model may immediately be able to respond to your question because it is content the model is confident in and has been trained to know. For example, a factual question—“How tall is Mount Sinai?” At other times, a question you ask demands a more sophisticated, real-time answer that goes beyond simple facts. For example, “What’s the current stock price for Microsoft?” AI now knows to search the open internet for that question. It does not know the answer itself, but it does know how to connect you to the answer by curating the correct information from the internet in real time. You might also ask AI to work more strategically, perhaps to create a plan involving a multi-step process, which will require AI to take more of a reasoning-based approach. It will “decide” on the right strategy to take to perform the specific task you have asked.

Your interaction with AI is heavily shaped by context, which is something like digital footprints. When you engage with an online platform like a search engine, it “learns” things about you. These platforms have access to your past conversations and demographic details about who you are. You might intentionally include added context about your role in a given task, such as that you may be working for a nonprofit that supports a school’s fundraising efforts, which in turn will allow AI to give better responses tuned to that context. Based on your specific request and all of the context available, AI then responds to you, the user, and stores the conversation history for future use and training. When you have had a conversation with an AI assistant, it is storing that conversation history so that your future interactions can be improved.

It is important to realize that something called “mediation” happens in the background before AI responds to your input. AI tools have guardrails that are programmed into them to control what they can and cannot do. For example, AI tools will not tell users how to build a bomb, nor will they answer questions that are recognized as prejudiced in a way that the makers of these tools do not support. There is not a direct line between what you ask and the LLM; a level of control has been intentionally built into the platform to mediate the process. Mediation can also include the level of personability that an AI tool displays when answering your question. Will it call you by your name? Will it be kind and warm to you, or will it be more factual and sterile in its response? Be aware that these are all features that can be tuned by the makers of these technologies, often to best fit their goals and worldview.

Reflections

With this brief overview of how AI works, let’s pause for a couple of reflection questions.

First, does AI truly “know” things? The important thing to realize is that these LLMs are prediction engines. An AI does not “know” things. It is a very sophisticated prediction engine. “I predict the answer is this. I predict the answer is that.” It is looking across the datasets it has been trained on and saying, “I predict this is the answer,” which answer is then put forward as truth. Or perhaps better put, “truth” as determined by an algorithm. In contrast to Google, where you must review results and synthesize them yourself, AI tools are happy to handle that synthesis and direct you to their recommended answer with confidence.

It is also important to ask, “Will AI tell you when it does not know something?” These tools are optimized to keep you using them, and for that purpose, AI tools are trained to project confidence. It is only in the rarest of circumstances that AI will admit that it does not know something. You have likely seen humorous examples of AI getting very basic math problems and questions wrong, such as “How many R’s are in the word strawberry?”—a question to which ChatGPT’s answer for months was a firm “two.” When you interact with AI today, what you will see most frequently is bold proclamations that, “This is the answer.” If you confront it and say, “That’s not the answer, because I happen to know,” it will apologize profusely and self-correct. The takeaway is that it is on the user of these tools to hold AI accountable. AI will not typically be transparent with you in terms of what it does and does not “know” unless very directly asked.

Lastly, an important question is, “Are AI responses neutral?” We have talked about how mediation happens and how the makers of these platforms control how they respond. For example, consider a recent article from The Gospel Coalition detailing a project in which Christian theologians were asked to judge the responses of seven different AI chatbots to basic questions about the Christian faith.1 The article illustrates the rather alarming degree of influence that platform developers have in shaping responses about questions of faith. AI tools ought to be used with the assumption that neutrality is not built in, but that the creator’s particular leanings or agendas will be influencing your experience.

CONCLUSION

This first article has provided a broad overview of AI as a technology starting with definitions, moving into a contrast between types of AI, and then diving a touch deeper into how the technologies work. It also outlined a framework that can be used to evaluate technologies— purpose, provenance, potential, and pitfalls. As we close out this section, we hope that the knowledge you have just absorbed gives you a more informed look at the question of “How do I think about AI tools?” In the next article, we will move on to some guiding principles for understanding and using AI.

1 Mike Graham / AI Christian Benchmark, “Evaluating 7 Top LLMs for Theological Reliability,” The Gospel Coalition (September 2025), https://www.thegospelcoalition.org/ai-christian-benchmark.