Introduction to AI in 2025

Given the topic of these essays, I thought it would be wise to provide an introduction to AI in 2025 as a guide for some of the ideas that will be discussed going forward. Typically, these essays will start and end with Scripture, and be theological throughout, but this one will break that mold just to explain some basic concepts. Rest assured, this will be the most techie post written here, as that is not the purpose of this series.

The state of Artificial Intelligence (AI) in 2025 is dominated by something called Generative AI. So that is where we will begin. Services like OpenAI’s ChatGPT, Google’s Gemini, xAI’s Grok, and Anthropic’s Claude are all within this category. These services, and the models that back them, are designed to generate content. The leading services all have text and image generation capabilities as of the time of this writing, with video services well on the way. These same services also now incorporate web searching capabilities, to help ground their generated text in factual information. Each of the products does this slightly differently: Grok does this automatically and links all of its sources at the top of its content, Gemini provides a “double check” option that will verify the information provided, and ChatGPT provides a button to allow the model to search the web.

What exactly does it mean to “generate” content? How is this done? As it turns out, human language is predictable. If I give you this fill-in-the-blank sentence, “Why did the chicken cross the ____”, your prediction is going to likely be “road”. This is roughly how text generative AI works. These models have been trained on trillions of words of written language, and operate as “next word predictors”. These predictive operations can be chained together and looks something like this:

Why did the chicken cross the ____ -> road?
Why did the chicken cross the road? ____ -> To
Why did the chicken cross the road? To ____ -> get

… and so on.

Image generation models are slightly different, but operate on a similar principle. They consume millions of categorized images, and are then able to coalesce an image based on the user’s request. In practice, this means that some very controversial images can be generated, which I obviously will not be sharing here. Video models then do something much like this, but on a larger scale generating individual images in a sequence to form a video. Audio generation models, such as those provided by ElevenLabs are also similar in function, and can be custom trained to mimic a specific individual’s voice and speak any text.

A major piece of the generative AI conversation is “prompting”. Simply put, a prompt is the set of instructions that the user gives to the service to guide it as it generates content. How the user builds the prompt can greatly affect the quality of the generated text from the service, and an entire cottage industry has sprouted around “prompt engineering”. I am not going to spend any time discussing this, but there are lengthy videos and articles all over the internet that provide guidance on this if it is of interest to you.

Another big term currently being thrown around, and one that will be particularly relevant to future discussion here, is “agent”. For our simplified purposes, an agent is simply an autonomous system that performs a task. This is one step past just sending and receiving messages from the AI service. An Agent receives a message, evaluates what the intent of the message is, and then it may repeatedly prompt itself to accomplish the overall task while also interacting with other tools, like Google Search, Wikipedia, or even a business’s internal documents. When talking heads discuss AI replacing jobs, this is what they’re pointing at.

The future of AI in the world looks very much like autonomous AI systems interacting on their own performing work that would have been previously done by humans. Tools like Cursor do this for writing code, and are already quite good. Adobe has natively incorporated AI into their products, providing generative image tooling that is frankly astounding. Google Gemini can produce multi-page research reports in minutes that cite sources, and provide in-depth analysis on topics that would have taken a researcher several hours to do. OpenAI’s o1 (and other models) are capable of approximating human reasoning with what is called “chain-of-thought”. These models have passed pretty much all of the professional certification and specialized graduate school entrance exams. Google’s services are capable of viewing your screen and helping you troubleshoot issues, as well as taking direct control and performing tasks when requested. All of these models are capable of analyzing images, extracting information, and taking action. I have personally used Google’s service to troubleshoot my broken dryer, among other tasks, via image analysis and even voice conversation with the service.

The last thing I want to do is scare anyone. AI today is powerful, and will change the way we interact with technology. It will also seep its way into the ways we interact with each other. Already, my phone makes response recommendations in my texts and emails and this is only going to increase. There is significant upside potential to AI, but like any powerful tool, there is also significant downside potential. This necessitates the need for thought around how and where we utilize these tools in the process of building Christ’s Church and serving Him here on earth.

I hope this article has been helpful. If you have additional questions, please leave comments and I will address them. This is obviously not intended as a full explanation of these concepts, but hopefully it has been useful and will lay a groundwork for some of the terminology in the essays to come.