Addiction is a difficult problem to understand through any single lens. What happens in a person’s brain matters, but so do family relationships, organizations, neighborhoods, treatment systems, and policy. Then there is another question beneath the research: What do we believe recovery ought to look like?
That question helps explain why secular and Christian approaches to addiction treatment can talk past one another. They may recognize some of the same patterns while using different language and aiming toward different ends. A productive comparison needs both a systems lens, to trace how the parts interact, and a worldview lens, to examine the values behind proposed solutions.
AI makes it possible to explore those connections much faster. The goal is not to ask a chatbot for the final answer. It is to use AI within repeated action research cycles: frame a question, test it, examine the result, and ask a better question. Here is how I use that process.
Key Takeaways
- Use short action research cycles to turn an initial question into a stronger one.
- Make worldview assumptions explicit when combining systems models and comparing proposed solutions.
- Distinguish a model’s description of addiction from its recommendations about recovery.
- Treat AI outputs as material to examine, verify, revise, or discard.
Table of Contents
- Make iteration speed the priority
- Design prompts with a worldview and compatible theories
- How the addiction case developed through research cycles
- Turn technical models into understandable questions
- Prompts to try in your own systems research
- Practice with epistemological humility
Make iteration speed the priority
Research often begins with genuine curiosity. A course topic catches your attention, or two ideas seem as though they might fit together. Some of those ideas lead nowhere. Others reveal a question worth pursuing. You cannot always tell which is which before doing the work.
That is why I prioritize iteration speed. Instead of spending months perfecting one question before testing it, I develop a workable question, run a research cycle, and learn from what comes back. This combines the reflective rhythm of action research with the testing mindset of lean startup.
The cycle has four practical stages:
- Frame the problem. State what you want to learn, compare, or demonstrate, along with any initial hypothesis.
- Test the question. Run a deep research report when appropriate, using a prompt specific enough to produce a useful first result.
- Assess the output. Skim or read the report, examine its sources and reasoning, and decide whether the prompt needs revision.
- Synthesize and reflect. Identify what is missing, what surprised you, and which question should come next.
For a substantial report, I often add the material to a Gemini notebook and generate an audio overview, sometimes alongside a video overview, slides, or an infographic. I may skim the written report but listen to the entire audio overview. Hearing an explanation can surface connections or analogies that I missed on the page. Those artifacts are useful for learning; they are not substitutes for checking the underlying work.
A report that is too technical or wanders into an irrelevant domain is not a disaster. It tells me how to revise the prompt. Sometimes the right move is to discard the report entirely.
Design prompts with a worldview and compatible theories
Prompts do more than identify a topic. They influence which concepts the AI emphasizes and what kind of explanation it produces. For worldview analysis, I think of prompt design as combining three elements:
Worldview × theory or model × theory or model
In the addiction case, I combined a worldview perspective with a socioecological model and system dynamics. The socioecological model helped me ask about different levels of a problem. System dynamics helped me ask how relationships, feedback loops, and changes over time connect those levels.
The combination matters. Theories should illuminate one another, not simply make a prompt sound sophisticated. Socioecological analysis and system dynamics both address interconnected systems, so they offer a sensible pairing. An unrelated theory would be less likely to clarify the question.
It also matters whether the worldview is explicit. A prompt that asks only for a scientific account of addiction may produce a different kind of answer from one that asks for a Christian interpretation. Neither result should be mistaken for a neutral view from nowhere. As the researcher, you need to notice which assumptions you introduced and which ones appeared in the output.
How the addiction case developed through research cycles
My starting question was how to understand the differences between secular and Christian approaches to addiction using system dynamics and worldview analysis. I did not try to settle that in one enormous prompt. I built a sequence of reports, using each result to shape the next question.
1. Map addiction across the system
First, I asked for an integrated system dynamics and socioecological model of addiction at the micro, meso, and macro levels. That meant looking from individual processes and close relationships outward toward institutions and larger social forces.
The aim was a comprehensive account of how the system works, not yet a comparison of treatment philosophies. It gave me a model to inspect: Which relationships were included? Where were the feedback loops? What did the model explain well, and what did it leave out?
2. Develop a Christian interpretation
Next, I attached the first model and asked for a comparative Christian interpretation. I named concepts I wanted considered, including idolatry, spiritual warfare, and principalities and powers. I also specified the intended audience: leaders of evangelical gospel rescue missions.
That audience instruction was important. I wanted material leaders could actually use, rather than an unnecessarily academic treatment or an interpretation disconnected from orthodox Christian theology.
The question was not whether Christian language and scientific language are interchangeable. It was whether they sometimes describe related aspects of the same phenomenon, and where they make genuinely different claims. For example, someone might discuss addiction in terms of behavioral patterns while a Christian account also asks about disordered loves and spiritual life.
3. Separate what is from what ought to be
The next question was the turning point. A model may be excellent at describing what is: patterns of addiction, interacting causes, and the effects of an intervention. But its recommendations about what ought to be depend on goals and values.
I used Hume’s is and ought distinction to probe that gap. If a report describes a problem accurately, does its preferred solution follow from the description? Or have moral, political, or theological assumptions entered the argument without being acknowledged?
One early attempt at combining the perspectives gave me what I considered weak integration. Christian concepts appeared, but the overall framework recast them in largely secular terms. So I revised the question. I asked for a third perspective that would take scientific descriptions seriously while giving greater weight to a biblical understanding when making recommendations about ultimate ends.
That is what I mean here by strong Christian integration. It does not require the Bible to overwrite every empirical observation. It does require clarity about which worldview is guiding judgments about what people and communities should pursue. Science can help describe an outcome. By itself, it cannot settle every question about the good life.
4. Test the critique against interventions and evidence
After developing the philosophical critique, I wanted to examine actual interventions. I asked about Housing First, harm reduction, and medications for opioid use disorder, or MOUD, through a system dynamics lens. In particular, I wanted to investigate stocks, flows, and whether an intervention functions as a buffer or contributes to longer-term change.
That distinction raises useful questions. Does a program reduce immediate danger? Does it change the conditions that keep producing the problem? What happens while someone receives help, and what happens afterward? A system can show improvement on one measure while other underlying needs remain.
Those questions should not be turned into a blanket verdict about treatment. My AI-assisted analysis led me to challenge whether certain approaches address root causes, but a generated report is not, on its own, proof of a universal outcome. In particular, claims about MOUD need to be weighed against clinical evidence and the specific outcome being measured. The National Institute on Drug Abuse’s overview of medications for opioid use disorder provides important context when evaluating such claims.
A good action research cycle does not stop when a theoretical critique sounds compelling. It asks what the empirical evidence supports, where the comparison is fair, and what remains uncertain.
Turn technical models into understandable questions
A system dynamics report can become highly technical. Its diagrams may represent relationships that would require substantial mathematics to model formally. One advantage of AI is that it can help translate a complicated model into ordinary language so that you can think with it.
That does not mean you should assume every generated diagram or calculation is correct. It means technical output can become a starting point for better questions. An audio explanation might help you notice a feedback loop, a possible step change, or a relationship between two parts of the system. You can then return to the report and investigate whether that connection holds up.
In this project, I brought the useful reports together into a broader resource on how worldviews shape understandings of addiction’s causes and possible solutions. The finished resource mattered, but so did the questions that emerged while making it.
Prompts to try in your own systems research
You do not need to begin with addiction. Choose a topic where systems thinking has a meaningful application, such as organizational growth, family dynamics, mental health, or a particular social program. Then give the AI a focused job.
- Start with a literature review: Ask for work by Christian thinkers at the intersection of systems thinking, system dynamics, and one specific topic. Check the cited sources.
- Map the wider ecosystem: Ask what people, organizations, incentives, and conditions interact around an issue.
- Explore feedback: Request a causal loop diagram that identifies important relationships, then review whether the direction of each relationship makes sense.
- Investigate archetypes: Ask whether patterns such as limits to success, growth and underinvestment, shifting the burden, or fixes that fail may apply.
- Model an organization or program: For a rescue mission, include intake sources, completion rates, outcomes, and measures of holistic health, recovery, and spiritual capital.
- Rework a logic model: Take an existing linear program logic model and ask what feedback loops and delayed effects it may be missing.
Be especially careful when naming a systems archetype. If you ask an AI to show how a program is a “fix that fails,” you have already pushed it toward that conclusion. Investigate the program and the evidence first. Diagram the archetype only if the pattern fits. And if you generate a visual, check its labels and spelling as well as its logic.
Practice with epistemological humility
Fluency with AI-assisted research is like fluency with an instrument. You do not get there by planning the perfect performance before you touch the keys. You get there by practicing, hearing what works, noticing what does not, and trying again.
I have run many reports that I threw away. Even for the addiction resource, I explored far more reports than I ultimately used. That is part of the method, not evidence that the method failed. A quick, poor result can teach you something about your question. A promising result still needs scrutiny.
Epistemological humility means remembering the limits of what we know. Researchers have blind spots. AI systems have flaws. Audiences bring their own assumptions. None of us has a monopoly on truth. That calls for source checking, fair treatment of competing views, openness to correction, and grace when an early attempt is imperfect.
Use this approach throughout a course or doctoral project, not only when you need a publishable paper. Follow the questions that make you curious. Run the cycle, keep the insights that stand up to examination, and let the discarded attempts help you ask something better. The game changer is not one brilliant prompt. It is the ability to learn and improve, repeatedly, at speed.