The Epidemiology of Artificial Intelligence
This paper proposes a conceptual framework adapted from environmental epidemiology to study artificial intelligence as a novel determinant of health, distinguishing between ambient and personal AI exposures to better understand their population-level effects and guide future research, equity, and governance.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
The Big Idea: AI is the New "Air We Breathe"
Imagine you are walking through a city. You can't see the air, but it's everywhere. Some of it is fresh, some of it is smoggy, and it affects everyone differently depending on where they live and how long they stay there.
The authors of this paper argue that Artificial Intelligence (AI) has become exactly like that air. It's no longer just a cool gadget you turn on and off; it is now the invisible infrastructure of our daily lives. It decides what news you see, how long you wait in the emergency room, and even what medical advice your doctor gives you.
The problem? We have a whole science called Epidemiology (the study of how diseases spread and affect populations) that knows how to study air pollution, tobacco, and diet. But we have zero framework for studying how this new "AI air" affects our health.
The paper proposes a new way to study AI, borrowing tools from how we study pollution.
1. Two Types of "AI Exposure"
To understand how AI affects us, the authors split it into two categories, using the same logic we use for air pollution:
A. Ambient AI Exposure (The "Smog" You Didn't Choose)
This is the AI that happens to you, whether you want it to or not.
- The Analogy: Imagine you live near a factory. Even if you never go inside the factory, the smoke drifts into your neighborhood. You breathe it in just by living there.
- In Real Life: You might never have asked a chatbot a question. But an AI might have decided your insurance claim is valid, sorted your social media feed so you only see certain health news, or triaged you in a hospital waiting room. You didn't choose this, but it shaped your health outcome.
- Key Point: You can be "exposed" to AI even if you never touch a computer.
B. Personal AI Exposure (The "Smog" You Chose to Breathe)
This is when you actively use AI tools.
- The Analogy: This is like someone choosing to smoke a cigarette or go for a run. It's a voluntary action.
- In Real Life: You ask a chatbot for diet advice, use an AI app to track your mood, or let an AI write your emails.
- Key Point: This varies wildly. Some people use it for fun, some for health, and some rely on it for everything.
The Big Takeaway: Just because you don't use AI personally doesn't mean you aren't affected by it. The "Ambient" layer touches everyone.
2. Why Old Experiments Don't Work
Scientists used to study AI by doing small experiments, like showing a person a chatbot for 10 minutes and asking, "Did this change your mind?"
The authors say this is like studying the effects of a bad diet by only measuring one single meal.
- It tells you what happens in that one moment.
- It doesn't tell you what happens if you eat that way for 10 years.
- It doesn't tell you how the "recipe" (the AI model) changes over time.
AI is constantly updating. The AI you talk to today is different from the one you talk to next month. We need to study it over years, in huge groups of people, just like we study the long-term effects of smoking or pollution.
3. The "Recipe" of AI Exposure
The paper suggests we need to measure AI exposure in four specific ways, not just "Do you use it?"
- Access: Do you have the internet and a device? (Some people are left out entirely).
- Intensity: Do you use it once a year or every hour?
- Purpose: Are you using it to learn math, to feel less lonely, or to check a symptom? (Using it for emotional support is very different from using it for homework).
- Dependency: Do you need it to function, or is it just a tool?
The Data Surprise:
The paper looked at real data from the US and found some interesting patterns:
- Education Gap: People with college degrees use AI much more than those without. This might widen the health gap if AI helps educated people get better health advice while others get left behind.
- The Health Gap: Surprisingly, Black adults reported using AI for health reasons more than White adults. The authors suspect this is because these communities often face barriers to traditional healthcare, so they turn to AI first. This means if AI makes mistakes, these communities might get hurt first.
4. The "Broken Glass" Problem (Why AI is Tricky)
In traditional science, we assume that if I take a pill, it affects me, and it doesn't change what happens to my neighbor.
AI breaks this rule.
- Interference: If I ask an AI a question, the AI learns from my answer and changes what it shows my neighbor next. My action changes their "exposure."
- Moving Target: The AI changes itself every day. It's like trying to study the effects of a drug that rewrites its own chemical formula every time you take a dose.
5. What Should We Do?
The authors are calling for a massive shift in how we research and regulate AI:
- For Scientists: We need to start tracking AI use in big health studies, just like we track smoking or exercise. We need to stop relying only on data from tech companies (which only shows the people who are already using the app) and start surveying regular people.
- For Society: We need to realize that AI is a health issue. If an AI gives bad medical advice, that's a public health crisis, not just a tech glitch.
- For Companies: Just like drug companies must report side effects to the government, AI companies should be required to share data so independent scientists can study the "side effects" of their products on our health.
The Bottom Line
AI is no longer just a tool; it is an environment. It shapes our decisions, our moods, and our access to care.
We are currently flying blind. We know AI is everywhere, but we don't know the long-term health consequences. This paper is a call to action: We need to treat AI like we treat air pollution. We need to measure it, study its long-term effects, and protect the most vulnerable people from its potential harms, while ensuring everyone can benefit from its good parts.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.