The role of Artificial Intelligence in interventions to reduce alcohol consumption and smoking: A systematic review
This systematic review of randomized controlled trials in high-income countries found no strong evidence that artificial intelligence-assisted interventions effectively reduce alcohol consumption or smoking, largely due to mixed results, the inability to isolate AI's specific impact within multi-component interventions, and a high risk of bias across the included studies.
Original paper licensed under CC BY 4.0 (https://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
Imagine you are trying to quit smoking or cut back on drinking. You've heard about a new kind of "digital coach" powered by Artificial Intelligence (AI) that promises to help you succeed. This paper is like a massive detective report that went out to find all the scientific studies testing these AI coaches to see if they actually work.
Here is the breakdown of what the researchers found, using simple analogies:
The Big Picture: A Search for a "Magic Wand"
The researchers looked for studies in wealthy countries (like the US, UK, Germany, etc.) where people tried to use AI to stop smoking or drinking. They wanted to know: Does adding AI to a health app actually make it better than just using a regular app or getting no help at all?
They found 16 studies about drinking and 12 studies about smoking. In total, this involved nearly 73,000 people.
The "AI" They Found Wasn't the "Smart" AI You Might Think
When we hear "AI" today, we often think of super-smart chatbots like the ones that write essays or hold deep conversations (called Large Language Models).
However, the "AI" in these studies was more like a very advanced "Choose Your Own Adventure" book or a digital vending machine.
- Chatbots: These weren't thinking machines. They were mostly "rules-based." Imagine a flowchart: If you say "I'm stressed," the computer shows you a pre-written calming tip. It didn't actually learn or understand you; it just followed a strict script.
- Recommender Systems: These were like a music playlist algorithm. If you liked a certain song, it suggested another one. In these studies, if you clicked on a specific message, the system suggested a similar one next time.
Crucially, the researchers found zero studies using the modern, "super-smart" AI (like the ones you talk to right now) to help people quit.
The Verdict: Mixed Results and Broken Magnifying Glasses
The researchers tried to combine all the results to get a clear answer, but it was like trying to mix apples, oranges, and bananas into a single fruit salad. The studies were all so different (different apps, different people, different ways of measuring success) that they couldn't do a math calculation to get a single number.
Instead, they looked at the stories of each study:
- For Drinking: Some studies said the AI apps helped students drink less, while others said they did nothing. One study found a website that tailored its content (like a personal trainer for your browser) worked well, but the chatbot apps were hit-or-miss.
- For Smoking: The results were all over the place. Some studies showed the AI chatbots helped people quit, while others showed they were no better than a standard text message service or a regular app without AI.
The "Broken Magnifying Glass" Problem:
The biggest issue wasn't just that the results were mixed; it was that the studies themselves had major flaws. The researchers used a "risk of bias" tool (a quality check) and found that almost every single study was "high risk."
- The Analogy: Imagine trying to judge a race where half the runners quit halfway through, and the finish line judges are guessing who crossed first because they weren't wearing glasses. Because so many people dropped out of the studies and the data was self-reported (people just saying "I quit" without proof), the researchers couldn't trust the results completely.
What the Experts (Stakeholders) Said
The researchers also talked to public health experts and community leaders. These experts had some concerns:
- The "Co-Production" Gap: Many apps were built by tech people without asking the actual users (the drinkers or smokers) what they wanted. It's like a chef making a meal without asking the diner if they are allergic to nuts.
- The "Readiness" Gap: Just because an app exists doesn't mean people are ready to use it. Some people might need to learn how to use the tech first.
- The "Fairness" Gap: There was a worry that these fancy AI tools might accidentally help rich, tech-savvy people quit while leaving disadvantaged groups behind, making health inequalities worse.
The Bottom Line
There is no strong proof yet that AI helps people drink less or smoke less.
The paper concludes that while AI might be useful in the future, we currently don't have enough good evidence to say it works. The studies we have are too messy, too flawed, and too different from each other to draw a firm conclusion. Furthermore, we haven't even tested the "super-smart" AI of today yet.
In short: The digital coach is still in training, and we can't say for sure if it's ready to help you quit just yet.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.