The use of AI to detect alcohol with Samatha Salim and Benjamin RiordanThe use of AI to detect alcohol with Samatha Salim and Benjamin Riordan
Addiction Audio
Researchers discuss how different AI models detect alcohol in video images and why this matters for public health, film ratings and online policies. The conversation highlights strong model performance, practical applications, and how AI can support rather than replace human researchers.
21:04•1 May 2026
Spotting Booze on Screen: AI, Alcohol Images and Public Health
Episode Overview
- AI models can reliably detect alcohol in video frames, with the LAVA model reaching around 95% accuracy on complex movie images.
- Exposure to alcohol-related content in media has been linked in research to later alcohol use, making accurate measurement of that exposure important.
- Current film and media classification systems often overlook alcohol, even in content rated suitable for children, raising concerns about age-appropriate viewing.
- AI can drastically cut the time needed to analyse huge volumes of video and social media content, freeing researchers to focus on interpretation rather than manual coding.
- These AI approaches could be extended to other topics like vaping, smoking, gambling and broader unhealthy commodities, supporting both policy enforcement and parental control.
“You don’t need to spend your time just watching the media… you just use these AI models, make it zero-shot learning mode, and the system will do the work.”
What drives someone to seek a life without alcohol? For many people, one big factor is how often they see booze on screens – and this episode of Addiction Audio takes that idea seriously, but with a techy twist. Host Dr Tsen Vei Lim chats with PhD candidate Samatha Pararath Salim and research fellow Dr Benjamin Riordan from the Centre for Alcohol Policy Research at La Trobe University in Australia. Their focus?
Comparing different artificial intelligence models that can spot alcohol in video images. Samatha breaks down AI in plain language, explaining it as “a computer system in which we… provide intelligence to the system so that it can make decisions or recognise patterns,” linking it to everyday tools like Spotify and online shopping. From there, the conversation turns to why anyone would care about bottles and beer glasses in films and social media.
Ben points to research showing that “exposure to almost any alcohol-related content might, say, lead to subsequent alcohol use,” especially when young people are constantly online. The trio talk through three AI models, including a custom-built system called ABIDLA and two larger, general-purpose models (LAVA and CLIP). The surprise star is LAVA, a big pre-trained model that wasn’t even specifically taught to recognise alcohol yet still reached around 95% accuracy on complex movie frames.
As Samatha puts it, “we just prompt the model to do something. That’s it.” You’ll hear how these tools could help rating boards classify films more consistently, help platforms check if alcohol ads are slipping through to young viewers, and even support parents who’d rather keep “unhealthy commodities” off family movie nights. The team stresses that AI isn’t there to replace humans, but to save researchers from watching thousands of films frame by frame.
If you’re curious about how tech can support healthier alcohol policies without losing the human touch, this one’s worth your time.
Do you want to link to this podcast?
Get the buttons here!
More From This Show
The latest episodes from the same podcast.
Related Episodes
Similar episodes from other shows in the catalogue.