Found a research on GitHub about analyzing AI articles content on the BBC. Researchers checked thousands of articles for patterns that indicate AI usage.

I thought that it might me a good idea to build kind of a browser extension to flags potential AI content as you browse. Like you don't upload conttent to detector, but you open a page and a detector checks it in real time, without sending data to external servers.

But I'm not sure, pattern matching be sufficient or do you need ML models running locally? And probably false positive rates would be awful, right?

Has anyone tried building or using something like this? Is this even worth prototyping or if the accuracy would be too unreliable

Here's the GitHub, I mentioned: https://github.com/isfakeai/bbc-ai-content-analysis

They also published a detailed report breaking down their methodology and findings.

rproffitt commented: Good idea. Keep going. +0

Recommended Answers

All 7 Replies

Interesting idea, but I agree the false positives would be a big challenge with pure pattern matching. You’d almost certainly need an ML model running locally for anything reasonably accurate, and even then results would be probabilistic. I think it’s worth prototyping as an experiment, but relying on it as a “detector” for users might be tricky.

I think that before something like this would be useful, what the world really needs is an accurate way of detecting AI-generated content that doesn't have an overwhelming amount of false positives or false negatives.

For now, I wouldn't really mind uploading to a content AI detector, as long as the detector was accurate.

If it was me, and I found some way of cracking accuracy, then I would create an API and charge a micropayment per request. I'd make a killing! I'd also create a browser tool based off of the API, and perhaps find a paying sponsor for it.

The first step is AI contect detection that actually works. Have that, and you can make bank. :)

My default position is to think most of media is either outright AI pulp fiction, or at the very least tainted with "suggestions".

For this local machines learning models can help but they are limited without access to large database

A real-time AI content detector browser extension is worth prototyping but may struggle with accuracy due to false positives and negatives. Pattern matching alone likely won't be enough and local machine learning models have limitations.

commented: Robot 1: " +0

If I were building this, I’d probably think about the user side first, not just detection accuracy. Like, even if it’s “mostly right,” what happens when it flags legit content? That could annoy people fast. I’d maybe frame it more as a “confidence signal” instead of a hard yes/no.

Be a part of the DaniWeb community

We're a friendly, industry-focused community of developers, IT pros, digital marketers, and technology enthusiasts meeting, networking, learning, and sharing knowledge.