AI Now Explains Why You Get Recommendations — But Can It Be Trusted?
AI is deciding why apps suggest what you buy, watch, or date — and it's not always right.
If you’ve ever wondered why Netflix suggests a movie that feels totally off-base, AI is behind the curtain making those guesses — and now it’s trying to explain its own logic. A new research paper tested if AI can reliably pick the best explanation for why apps recommend things, like 'this song is similar to your favorite artist' or 'people who bought this also bought that'.
Researchers had AI generate 18 different explanation styles and then asked 14 different AI models to rate which one made the most sense to a real person. Big AI models (the giants like GPT-4) performed slightly better than smaller ones, but none agreed perfectly with what real humans thought. In fact, AI and humans both missed fake or misleading facts in explanations about 30% of the time.
The team’s advice for companies using AI explanations: keep the instructions short, use bigger AI models, test the explanation styles before rolling them out, and double-check facts — because even AI’s 'judgments' aren’t foolproof.
So while AI can help explain why you’re being recommended something, it’s still not ready to be your personal critic — and it might be making up reasons that sound good but aren’t true.
- AI is now used to explain why apps recommend movies, products, or content — like a built-in translator for algorithmic decisions
- A new study found AI judges only moderately agree with real people and miss fake facts in explanations about 30% of the time
- Bigger AI models perform slightly better at rating explanations, but no AI is reliable enough to replace human judgment yet
Why It Matters
AI explanations shape what you see online — but they’re not always honest, so your choices might be nudged by guesses.