Author/Source: Adversarial examples are tricky inputs designed to fool an AI. Confidence scores are how sure an AI is about its own decision. See the full link here
Takeaway
This article talks about how we decide to trust artificial intelligence (AI) systems, especially when they make decisions that could be risky. It questions whether current approval methods for AI are good enough and suggests we need better ways to check if AI is safe and reliable.
Technical Subject Understandability
Intermediate
Analogy/Comparison
Deciding whether to trust an AI is like deciding whether to trust a new doctor. You want to know if they are qualified and if their methods are safe before you let them make important decisions about your health.
Why It Matters
If we don’t have good ways to check AI, we might end up trusting systems that make mistakes or cause harm. For example, an AI used in a self-driving car needs to be thoroughly tested to make sure it won’t cause accidents.
Related Terms
Adversarial examples, confidence scores
Tech Teacher 
Leave a comment