Bias in AI: What's Really Going On?

Ever thought that AI systems could be biased? Well, they can. Even if they pass tests that check for obvious bias, they can still have hidden biases. These biases are like the ones humans have, even if they claim to be fair. It's tricky to spot these hidden biases. As AI systems get more secretive, it's hard to see what's really going on inside them. Plus, these biases only matter if they affect the decisions these systems make. To tackle this problem, researchers created two new ways to measure these hidden biases. The first is called the LLM Word Association Test. It's like a game where the AI has to quickly link words together. The second is the LLM Relative Decision Test. This one checks if the AI treats different groups differently when making choices. These tests are based on how psychologists study human biases. The Word Association Test is like a game humans play to see what words they link together. The Decision Test looks at how people judge two things side by side, not just one thing at a time. This can show hidden biases. Using these tests, researchers found that AI systems have biases that mirror those in society. They looked at eight different AI systems and found biases in areas like race, gender, religion, and health. These biases were in things like linking race to crime, gender to science, and age to negative things. These tests are important because they show that even AI systems that seem fair can have hidden biases. This is a big deal because it means we need to be careful about how we use these systems. We need to make sure they're not making decisions based on these hidden biases. It's important to note that these tests are based on a lot of research in psychology. They show that even AI systems can have biases that affect how they make decisions. This is something we need to be aware of as we use these systems more and more.