How do AI systems typically determine what is considered good or bad during training?

How do AI systems typically determine what is considered good or bad during training?

• A. Hard-coded ethics rules
• B. Random chance
• C. Patterns in the training data reflecting human judgments about good and bad
• D. Only by user feedback after deployment

Answer: (C)

During training, AI systems learn from data that encodes human judgments about what is good or bad. This means the training data includes examples labeled or ranked by humans, or demonstrations of the preferred behavior. The model adjusts its parameters to imitate or optimize those judgments, so it learns the patterns humans use to distinguish good outcomes from bad ones. Because this signal comes from large amounts of labeled or ranked data, the model generalizes those judgments to new situations it hasn’t seen. Hard-coded ethics rules are not the typical basis for training, since it’s hard to list all cases; relying on random chance wouldn’t guide learning toward desirable behavior; and while post-deployment user feedback can refine the model, the primary determinant during training is the human-annotated data that reflects good and bad.

During training, AI systems learn from data that encodes human judgments about what is good or bad. This means the training data includes examples labeled or ranked by humans, or demonstrations of the preferred behavior. The model adjusts its parameters to imitate or optimize those judgments, so it learns the patterns humans use to distinguish good outcomes from bad ones. Because this signal comes from large amounts of labeled or ranked data, the model generalizes those judgments to new situations it hasn’t seen. Hard-coded ethics rules are not the typical basis for training, since it’s hard to list all cases; relying on random chance wouldn’t guide learning toward desirable behavior; and while post-deployment user feedback can refine the model, the primary determinant during training is the human-annotated data that reflects good and bad.