Part 4/14:
Instances of AI's fallibility highlight its ignorance. A hiring algorithm favored men because it learned from biased historical data. An image recognition system misclassified a Husky as a wolf because of snow-covered backgrounds. These errors stem from the AI's repetition of biases present in its training data, effectively functioning as a black box: even its creators often don’t fully understand how decisions are made.
Legal and ethical challenges emerge clearly in scenarios like self-driving cars, where assigning responsibility—manufacturer, owner, or passenger—in accidents remains unresolved.
Ethical AI: Implementation and Oversight
To harness AI ethically, several principles must be adhered to: