What is Artificial Intelligence (AI)?

AI systems perform tasks once limited to human cognition, but major limitations persist, many of them systemic and high-risk. For example, facial recognition error rates can reach 35% for dark-skinned women (Buolamwini, 2019), while 97% of image classifiers fail under adversarial perturbation (Papernot et al., 2015). Large language models like GPT-4 hallucinate information, posing risks in domains like healthcare and law (OpenAI, 2023; Ji et al., 2023).