Which of the following is a reason for AI system failures?

Brittleness and embedded bias are significant reasons for AI system failures. Brittleness refers to the lack of flexibility in AI systems; they can perform well under the conditions they were trained on but fail dramatically when faced with new, unforeseen scenarios. This rigidity can lead to critical errors, especially in high-stakes environments where adaptability is necessary.

Embedded bias is equally concerning, as it arises from the data used to train AI systems. If the training data contains biases—whether intentional or unintentional—the AI system can learn and perpetuate these biases in its operations, leading to unfair or prejudiced outcomes. This can cause failures not only in performance but also in ethical standards, two critical aspects for the responsible deployment of AI.

In contrast, while high computational power can enhance the performance of AI systems, it is not a direct reason for their failure. Overactive human involvement may lead to complications or inefficient processes but does not inherently cause failure in AI systems. Similarly, a lack of machine learning algorithms may limit the capabilities of an AI application but does not directly equate to a failure within existing systems. Thus, brittleness and embedded bias are key reasons that can lead to real-world AI system failures, making this the correct choice.