A healthcare startup is developing a generative AI system to assist doctors in diagnosing rare diseases. According to Google's AI Principles, what is the MOST important requirement before deployment?
Trap 1: The model must achieve at least 99% accuracy on a held-out test set
Accuracy alone is insufficient; safety, fairness, and accountability are also required by Google's AI Principles.
Trap 2: The model must be trained on the most recent medical literature
While up-to-date data is important, safety and human oversight are the primary concerns for high-stakes decisions.
Trap 3: The startup must publish the model's architecture in a…
Publication is not a requirement of Google's AI Principles; the focus is on safety, accountability, and societal benefit.
- A
The model must achieve at least 99% accuracy on a held-out test set
Why wrong: Accuracy alone is insufficient; safety, fairness, and accountability are also required by Google's AI Principles.
- B
The model must be trained on the most recent medical literature
Why wrong: While up-to-date data is important, safety and human oversight are the primary concerns for high-stakes decisions.
- C
The startup must publish the model's architecture in a peer-reviewed journal
Why wrong: Publication is not a requirement of Google's AI Principles; the focus is on safety, accountability, and societal benefit.
- D
The system must include a mechanism for human review of all diagnostic suggestions
For high-stakes AI decisions like medical diagnoses, human oversight is essential to ensure accountability and patient safety.