If An Internal Model Discovers A New AI Architecture 1,000x Better Than Current Architectures, But No Human Can Understand It, Should We Refuse To Build It?

Terrence Tao's recent post stating that it would be irresponsible for cancer patients to accept an AI discovered cancer cure (even if it works) if no humans can understand it made me realize that interpretability and the need to understand a models thought process will become impractical and impossible as we arrive near true ASI. Imagine an AI discovers a completely new architecture for building AI models. The resulting models (if built on the new architecture), are 1,000x more capable, dramatically faster, and cost a fraction as much to train and run. However, the reasoning that led to the model discovering this new architecture is impossible for humans to understand, just as it would be impossible for an Ape to understand how a transformer works. And, imagine that internal model cannot simplify/explain the reasoning mechanism for this new architecture, just as it would be impossible for any of you to explain to an Ape how the transformer works, even if you dumb it down as much as possible and use countless of examples/analogies. Should Humanity refuse to build architectures it cannot understand because of interpretability/safety fears, even if these new models would lead to ever accelerating intelligence, cures of all diseases, unlimited energy & abundance?