When it comes to high-stakes AI (medical, finance, etc), what would help you trust it the most?