Trustworthy AI: From principle to practice