Given the failure of US-based AI forecasters to predict Chinese AI decisions, "mind-reading" should be downgraded compared with analysis of stated and revealed preferences. Translation and literature search are easier than ever, so there are few excuses to neglect such analysis.