I would love some extra sources for this. A quick reading of the article makes me understand that the researchers have a lot of confidence in the correctness of the AI gradings of policies, which I have my doubts about.

The article worries about the fact that the energy policies score worse than expected, but this assumes that these scores are valid in the first place. There is lots of interesting research happening in trying to use AI for varying problems, but in each case results go from bad to promising, but never outright certainty. The fact that this article almost immediately gives the vibe that the AI results are trustworthy and a valid and desired alternative to the slow human analysis really rubs me the wrong way.