Clickbait: Normative value is algorithmically complex: there is no simple way to describe the thing we should want AIs to want.
The 'Complexity of Value' proposition is true if, relative to viable real-world Methodologies for AI development, there isn't any knowable way to directly specify by hand the AI's [object-level preferences] as a structure of low complexity, and let the result of running that AI being the Achievement of Enough of the possible value.