I have problems with the EU High-Level Expert Group on AI's Policy Recommendations, and have a paper on my views forthcoming in @EurJRR. The preprint is now up on @lawarxiv at osf.io/preprints/lawa…. Here's a summary (short thread): 
I argue (with the help of a Discworld metaphor, cc @Jackstilgoe) that we don't need policy interventions to help us know when to use AI as much as when *not* to use it (and to frame the problem properly)
Where are the 'low level experts'? (thanks, @sedyst) '720 more professorial chairs in applied ethics' are recommended (!): but where will skills to organise to affect business models like #TechWontBuildIt and #GoogleWalkout come from? And where are affected communities?
The HLEG roundly ignore the role of infrastructure: not just compute but mostly data collection, experimentation, tracking, in AI, fetishising the models rather than the business models. They're intertwined. Separating them is just what lobbyists want you to do.
The HLEG do not recommend increasing regulatory enforcement capacity. We have laws. They're unenforced. This would have been an incredibly simple recommendation, and its absence says it all really.
There is a power outage throughout the report. As soon as power is even approached, the words weaken to 'consider' instead of 'develop' or 'adopt'. The surreal experience of looking for power in the EU HLEG AI recommendations would be amusing if it wasn't so tragic.
The recent German #DatenEthikKommission report stands in stark contrast to the EU HLEG AI report. Read them side-by-side and see for yourself. @mireillemoret has helpefully threaded key pieces here.
