The Parliamentary Research Service of the EU has issued a briefing last month. I’ve written previously on the European Union recommendations for trustworthy Artificial Intelligence. This briefing starts to consider how to implement them. A few things are of note here. The first is the recognition that EU ethics guidelines are non-binding and there is no regulatory oversight to implement them in any case. In addition, there is a recognition that may countries, both inside and outside the EU, are each doing their own thing on AI on a fairly uncoordinated basis. The briefing provides a nice summary of each of these with links to relate documentation. (As I’ve mentioned previously, Canadian efforts compare weakly, if only a volume basis, but our contribution to guidelines for government agencies are noted). Across the globe, most ethical principles are identified as being voluntary in nature.
The briefing notes that a “number of legally binding instruments could be adopted to translate ethical rules into hard law and make them mandatory for the most influential AI industry players in the EU.” It also notes that across EU jurisdictions the proposed ethical principles line up rather closely. This would make EU wide regulations of course easier. It’s also not very surprising since there aren’t really a lot of outlier views on the ethical front.
Also, because it’s nice to be right, there’s a good section on standardization. I mentioned last week that standardization could be a back door into regulation. The briefing notes that standards could become mandatory though at this point there is not enough consistency or we are too early in standars evolution to do that.
The briefing highlights three areas where AI legislation proposals are more robustly discussed:
-on transparency of decision-making systems (with a regulatory bord to oversee this);
-with respect to healthcare; and
-with respect to face recognition technology.
This seems like an oddly truncated list, leaving out fundamentally important areas like banking and financial markets or transportation. It’s also notable that the limited Canadian contribution–guidelines for usage of AI by government–doesn’t make it into this list. As much as my comment son Canadian endeavours in this field are not complimentary, this contribution on guidelinces for public services could actually be something that could be easily replicated.