Thread by @St_Heumann

A few thoughts on the leaked EU Commission's White Paper on a European Approach towards #AI
(thanks to @F_Kaltheuner for pointing it my way)

THREAD 👇

The white paper references the Commission's ongoing activities on #AI. But the main thrust is the analysis of different regulatory options to make good on @vonderleyen pledge to present legislation on #AI within the first 100 days in office.

The big story in the media such as this @POLITICOEurope piece (where you can find a link to the leaked draft) has been the consideration of a temporary ban of facial recognition in public spaces. politico.eu/article/eu-con…

While prominently reported in the media, the reference to facial recognition is rather short and well buried in the document. And the language is rather skeptical - calling the measure "far reaching" and expressing preference to deal with facial recognition through #GDPR.

Though this story might change the dynamics of the debate and push for more drastic measures like a ban of facial recognition technology that it seems the Commission actually wanted to avoid. #Clearview nytimes.com/2020/01/18/tec…

The far less reported story of the draft paper is that it reveals how difficult it is to make good on @vonderleyen's pledge. While the hype of #AI makes it appear unique & unprecedented, the truth is that there is already a lot of existing regulation & laws that would apply.

The document does, however, do a good job of fleshing out existing gaps in legislation that need to be addressed. But this requires a much more piecemeal approach - a far cry from the sweeping legislation that @vonderleyen called for.

Here is a very good list of gaps in current legislation that do not adequately address the nature and risks associated with #AI - it includes shortcomings in the Charter of Fundamental Rights (which mostly binds the public sector), product safety legislation, enforcement ...

The white paper also acknowledges the difficulty of addressing the entire lifecycle of #AI systems. I particularly like the principle of attributing responsibility with actors best placed to address the problem. - This means obligations for both developers and users of AI.

And then there is a long list of ex ante regulatory requirements. That's the stuff that will cause conflict with those favoring a "light touch" regulation (US government, tech companies)

The paper discusses five regulatory approaches
1. voluntary labelling
2. focus on public sector regulation and facial recognition (this is where the potential ban is brought up)
3. risk-based regulation
4. safety and liability
5. oversight & enforcement

The Commission states a clear preference for Option 3 (risk based regulation) combined with Option 4 (safety and liability and Option 5 (oversight and enforcement)

Again a clear rejection of the light touch approach favored by the United States (Option 1 - voluntary labelling). And in line with the recommendations of the German Data Ethics Commission - see the graphic below on its risk based regulatory approach to #AI.

The rejection of Option 2 (focus on public sector) indicates that the Commission will seek to avoid a fight with member states on adoption of #AI by security agencies and public administrations. This might not bode well for the proposed ban of facial recognition in public spaces

But a ban of facial recognition still can (and in my opinion should) be on the table within the risk based approach - as the broader risks to #privacy, #freedomofexpression and other fundamental rights are obvious.

Enforcement (as we know from #dataprotection) will be crucial. Here again a patchwork is most likely as "Member states will be free to decide that these tasks should be entrusted to existing authorities" or could "appoint authorities responsible for monitoring

the overall application and enforcement of the future regulatory framework for artificial intelligence." The case of Ireland (HQ to most big tech companies in Europe) in enforcement of #GDPR is telling. Story: politico.com/story/2019/04/…

Final thoughts. Analysis of the White Paper is straight forward & clear. Commission knows what it is talking about. The White Paper provides a thorough analysis on which the development of the EU's regulatory agenda can be based - though no blueprint for a single #AI legislation.

The introduction places this regulatory approach in the broader EU strategy on #AI. It is very clear about the investment challenge where the EU as a region is far behind North America and Asia.

It is also serious about the challenge of building competitive data infrastructures and computing capabilities to compete on #AI. Here it makes a bold prediction on the trend of edge computing leading to 80% (!!) of global data to be stored locally by 2025.

What I particularly like is that the section on promoting research and uptake on #AI does not just present current initiatives and thinking but ends with key questions. Would be curious what people think about those. But they make clear that data is at the heart of #AI policy.

A lot to dig from this White Paper. Surprised that I have not seen much more about it in the media or my twitter timeline. Please let me know what I have overlooked and share interesting analyses that you have seen. This is a fascinating space to watch. END.