The European Parliament broadly supports the Commission's risk-based approach to artificial intelligence (AI). This was revealed in a first round of discussions in the European Parliament's Legal Affairs Committee on Tuesday 12 March.
The starting point: the February package
On 19 February, the von der Leyen Commission presented a package of documents on the technologies of the future (see EUROPE 12429/5). The aim was to open discussions, including a public consultation, with a view to possible legislative proposals by the end of 2020. However, with the Covid-19 pandemic, there is a good chance that the schedule will change.
Although the division of this dossier between committees has not been easy, the European Parliament is now fully involved in the matter. While the negotiations for a special commission are more or less at a standstill, a dozen or so substantive reports are now being prepared.
The Legal Affairs Committee held a first round of discussions on Tuesday (12 May) on three of them for which it is responsible, which concern ethical issues, civil liability regimes and intellectual property rights. MEPs broadly welcomed the risk-based approach proposed by the Commission and included in the draft reports by Iban Garcia del Blanco (S&D, Spain) on ethics and Axel Voss (EPP, Germany) on civil liability.
Control and risk-based liability
Mr Garcia del Blanco suggested that high-risk technologies should be subject to a conformity assessment carried out on the basis of "objective and harmonised criteria" laid down by a new AI agency and "carried out and verified by national supervisory bodies".
Mr Voss, on the other hand, proposes to adapt the Product Safety Directive to include a strict liability regime for high-risk AI "deployers". Subject to compensation requirements (maximum €10 million in the event of death and €2 million in the event of damage caused to property), these deployers would be obliged to take out liability insurance comparable to that which exists for owners of a car (it is, moreover, through this analogy that he illustrates the notion of “deployer”. For others, deployers of technologies that are not high-risk, the rapporteur provides for a "presumption of fault" and defers to national law.
Both rapporteurs opt for legislation in the form of a regulation, rather than a directive.
Amendments to be foreseen
However, during the tour de table organised on 12 May, a large majority of political groups questioned the idea of a new European AI agency (EPP, Renew Europe, ID and RCE). The rapporteur has said that he is prepared to revise his copy, provided that this request is not justified by accounting logic. "If we don't go to an agency, it should only be because there is no added value", Mr Garcia del Blanco told his colleagues.
Many MEPs also criticised the lack of coordination between the two reports as regards definitions, including systems at risk. In an annex, Mr Voss proposes a closed list which could be amended by the Commission with the help of a standing committee of experts and stakeholders from the Member States. At this stage, it identifies autonomous robots, autonomous public place cleaning devices, autonomous cars (level 4 and 5), autonomous traffic management systems and unmanned aircraft. For the S&D Group, Tiemo Woelken also notes that the responsibility should not lie exclusively with the deployers.
A quick discussion also took place on the report on intellectual property by Stéphane Séjourné (Renew Europe, France), while the discussion on Gilles Lebreton's working paper was postponed.
Facial recognition and lethal weapons
So far, only Mr Garcia del Blanco's report tackles the delicate issue of facial recognition, on which the European Commission has so far refused to take a position. The Socialist rapporteur refuses to ban it. Its draft report addresses "remote recognition technologies, such as biometric recognition, (which) are deployed or used by public authorities in Member States for the purpose of responding to a national emergency". In such cases, notes the rapporteur, the authorities should ensure that their deployment is limited to specific objectives, limited in time and carried out in accordance with European values.
For his part, Axel Voss focuses on damage to life, health, physical integrity and property and excludes the use by AI systems of biometric data or facial recognition techniques already covered by other rules, such as the general data protection regulation.
However, this issue may resurface in other parliamentary reports on AI. Reports are still awaited on: - education (Committee on Education); - safety and liability (Committee on the Internal Market and Consumer Protection); - excellence and trust (Industry Committee); - criminal law (Civil Liberties Commission).
As for the working paper by MP Gilles Lebreton (ID, France), it addresses the military aspects of AI, which he considers to be absent from the Commission's work. The Frenchman points out that the European Parliament voted in 2018 in favour of an international treaty banning "autonomous lethal weapons systems", but that this is unlikely to see the light of day, given the positions of the major military powers. "If the European Parliament wants to influence the debates of the group of governmental experts set up by the UN, it is preferable that it puts forward constructive proposals to regulate the use of these systems", notes Mr Lebreton.
Uncertainties about the Commission's timing
At present, the European Commission is continuing its consultations. It has just decided to extend the deadline for replies for a second time - until 14 June.
According to a revised draft 2020 work programme several weeks ago, the Commission could postpone certain initiatives until 2021. This could be the case for the 'Horizontal Initiative on Artificial Intelligence', the 'Responsibility' initiative as well as the revision of the Machinery Directive (2006/42/EC) and the General Product Safety Directive (2001/95/EC). According to the same document, delegated acts under the Radio Equipment Directive (fraud, privacy and data protection, connected devices) would be maintained in 2020.
However, it will be necessary to wait until the end of May, when the Commission will publish the final version of its revised work programme, to confirm these deadlines.
Link: https://bit.ly/2T1YCza (Original version in French by Sophie Petitjean)