Ref. Ares(2020)3380274 - 29/06/2020
CAB BRETON 230
Lucia Caudet meeting with
Google
Brussels, 4 May 2020
Line to take
x The Commission supports a regulatory and investment oriented approach with
the twin objective of promoting the uptake of Artificial Intelligence (AI) and of
addressing the risks associated with certain uses of this technology.
x The White Paper published on 19 February contains policy proposals to
implement the EU’s human-centric approach to AI. The proposals aim to create
an ecosystem of excellence, which can only be realised if accompanied by an
ecosystem of trust.
x According to the Commission’s approach to AI, a framework based on
fundamental EU values and rights and on robust safety requirements will give
citizens the confidence to embrace AI-based solutions while encouraging
businesses to develop them.
x Issues related to digital services, and online platforms in particular, will be
addressed in the forthcoming Digital Services Act (DSA) package. We are,
amongst other, looking at the transparency and reliability of a series of
algorithmic processes, not least related to online advertising, recommender
systems of content moderation.
x The open public consultation for the DSA should start in the coming weeks and
we welcome contributions and engagement on the specific issues.
Scene Setter
AI
Google- and even more so its mother company Alphabet, is very active in the field of
AI.
Google has a lot to lose from possible regulation on digital issues. Therefore, it is
generally supportive of non-legislative actions reassuring the public. In particular, on AI
it has been proactively promoting ethical principles and their implementation.
The seven AI principles of Google (launched in June 2018, see annex) are non-
controversial but generic. Similarly, Google’s red lines are sound but worded in a way
avoiding stringent commitments.
On 21 November 2019, Google introduced its “Google Cloud AI Explanations” tool
which is supposed to quantify each data factor’s contribution to the output of a machine
learning model. It also introduced “Model cards”, a kind of standardised product
information for AI products.
Both initiatives have the advantage for Google of being proactively ethical and
reinforcing its position vis-à-vis start-ups that do not have the resources to launch
similar projects.
Google’s White Paper on AI governance (February 2019) also calls for regulation
through dialogue with industry rather than legislation.
Page
1 of
7
CAB BRETON 230
Lucia Caudet meeting with
Google
Brussels, 4 May 2020
Google is represented in the AI High-Level Expert Group (HLEG) established by the
EC in June 2018. Therefore it could not participate in the deep dives (around 50
organisations, not part of the HLEG, were selected for an in depth evaluation of the
trustworthy AI assessment list developed by the HLEG), but it did participate in the
online survey of the assessment list. In its submission, the company agrees with the
topics covered by the EU Ethical Guidelines Assessment List. In establishing Google’s
procedures to implement their AI principles they had similar considerations — albeit in
a manner tailored to specific applications and issue areas rather than a generic
standardised list. Because of this, the Assessment List in its current form seems less
effective for Google to adopt than their application and context specific approach.
However, they agree that a revised version of the Assessment List could be a helpful
starting prompt for those organisations who have not yet established procedures for
ensuring responsible AI implementation. More specifically Google proposes
clarifications and simplifications to the list.
Google (following a discussion with CNECT services) submitted two initiatives to the AI
& COVID initiatives repository, namely: Kaggle COVID-19 Open Research Dataset
Challenge, and Google Cloud Rapid Response Virtual Agent.
ONLINE PLATFORMS
The Commission announced, in the Communication on Shaping Europe’s Digital
Future, a Digital Services Act Package, including two instruments:
1. New and revised rules to deepen the Internal Market for Digital Services, by
increasing and harmonising the responsibilities of online platforms and information
service providers and reinforce the oversight over platforms’ content policies in the
EU.
2. The Commission will further explore, in the context of the Digital Services Act
package, ex ante rules to ensure that markets characterised by large platforms
with significant network effects acting as gatekeepers, remain fair and contestable
for innovators, businesses, and new market entrants.
The package is announced in the Commission’s Work Programme for Q4 2020.
Commissioner Breton announced in the French Senate that some minor delay might
follow. No formal update to the Commission Work Programme was made to date.
Page
2 of
7
CAB BRETON 230
Lucia Caudet meeting with
Google
Brussels, 4 May 2020
Background
Google’s AI principles
1. Be socially beneficial.
2. Avoid creating or reinforcing unfair bias.
3. Be built and tested for safety.
4. Be accountable to people.
5. Incorporate privacy design principles.
6. Uphold high standards of scientific excellence.
7. Be made available for uses that accord with these principles.
Google’s red lines
1. Technologies that cause or are likely to cause overall harm. Where there is a
material risk of harm, we will proceed only where we believe that the benefits
substantially outweigh the risks, and will incorporate appropriate safety constraints.
2. Weapons or other technologies whose principal purpose or implementation is to
cause or directly facilitate injury to people.
3. Technologies that gather or use information for surveillance violating internationally
accepted norms.
4.
Technologies whose purpose contravenes widely accepted principles of
international law and human rights.
Page
3 of
7
CAB BRETON 230
Lucia Caudet meeting with
Google
Brussels, 4 May 2020
EU strategy on Artificial Intelligence
On 25 April 2018 the Commission adopted the
Communication "Artificial
Intelligence for Europe", outlining three main priorities: (i) boost the EU's
technological and industrial capacity and AI uptake across the economy; (ii) prepare for
socio-economic changes brought by AI; (iii) ensure an appropriate ethical and legal
framework.
The Commission then adopted a
Coordinated Plan on the development of AI in
Europe to maximise the impact of investments and to collectively define the way
forward on 7 December 2018. It comprises some 70 individual measures, including
research, investment, market transfer, skills and talent, data, ethical and legal
framework and international cooperation.
Regarding
safety and liability the Commission established an expert group on
“Liability and New Technologies”, in two distinct formations. As a result of the expert
work, the Commission will publish interpretative guidance on the Product Liability
Directive and a Report on the broader implications for, potential gaps in and
orientations for, the liability and safety frameworks for Artificial Intelligence, Internet of
Things and robotics.
In April 2019
Ethics Guidelines for Trustworthy AI (accompanied by an
assessment
list aiming at operationalizing them) were published by the High-Level Expert Group on
AI established by the Commission in June 2018. The Guidelines are addressed to all
relevant stakeholders developing, deploying or using AI, encompassing companies,
organisations, researchers, public services, institutions, individuals or other entities.
The guidelines are based on fundamental rights and ethical principles.
Policy and investment recommendations were also developed by the High-Level
Expert Group, who looked at the main goals or impacts that they would like to achieve
for AI, and at the main enablers that can make such impact happen such as funding,
skills, data and infrastructure as well as regulation. The final version of the policy and
investment recommendations was presented on 26 June 2019. It was addressed to
Member States and to the new Commission.
On 19 February 2020 the Commission presented a digital package containing inter alia
a
White Paper on Artificial Intelligence, a data strategy and a report on the safety
and liability aspects of artificial intelligence. The White Paper launched a wide public
consultation that will run until 14 June 2020.
This White Paper contains policy proposals to implement the EU’s human-centric
approach to AI. The proposals aim to create an ecosystem of excellence, which can
only be realised if accompanied by an ecosystem of trust.
According to the Commission’s approach to AI, a framework based on fundamental EU
values and rights and on robust safety requirements will give citizens the confidence to
embrace AI-based solutions while encouraging businesses to develop them.
To create an ecosystem of
excellence the Commission proposes to use Horizon
Europe and the Digital Europe programme to support:
x A new public-private partnership in AI, data and robotics;
x The strengthening and networking AI research excellence centres;
x The set up of AI testing and experimentation facilities;
Page
4 of
7
CAB BRETON 230
Lucia Caudet meeting with
Google
Brussels, 4 May 2020
x AI-focused digital innovation hubs, which facilitate the uptake of AI by SMEs
and public administrations;
x An alliance of universities for the strengthening of AI (data science) skills
x An increase in the provision of equity financing for innovative developments
and deployment in AI through InvestEU.
x Sector dialogues to facilitate the development of a new programme (‘Adopt
AI’) to support public procurement of AI systems.
x International
cooperation
with
like-minded countries, companies and civil
society on AI based on EU rules, values and safety requirements.
To create an ecosystem of
trust the Commission proposes the following:
x A risk-based and proportionate regulatory approach;
x The identification of high-risk AI systems through a combination of (i) sector
and (ii) concrete use of the system;
x In certain cases, AI systems can be considered high-risk, irrespective of the
sector (e.g. recruitment processes and remote biometric identification);
x Mandatory requirements for high-risk AI systems only:
o Training data should be of high quality and respect EU’s rules and
values
o Record keeping of the relevant data sets and of the programming and
training methodologies
o Provision of information about the AI system’s performance
o Robustness
and
accuracy
o Human
oversight
x For AI systems that are not high-risk, a voluntary label could be considered.
x For remote biometric identification (e.g. facial recognition), the White paper
is launching a broad European debate.
x The goal is to intervene only where this is strictly needed and in a way that
minimises the burden for the economic operators.
x Thus, the EU is supporting a regulatory approach that is proportionate, not
excessively prescriptive and does not hamper innovation worldwide.
With the White Paper the Commission has launched a wide public consultation ending
on 14 June 2020. We invite comments from every strand of society to come back to us
with their input, in order for us to assess it and start preparing what might be needed in
legislation as well as to launch investment and funding that will allow us to make the
most of AI.
The EU is also closely cooperating at the international level with the OECD, G7, G20
etc. and there are many bilateral contacts, for example with Canada, Japan and
Singapore. It is important to have (at least a partial) consensus on ethical values at
global level.
Page
5 of
7
CAB BRETON 230
Lucia Caudet meeting with
Google
Brussels, 4 May 2020
Contact
- For
AI
, CONNECT A2, Tel.
-
For algorithmic processes on digital services and the Digital Services Act:
, CONNECT F2, Tel.
Page
7 of
7
