EU 消費者保護の新側面 デジタルサービスとAI

こんにちは、丸山満彦です。

EUが、人工知能によって提供されるデジタルサービスに関連して、特に消費者保護、データ保護、プロバイダの責任に関する新たな課題と機会に関する報告書を公表していますね。

この報告書では、デジタルサービスが消費者のデータを処理したり、広告やその他のメッセージで消費者をターゲティングしたりするためにAIに依存する場合について、消費者のプライバシーや自律性に対するリスクや、消費者に優しいAIアプリケーションの開発の可能性について言及していますね。

また、違法・有害なコンテンツを検出して対応するためのAIシステムの使用に関連して、サービス提供者の責任に対するAIの関連性についても言及していますね。

● European Parliament - Think Tank

・2020.04.15 (Study) New aspects and challenges in consumer protection - Digital services and artificial intelligence

・[PDF]

ーーーーー

	EXECUTIVE SUMMARY	エグゼクティブサマリー
1	INTRODUCTION	序論
2	DIGITAL SERVICES ANDCONSUMERS: THE RISE OF INFLUENCE MACHINES	デジタルサービスと消費者：影響力を持つ機械の台頭
3	AI AND TARGETED ADVERTISING	AIとターゲティング広告
4	FROM ADVERTISING TO FILTER BUBBLES	広告からフィルターバブルまで
5	A NEW LANDSCAPE	新たな景色
6	MONETISING VS NON-MONETISING CONSUMER DATA	収益化する消費者データと収益化しない消費者データの比較
7	CONSUMERS’ PROFILING AND CONSENT	消費者のプロファイリングと同意
8	FROM DATA PROTECTION TO CONSUMER PROTECTION	データ保護から消費者保護へ
9	FROM CONSUMER PROTECTION TO CONSUMER EMPOWERMENT	消費者保護から消費者エンパワーメントへ
10	PROVIDERS LIABILITY AND THE DIGITAL SERVICE ACT	プロバイダ責任とデジタルサービス法
11	ISSUES ON INTERMEDIARY LIABILITY	代理人に関する課題
12	AI IN CONTENT FILTERING/MODERATION AND CONSUMER PROTECTION	AIによるコンテンツフィルタリング/モデレーションと消費者保護
13	RECOMMENDATIONS	推奨
	REFERENCES	参考文献

EXECUTIVE SUMMARY

Background

Online consumers find themselves in an unbalanced relation to service providers and traders. A range of powerful intermediaries has emerged, which deliver key services, such as access to the Internet infrastructure, online search, contentsharing, cloud computing, and online payments. Some of these services are offered for free to final users, being supported by advertising revenue. Ads are automatically targeted to individual consumers, the targeting being based on information collected by tracking them.

AI has provided technologies with which to exploit the wealth of consumers’ information so as to better target individuals. In particular, machine learning has enabled traders to grasp correlations between consumer data (purchases, sites visited, likes on social networks) and possible responses to ads. The ability to predict consumers’ reactions provides traders with the ability to trigger such reactions through appropriate ads and other messages. This ability could morph into manipulation, as consumers’ responses could be based on irrational aspects of their psychology, on a lack of information, or on a situation of need.

A widespread mechanism for changing behaviours has emerged whose final purpose is to modify people’s purchasing behaviour through targeted ads. Thanks to big data and AI, traders may come to know what may influence particular consumers or groups of them one way or the other. More generally, a personal data economy is emerging where all kinds of personal data are collected and exchanged, their value consisting in their possible uses to anticipate and modify the behaviour of people.

The business model based on providing “free” services paid through advertising has an impact that goes beyond e-commerce. In order to expose consumers to ads, platforms have to attract and keep consumers on their websites. AI can discover what kinds of messages and information are more likely to achieve this goal. These tend to include messages —among which rumours or fakes —that please or excite users, confirm their prejudices, trigger negative feelings (such as rage or disgusts), and provide for additive symbolic rewards and punishments. Moreover, individuals tend to be served with kinds of content and messaging that have attracted or pleased similar people in the past. This may lead to separation and polarisation in the public sphere.

AI technologies are also increasingly used by online service providers, to detect and react to unlawful and inappropriate online behaviour. While AI technologies can contribute to effective moderation, enabling providers to cope with the huge growth and accelerated dynamic of user-generated content, they may also deliver inaccurate, biased or discriminatory responses, to the detriment of freedom of speech and users’ rights.

Aim

The aim of this report is to identify key issues concerning the situation of consumers relative to service providers and traders as well as the ways in which AI is impacting relations and interactions, and to propose possible solutions.

Presently, consumers’ personal data are most often extracted from online servicesat no cost, and then used and exchanged to the benefit of providers and traders. One way out of this predicament consists in accepting that personal data are a tradable commodity, but ensuring that data subjects draw some benefit from the use made of their data, while also enabling them to exercise some control over these data. The other way out consists in ruling out the possibility that personal data can be a tradable commodity, i.e., in barring vendors from offering services or benefits in exchange for personal data. On the latter approach, personal data should be used only when necessary to deliver a service requested by consumers, not as something given in exchange for a different service. EU law has not yet chosen between these two models, nor has it found a way to reconcile them. This report argues that guidance should be provided in this regard. Consumer choice can play an important role, whichever approach is adopted, but effective protection of consumer privacy requires that consumers should not be deceive by “design tricks” or “dark patterns” that stealthily induce them to consent to the processing of their data.

The AI-based processing of consumer data is relevant to the main goals of consumer protection law, such as protection of the weaker party, regulated autonomy, and non-discrimination. First, as noted above, the use of AI by vendors/retailers and service providers may introduce further imbalances between these parties on the supply side and consumers on the demand side. Second, the manipulative use of big data and AI may limit the independence of consumers. Third, automated decisions may work to the disadvantage of certain individuals and groups, and without any acceptable rationale. In this regard, some clarification is also needed, possibly through soft law instruments.

Even though the risks that AI poses to consumers are significant, no less important are the opportunities opened up by AI. AI can support citizens and their organizations so that they may not only make better use of the opportunities available in the market, but may also resist and respond to unfair and unlawful behaviour by AI-powered companies. Consumer-empowering AI technologies can support consumers in protecting themselves from unwanted ads and spam; they can enable consumers to identify cases where unnecessary or excessive data is being collected or where fake and untrustworthy information is provided; they can enable support consumers and their organisations in detecting violations of the law, assessing compliance, and obtaining redress. The public could support and incentivise the creation and distribution of AI tools for the benefit of consumers, as data subjects and citizens.

In the domain of Internet moderation, AI may enhance the capacity of providers to detect and react to unlawful and inappropriate online content and behaviour. AI systems can filter out some unlawful content or flag content for human moderation. AI can also assist human moderators by increasing their productivity. It may also be used to reduce the potentially harmful effects of content moderation of individual moderators.However, there is the risk that AI-based moderation may lead to outcomes that are inaccurate, unfair or discriminatory, to the detriment of freedom of expression and information.

Recommendations

The report includessome policy recommendations as follows:

a) Consumers should have the option not to be tracked and (micro)-targeted and should have an easy way to express their preferences;

b) The grounds on which service providers and traders cannot price-discriminate should be specified;

c) It should be considered how discrimination in ad targeting is to be addressed;

d) Guidance should be given concerning what algorithmic practices count as instances of aggressive advertising;

e) Guidance should be given concerning cases in which consumers have a right to contest a decision that undermines their interests;

f) Consumers should be provided with information on whether and for what purposes they are tracked and on whether they are receiving information for advertising purposes;

g) Protection of consumer privacy requires preventive risk-mitigation measures in combination with collective redress;

h) The development of consumer-friendly AI-technologies should be encouraged and supported. Service providers should be prevented from blocking legitimate tools for the exercise of consumer rights;

i) Liability limitations for online providers should also apply to “active” providers, such as search engines, online repositories, and social networks, regardless of whether user-generated content is organised, presented and moderated by humans, by algorithms or both;

j) Limitations on providers’ secondary liability should not apply when providers have failed to adopt reasonable precautionary measures that could have prevented that behaviour or mitigated its effects. This failure may also depend on not having adopted the most effective AI technologies;

k) The availability of AI technologies for detecting unlawful online content and behaviour should be encouraged, in combination with human judgment;

l) Third-party filtering/moderation should be encouraged so as to broaden access, and so should the sharing of datasets (to train AI classifiers) and software, so that both are accessible to small companies as well.