Telefónica has contributed to the European Commission (‘EC’) call for feedback on its Draft Ethics Guidelines for Trustworthy AI. The proposal issued by the EC is a suggestion on how AI must respect fundamental rights, applicable regulation and core principles and values; while being technically robust and reliable, in order to ensure that no harm comes out of its utilization.
Telefónica welcomes the commitment of the European Commission to advance on the social impact of AI and proposes certain considerations to aid on the adoption of those measures.
To improve the guidelines the following issues summarize Telefónica’s most relevant proposals:
- Inclusiveness, effectiveness and competitivity: The invitation to endorse the guidelines should not be targeting solely European organizations, but extended to all organizations with close ties with EU citizens, businesses and governments irrespective of where geographically based. The guidelines must also consider the policy realities in other markets around the world, in order to ensure the competitiveness of Europe, building on Europe’s ability to drive a collective agreement and act as a block. For this matter, it is important that in the unfortunate event that these guidelines turn into Regulation they are not a enforced only on European businesses but on all businesses in other regions of the world that are serving European customers.
- Monopolization of data: we believe that access to (behavioral) data is critical for the development and implementation of AI. Its monopolization in the hands of very few companies willlimit the emergence of AI solutions from other alternative players, thus enabling potential abuses on citizens and uneven distribution of AI benefits. Such situations should be addressed by mitigating dominance.
- Trust and quality of the business model: business models enabled by AI technology should not pursue an unethical purpose. On this regard, the “Do not harm principle” should prevail, while the opportunity and applicability of the “Do good principle” is to be challenged as boldly applying it would restrict companies’ freedom to innovate. Guidelines should not demand others or the technology to do good but to do no harm. Additionally, as profiling is widely used and needed for most commercial activities, and the “Do not harm principle” already enshrines eliminating all negative actions, reference to eliminate “negative profiling” should be eliminated: “profiling” is neutral from a normative perspective, and what makes it harmful is the purpose of the profiling.
- Progressive and individualized application: the guidelines should be applied with different intensity according to the specific impact of an AI based system throughout the levels in the AI system life cycle (development, deployment and usage). The same way cybersecurity requirements are different for a domestic watering IoT device and a nationwide energy grid, AI principles should apply differently depending on its impact. Requirements for trustworthy AI and the technical and non-technical methods to achieve trustworthy AI should be domain and application specific.
- Informed consent: the Guidelines consider informed consent as an ethical value that puts in practice the fundamental right of human dignity. This does not take into account that even the GDPR considers five legal bases for processing other than consent which of course do not negatively impact human dignity. We would recommend including a different example on how to go from fundamental right to principles and values
Overall the Guidelines provide a good and thorough approach to ensure that AI will have much more good use than bad (intentional or unintentional) use. We support that view, as can be ascertained from Telefonica’s AI principles:
The principles call for a trustworthy environment for our stakeholders regarding how we will develop and use AI, and by Telefónica’s more general Business principles, where transparency and acting in accordance with non-negotiable ethical standards are two core principles.
Read our complete proposal for the European Commission public consultation here.