The white paper

The European Commission published its long-awaited white paper on Artificial Intelligence on February 19, 2020 and has an open consultation for feedback until June 14, 2020. The white paper centers around two main pillars: excellence and trust, but given the earlier statements of von der Leyen about launching an AI regulation during her first 100 days, expectations were mostly focused on the approach for regulation (trust).

Much feedback has already been published by many different organizations. In this note, we briefly summarize the main messages of the white paper, express our support for some of the feedback published by others and provide our additional feedback that we haven’t seen elsewhere.

The white paper focuses on Excellence so that Europe becomes more competitive in the global AI ecosystem, and on Trust so that AI will be adopted at a large scale without causing negative impacts. We agree that Excellence is important and endorse the proposed actions to increase investments, skills and retention of world-class researchers and engineers. We also endorse the creation of a European center of excellence. However, we recommend the European Commission to let excellence prevail over politics, and to avoid the typical fragmentation. We need member states to think “Europe”, and don’t need another European Institute of Technology (EIT) that alludes to the world-renowned MIT, but due to its fragmentation is not even a shadow of it. We also applaud the focus on technology transfer from research to market, something where Europe has systematically failed in the past two decades. The Commission is creating local ecosystems called Digital Innovation Hubs that aim to bring AI technology to SMEs and the local market. We give the EC the benefit of the doubt and hope that this approach will make a difference compared to the current situation.

Regarding trust, the focus is on the avoidance of the potential negative impacts of AI, in particular related to autonomous decisions such as liability, explainability and transparency, bias and discrimination, and privacy. While not explicitly mentioned, the focus is on the risks of “narrow AI” powered mostly by deep learning and other machine learning techniques. The Commission aims to increase trust by distinguishing between high-risk and low-risk AI applications, and proposes regulation in case of high risk. An AI application is considered high risk if it belongs to a high-risk sector and at the same time concerns a high-risk use case. The Commission proposes to define an exhaustive list of high-risk sectors and use cases with periodical review. Examples of high-risk sectors would be health, transport, police, and legal. A typical example of a high-risk use case would be facial recognition.

Regulation reduces legal uncertainty, and this is important for companies to invest in AI technologies.

Many organizations already have published their feedback. Here we will summarize the points which we agree with.

Looking at the majority of the examples in the white paper and at the requirements on high-risk AI (training data, data and record keeping, robustness and accuracy, human oversight), it looks like AI regulation is mainly aimed at autonomous decision making, leading to typical problems such as black box algorithms (explainability & transparency) and bias (undesired discrimination). AI is not perfect and makes errors (false positives and false negatives), hence the importance of robustness and accuracy. Maybe we should come up with criteria for when autonomous decisions require HITL, HOTL or HOOTL. What types of decisions require a human IN the loop (HITL), approving every decision a machine makes? What types of decisions require a human ON the loop (HOTL), monitoring the AI outcome and correcting when wrong? And what types of decisions can be taken without any human intervention (H Out Of TL)?

Therefore, we recommend the Commission makes a thorough assessment of what currently operating AI systems would need to stop if regulated, and what would be the impact on societies and economies.

In case the Commission decides to introduce AI regulation, it becomes important to define how organizations will be audited (ex-ante or ex-post) for compliance. The white paper, however, only mentions “audit” once, without any further elaboration. Such auditing process would need to adequately cover the possible regulation (robustness, bias, transparency, liability, etc.) in a flexible framework that allows externalization to designated, competent auditing organizations. The objective would be to come up with a scorecard with relevant metrics which evaluate the trustworthiness of individual AI products.

In conclusion, we welcome the AI white paper, especially because it is generating an essential debate on the future of AI (regulation). And we are confident that the Commission will come up with an improved version based on the public consultation. However, we should not forget that even if the use of AI implies significant risks, it could still be better or generate less risk than when only humans are making decisions as it happens today. And we end with an open question: would the world be a worse or better place today if AI had been available 120 years ago?

This note expresses the opinion of the authors.

Richard Benjamins is co-founder, Vice-president and co-director of the Area of Ethics and Responsibility of OdiseIA.

Idoia Salazar is co-founder, President, and co-director of the Area of Ethics and Responsibility of OdiseIA.

Javier Fernández-Castañón is co-director of the Area of Health of OdiseIA

Related posts:

bibit jambu air dalhari tanaman buah 60cm mail %BERGARANSI HIDUP SAMPAI TUJUAN. Dikembangkan terutama di kabupaten Sleman, Jogjakarta Buah berwarna merah, dengan rasa yang manis dan berair Berukuran…

The decision on what my soul would believe was not wrapped in my own home traditions, or some near-death -experience and not even a reflective meditative journey. I was a teen, and remember the…

In the world of fashion, brands cluster into different segments. While Zara is fashionable and affordable, Chanel is timeless and costly. From how a store feels to product packaging, a brand takes a…