AI-Ethics landscape 1/2019

The landscape relies on the following color and symbol coding:

The data comes from a set of sources:

State actors

• Denmark: Published: January 2018
  Sources: Strategy for Denmark’s Digital Growth
  URL: https://eng.em.dk/media/10566/digital-growth-strategy-report_uk_web-2.pdf
  Summary: Denmark describes itself as fundamentally good at working responsibly. Accordingly, the statements on ethical procedures in the AI are quite brief. The explicit passage comprises a few lines and mentions general measures such as the establishment of data ethics – possibly including a data ethics code – and discussions on the topic in internationally relevant forums. Total expenditure of DKK 5 million (≈ EUR 670,000) is planned for 2020.
• Italy: Published: March 2018
  Sources: Artificial Intelligence. At the Service for Citizens
  URL: https://ia.italia.it/en/
  Goal: Ensuring that AI should always have the priority to serve the citizens
  Summary: Italy has one of the most comprehensive concepts for ethics in AI among the EU states and intends to use it specifically to align all AI strategies: Italy wants to start with a conscious reflection on how we want the world to be in the future. They refer to the virtuous ethics of Aristotle, but at the same time emphasize that an agreement on the improvement of human well-being is also possible with followers of other ethics such as Kant’s ethics, Confucian ethics, Shinto or Ubuntu. It is important to reaffirm the “anthropocentric principle” according to which AI should always serve the citizen first and foremost. In principle, the ethical application of AI in public administration seems to be a central point of reference for Italian considerations.
  Ethics in AI is challenge No. 1 in Italy’s AI strategy. They demand the formulation of general principles of justice as a basis for freedom as well as individual and collective law and propose four “elements” for public debate and scientific analysis. One concrete measure described is the establishment of a Trans-disciplinary Centre on AI, with the following objectives: To promote a debate on the evolution of ethics, to support critical reflection on emerging ethical concerns and to facilitate the involvement of experts and citizens in deriving regulations, standards and technical solutions based on technical and social aspects.
• France: Published: March 2018
  Sources: AI for Humanity
  URL: https://www.aiforhumanity.fr/en/
  Summary: France’s AI strategy is based on the report by mathematician and politician Cédric Villani. In the section “Ethical Considerations of AI Recent”, for example, he proposes a kind of special task force of certified experts who are commissioned to audit and test algorithms for constitutional issues. For the training of AI experts, he recommends Ethics by Design as an obligatory component in order to raise awareness of ethical issues right from the start. He also recommends a “digital technology and AI ethics committee” to lead a transparent public debate. Despite all the urgent questions, this debate should give rise to a long-term perspective.
  President Macron follows these and other recommendations. His official strategy includes the idea of an international group of experts similar to the Intergovernmental Panel on Climate Change (IPCC). The group should develop independent findings on transparency and fairness and incorporate them into educational programmes. Furthermore, research on more comprehensible AI is needed: a more explainable model, more interpretable interfaces and a better overall understanding of the underlying mechanisms. He would like to offer AI developers ethics training and a discrimination impact assessment for their algorithms. For the public sector, he plans the clear regulation of prediction algorithms in law enforcement and an intensive international debate on any development from LAWS to an observatory to curb their proliferation. As further conditions of ethical AI he names diversity and inclusion. Specifically, he wants to raise the proportion of women in digital engineering courses to 40% by 2020. In addition, he would like to draw trainers’ attention to the importance of diversity through guidelines and training. The government should promote the use of AI in social innovation programmes so that workers in the social sector also benefit from the technology.
• Sweden: Published: May 2018
  Sources: National approach to artificial intelligence
  URL: https://www.regeringen.se/4aa638/contentassets/a6488ccebc6f418e9ada18bae40bb71f/national-approach-to-artificial-intelligence.pdf
  Goal: Help to create sustainable AI
  Summary: Sweden names “sustainable” AI as its overarching motto and stresses that ethics, protection and safety must not be a downstream consideration, but an integral part of early stages of AI development. Furthermore, the development of interdisciplinary AI knowledge for engineers and decision-makers is desired and transparent and understandable algorithms are required for the development of rules, standards, norms and ethical principles. However, the ethical efforts must remain in balance with the urgent need for data in order to realize the social benefits of AI.
• Finland: Published: May 2018
  Sources: Work in the age of artificial intelligence Publications of the Ministry of Economic Affairs and Employment: Four perspectives on the economy, employment, skills and ethics
  URL: http://julkaisut.valtioneuvosto.fi/bitstream/handle/10024/160980/TEMjul_21_2018_Work_in_the_age.pdf
  Goal: Use technology for valuable social objectives
  Summary: Finland is committed to using the “ambitious goal” to use technology to achieve valuable social goals, with a strong focus on work design. A social vision is a prerequisite for the government to influence the development and application of AI in a meaningful way. They criticize that at this point in time none of the numerous international reports contains a strategy or action plan for achieving the often desired good artificial intelligence societies. For such a society they propose the values “transparency”, “responsibility” and “extensive societal benefits”. They emphasize that the legal and moral responsibility of algorithmic applications must clearly remain with the human being.
  If the future developments lead to job losses, support for the people affected is necessary, because after all they have paid for the progress with their taxes. Society should benefit equally from the advantages of AI. Mechanisms for sharing these benefits as well as income transfers are mentioned for this purpose. Developments in the world of work should be discussed at the level of the society as a whole, but also at the level of individual work organisations.
  For individual companies, the ethical handling of AI would increasingly influence the brand value. The large companies have already understood this, but further research from the perspective of technology, business management and political science is still necessary. This should be expanded into international cooperation, whereby Finland would also like to take on a pioneering role in the implementation of the AI strategies of the EU.
  The success of the strategy presupposes a strong education of the population on the topic of AI. Finland has presented a unique programme to increase understanding, diversity and inclusion in AI: With an AI Challenge since May 2017, Finland has supported a total of around 55,000 citizens (1% of society) to develop AI solutions themselves for their personal tasks.
  Other planned measures include the establishment of a parliamentary monitoring group to monitor technological progress, the formulation of a value base geared to the common good (with a centre of excellence specifically for the B2B sector) and the regulation of online platforms based on the example of the European Data Protection Regulation (GDPR). All in all, despite the hardly foreseeable developments in the field of AI, experiments should continue and experience should be gained. In doing so, one could rely on the support of state institutions in critical situations.
• UK: Published: June 2018
  Sources: Data Ethics Framework
  URL: https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/737137/Data_Ethics_Framework.pdf
  Goal: Become the world leader in ethical AI
  Summary: The United Kingdom unequivocally formulates its claim to be the “world leader in ethical AI”. The focus is on the ethical handling of confidential data to facilitate its availability and exchange. Central instruments for securing ethical AI include Data Trusts and a Centre for Data Ethics and Innovation (CDEI). The latter already seems to be the official information point for the concrete ethical approach of the United Kingdom. It is intended to act as a long-term interface between government, business and civil society. Specifically, it aims to assist the government with legislation, work with research institutions and think tanks to identify national challenges and global opportunities, and provide leading companies with a knowledge base and exchange platform for best practices.
  The CDEI’s statements on its basic ideas on ethical AI are broadly in line with the general discourse: to increase diversity among developers and to ensure fairness, transparency, liability and accountability. The following fields of action receive comparatively more attention: Targeting of citizens by advertisers, frameworks for the exchange of data, intellectual capital and property rights for innovations generated by AI.
• Netherlands: Published: October 2018
  Sources: AI voor Nederland. Vergroten Versnelen en Verbinden
  URL: https://order.perssupport.nl/file/pressrelease/a253ded8-1863-4d07-b79d-72639baefc51/6be5dcdc-b9d8-4675-bd80-9425826c840c/AIvNL20181102final.pdf
  Summary: The public-private partnership AINED, together with the Boston Consulting Group (BCG), has published a first report for the Netherlands. The report merely contains the reference to continuously gather experience with AI and its social consequences and then formulate regulations for the technology. Observers see an urgent need for action for the Netherlands not to move further away from its formerly good competitive position. AINED and BCG have announced a national strategy in the coming months.
• Austria: Published: November 2018
  Sources: Die Zukunft Österreichs mit Robotik und Künstlicher Intelligenz positiv gestalten. White Paper des Österreichischen Rats für Robotik und Künstliche Intelligenz
  URL: https://www.acrai.at/images/download/ACRAI_whitebook_online_2018.pdf
  Goal: Legal and moral framework for shift of decision-making authority to AI systems
  Summary: Austria is the only country examined that addresses robotics in its ethical approach to AI on an equal footing. It is interesting to note that the moral status of robots and AIs is “currently unclear”. Most other authors of ethical procedures would probably currently contradict this and exclude any moral status. Austria is also the only EU state to categorically ban LAWS. Apart from that, the other statements seem to be in line with the general discourse.
  In its ethical approach, Austria refers from the outset to the values of the European Human Rights Charter. These include, for example, justice, fairness, diversity and inclusion. A prerequisite for this is the transparency and security of the technology to be achieved. Currently, people always have to make and take responsibility for every ethical decision with regard to AI. A shift of decision-making authority to the AI system requires a clear legal and moral framework. In addition, “computational precautions for ethical behaviour” should be directly integrated into machine architecture and design (ethics by design). In order to align all further decisions with the common good of the population, Austria also plans to hold extensive expert consultations and public discussions.
• Germany: Published: November 2018
  Sources: Strategie Künstliche Intelligenz der Bundesregierung
  URL: https://www.bmbf.de/files/Nationale_KI-Strategie.pdf
  Goal: Use ethics guidelines to contribute to trademark “AI made in Europe”
  Summary: The German AI Strategy contains the most concrete measures and budget data of the compared states. The description of the ethical approach, on the other hand, corresponds approximately to the European average in its degree of abstraction.
  In principle, the regulatory framework is already very stable due to German legislation and the GDPR, but gaps in algorithm- and AI-based decisions, services and products would be examined. As with many other ethical procedures, transparency, traceability and verifiability of the algorithms are required as protection against bias and manipulation. It will be examined whether government agencies should also be established or expanded to safeguard these features. In any case, auditing standards and standards for impact assessment are planned. Technological possibilities for the enforcement of law and ethical principles are also to be explored.
  A focus will be on the digital education of citizens to empower and self-determine individuals in relation to AI. Ethics issues and sociological correlations are to become part of teaching, training and further education. Information campaigns are to be launched to promote the digital and media competence of citizens and to enable and motivate critical reflection on the basis of realistic application scenarios. The government emphasises the need to educate society and to continuously discuss ethical frameworks with equal involvement of the technical and natural sciences as well as the humanities and social sciences. For this purpose, forums and debates are to be made available. The Federal Government itself describes its approach as “ethics by, in and for design” and would thus like to make a contribution to the trademark of “AI made in Europe”.

Large state actors

• China: Published: July 2017
  Sources: New Generation Artificial Intelligence Development Plan
  URL: https://chinacopyrightandmedia.wordpress.com/2017/07/20/a-next-generation-artificial-intelligence-development-plan/
  Summary: China is often presented as a negative ethical example due to the Social Credit System announced for 2020. But even there, the ethical debate is intensifying based on the topic of moral agents. At least with regard to the quantity and concreteness of information on ethical procedures, the AI development plan presented in 2017 is absolutely comparable with European strategies. As with all other strategies, ethical norms, laws and regulations are sought. Their continuous optimisation through research activities is planned until 2030. In addition, an ethical framework for human-computer collaborations and an ethical code of conduct for research and development on AI products are to be developed. Potential risks of AI should be made assessable and there should be solutions for emergency scenarios. In industry and business, China is focusing on promoting self-discipline and strengthening manager morale. The latter was not formulated in such a way in any other ethics approach examined.
• USA: Published: December 2017
  Sources: FUTURE of Artificial Intelligence Act
  URL: https://www.congress.gov/bill/115th-congress/house-bill/4625/text
  Goal: Become the international leader in military ethics and AI security
  Summary: In the USA, state responsibility for AI currently lies primarily with the Pentagon. In January 2019, the United States began developing ethical principles for the military use of AI. The FUTURE of Artificial Intelligence Act, the central document of the state’s AI strategy to date, recommends ethics training for AI developers, but also recommends examining whether (!) and how ethical standards can be integrated into the development and application of AI. In February 2019, the Pentagon published the Summary of the 2018 Department of Defense Artificial Intelligence Strategy. Harnessing AI to Advance our Security and Prosperity. The goal is to become the international leader in military ethics and AI security. Stated measures are consultations by experts, investment in research and development and the provision of the results to create international standards for the use of AI in the military field. The US government’s commitment in the field of AI ethics, which still appears to be low in the public perception, is contrasted by a large number of non-governmental institutions with international reach.
• EU: Published: April 2019
  Sources: Ethics Guidelines for Trustworthy AI
  URL: https://ec.europa.eu/futurium/en/ai-alliance-consultation
  Goal: Development of a holistic framework for trustworthy AI systems
  Summary: The guidelines of the European Commission, prepared by the High Level Expert Group (HLEG), on ethics in AI were published in April 2019. According to the report, a “trustworthy” AI should contain three components over its entire life cycle: it should be lawful, ethical and robust. The report particularly emphasizes the necessary orientation of AI towards people and their needs (“human-centric design“). Only if AI is perceived as “trustworthy”, i.e. as benign and technically robust, would people want to get involved. The EU thus emphasises the necessity of ethics in AI for its dissemination and further development.
  The section on ethical AI proposes ethical guidelines to be taken into account in the development process and in the use of artificially intelligent systems. They should be developed according to the principles of respect for human autonomy, prevention of harm, fairness and explicability. In addition, the development and use of AI systems should focus on diversity. This should prevent that unprotected or historically disadvantaged groups, as well as situations with asymmetries regarding power or information, are included in the system with a certain bias. In addition, risks that could potentially arise through AI systems and thus no longer benefit individuals or society should be considered and mitigated with foresight. These ethical principles are translated into concrete requirements for which technical and non-technical implementation methods are presented.
  In addition, the EU envisages further concrete points to develop trustworthy AI systems: these include the promotion of research and innovation, training and education, the establishment of expert groups, as well as a transparent communication regarding capabilities and limitations of ethical AI systems.
  With its report the HLEG offers the most complete framework for the implementation of ethical AI among the examined ethical procedures. It represents a process of top-down implementation (from ethical intent to concrete implementation in the situation), but which itself was created bottom-up (from the concrete human situation to ethical intent). The EU Commission intends to continue this dialectical and dialogical process on an ongoing basis. For its part, HLEG proposes to communicate its approach in training courses and educational offers for AI developers and users. In addition, the European Commission plans to play a leading role in the global debate on ethical principles and to support relevant research.

Institutions

• Future of Life Institute: Published: January 2017
  Sources: Asilomar Principles
  URL: https://futureoflife.org/background/existential-risk/
  Goal: Goal of AI research should be to create not undirected intelligence, but beneficial intelligence
  Summary: Founded in Cambridge in 2014, the Future of Life Institute is a non-university research institution dedicated to the existentially threatening risks of nuclear wars, biotechnology, climate change and artificial intelligence. With the financial support of Elon Musk, the Institute has supported a total of 47 beneficial AI projects since 2015, with grants ranging from $100,000 to $1.5 million (Future of Life Grants). As part of the 2017 Asilomar Conference, the Institute developed the Asilomar Principles.
  The 23 Asilomar Principles are the best known list of ethical principles for beneficial AI. They were formulated in January 2017 at the Asilomar Conference in California by approximately 100 experts from AI, business, industry, philosophy and social sciences. Among the nearly 4,000 signatories to the principles are Elon Musk, Ray Kurzweil and Stephen Hawking. The principles state the importance of safe AI systems, transparency of the algorithms, security of personal privacy, and the overall benefit of AI for all of humanity.
• FAT / ML: Published: July 2016
  Sources: Principles for Accountable Algorithms and a Social Impact Statement for Algorithms
  URL: http://www.fatml.org/resources/principles-for-accountable-algorithms
  Goal: Provide AI developers and product managers with concrete ethical orientation
  Summary: FAT/ML stands for Fairness, Accountability and Transparency in Machine Learning. The conference, which has been held annually since 2014, is aimed at international AI researchers dealing with informatics methods for ensuring ethical AI (Ethics by design). The Principles for Accountable Algorithms and a Social Impact Statement for Algorithms were formulated in July 2016 and are intended to provide AI developers and product managers with concrete ethical orientation.
• Ethics in AI Initiative: Published: 2015
  Sources: Ethics in Artificial Intelligence Initiative
  URL: https://www.cs.ox.ac.uk/efai/
  Goal: How can ethical codes for AI be developed?
  Summary: The Ethics in Artificial Intelligence Initiative was founded in 2015 with funds from the Future of Life Grants at Oxford University. The initiative focuses less on the ethical challenges associated with AI than on how ethical codes can and should be developed. A comprehensive overview of ethical codes as well as links and documents on the creation of ethical codes are offered online.
• Partnership on AI: Published: September 2016
  URL: https://www.partnershiponai.org/
  Summary: The Partnership on AI is one of the largest AI initiatives with over 80 partners from different industries. It was founded in September 2016 by Facebook, Amazon and Google, among others. Together with the partners, its goal is to establish best practices for AI and to inform the population about the possibilities and scope of the technology. The partners have committed themselves to eight ethical principles.
• AI4ALL: Published: 2017
  URL: http://ai-4-all.org/
  Summary: AI4ALL is a non-profit education initiative founded at Stanford University in early 2017. Its aim is to increase diversity and inclusion in AI teaching, research, development and legislation through qualification and the establishment of role models. Its educational offerings are aimed specifically at population groups that are underrepresented at the university.
• Ethics and Governance of Artificial Intelligence: Published: 2017
  URL: https://aiethicsinitiative.org/
  Summary: The Harvard-based Ethics and Governance of Artificial Intelligence Initiative has been operating since 2017. It was founded in cooperation between the Berkman Klein Center and the MIT Media Lab with a USD 27 million grant from the Knight Foundation. It researches opportunities for private and public decision-makers to use AI in the public interest. In an open competition, the initiative offers a total of 750,000 USD for projects on four ethical challenges.
• AI Ethics Lab: Published: January 2018
  URL: http://aiethicslab.com
  Summary: The AI Ethics Lab was also founded at Harvard University in January 2018 and offers researchers and practitioners advice, training, workshops and lectures on the ethical implementation of AI. Its mapping workshop helps AI developers identify ethical sources of conflict in their projects and develop solutions without compromising technical integrity.
• MIT Stephen A. Schwarzman College of Computing: Published: October 2018
  URL: http://news.mit.edu/2018/mit-reshapes-itself-stephen-schwarzman-college-of-computing-1015
  Summary: In October 2018, the Massachusetts Institute of Technology (MIT) announced an investment of $1 billion to address global opportunities and challenges through AI. The focus is on the establishment of the Stephen A. Schwarzman College of Computing, named after the CEO of asset manager Blackstone, who personally donated USD 350 million. The promotion of ethics in AI will be an important part of the college’s activities. For example, new curricula are to be developed that combine AI with other disciplines, or forums are to be organized in which leading personalities from politics, business, research and journalism discuss the possible consequences of the latest AI advances and derive the need for action. The work is expected to begin in September 2019.

Tech companies

• Facebook: Published: May 2018
  URL: https://www.forbes.com/sites/samshead/2018/05/03/facebook-reportedly-has-a-dedicated-ai-ethics-team/#5690de1d2e5c
  Summary: Facebook announced at its F8 developer conference in May 2018 that it has formed its own Ethics Team. This team cooperates with development teams throughout the company and has developed the software Fairness Flow, which can detect bias of another AI. Fairness Flow has been integrated into FBLearner Flow, the software Facebook uses to train its algorithms. Scientific publications will follow. Facebook plans to further minimize the bias of its software by increasing the diversity of its workforce, for example by opening offices far from Silicon Valley. In January 2019, facebook announced that it would finance an ethics chair at the Technical University of Munich for an initial period of EUR 6.6 million over five years to develop ethical guidelines for AI, among other things.
• Google: Published: 2018
  Source: Responsible AI Practices
  URL: https://ai.google/education/responsible-ai-practices
  Summary: Google has already established an AI Ethics Board in 2014. Both members and tasks are unknown. In June 2018, Google published its Principles for Ethical AI and shortly thereafter made recommendations for Responsible AI Practices. In December 2018, Google announced a comprehensive set of measures via its blog. This includes training, an AI Ethics Speaker Series, a fairness module in the Machine Learning crash course, an interdisciplinary Responsible Innovation Team, a group of experienced senior consultants, and an advisory board of experienced executives for more complex issues involving multiple products and technologies.
• Microsoft: Published: 2017
  URL: https://www.microsoft.com/en-us/research/group/fate/
  Summary: Microsoft founded the Fairness Accountability Transparency and Ethics group (FATE) in 2017. The group aims to develop informatics methods that are both ethical and innovative and plans to publish about them. In the same year, AI and ethics in engineering and research (AETHER) was added, a company-wide recruited advisory board of experienced executives to identify and address ethical issues at an early stage. Microsoft published its own ethical principles in January 2018 and formulated ethical guidelines for the development of AI chatbots in November of the same year. In December 2018, ethical principles for the use of face recognition were added.

Others

• Ethically aligned design: Published: 2016
  Source: Ethically Aligned Design: A Vision for Prioritizing Human Well-being with Autonomous and Intelligent Systems
  URL: https://standards.ieee.org/industry-connections/ec/ead-v1.html
  Summary: The brochure Ethically Aligned Design: A Vision for Prioritizing Human Well-being with Autonomous and Intelligent Systems is based on the input of several hundred experts from six continents. The first version was released in 2016. A second version was released in 2017 with a request for feedback from the public. The revised version is to be finalized in 2019. It focuses on five principles that provide an ethical framework for the design, development and application of AI. The brochure is published by the Ethics in Action Initiative of the American Institute of Electrical and Electronics Engineers (IEEE), the world’s largest association of technical professions. Since June 2016, this association has also been working on the P7000 standard series, which develops process models for addressing ethical considerations in the conception and design of IT systems.
• Toronto Declaration: Published: May 2018
  Source: Toronto Declaration for Human Rights and Artificial Intelligence
  URL: https://www.accessnow.org/cms/assets/uploads/2018/08/The-Toronto-Declaration_ENG_08-2018.pdf
  Summary: At the RightsCon Conference in May 2018, Amnesty International and AccessNow initiated the Toronto Declaration for Human Rights and Artificial Intelligence. It calls for greater cooperation between the public and private sectors to safeguard human rights in the age of artificial intelligence and recalls, for example, the Vienna Declaration of the UN Human Rights Committee or the International Covenant on Economic, Social and Cultural Rights. Other supporters include Human Rights Watch and the Wikimedia Foundation.
• Montreal Declaration: Published: Dezember 2018
  Source: Montreal Declaration for Responsible Development of Artificial Intelligence
  URL: https://www.montrealdeclaration-responsibleai.com/the-declaration
  Summary:
  The Montreal Declaration for Responsible Development of Artificial Intelligence was developed by the Université de Montréal together with the Fonds de recherche du Québec. It was presented to the Society for Arts and Technology in December 2018. It is the result of more than a year of consultation with philosophers, sociologists, lawyers and AI researchers as well as online and face-to-face surveys of the population. Its ten principles are intended to provide an ethical framework for the development and application of AI, serve as orientation in the digital transformation of society, and provide a basis for further debate. Continuous revisions of the Declaration are also planned for the future.