Meeting with – Senior Scientist at Google Research and PAIR (CDMA 06/144, 13 October 2017, 10:00 – 11:30) Participants: GOOGLE: (Senior Staff Research Scientist and member of the Google Brain Team) (Public Policy and Government Relations Analyst at Google) JRC: SUCHA Vladimir DELGADO SANCHO Luis, CHIRONDOJAN Dan, Chef de File: A.2 Contribution: DIR A, DIR B, DIR E, DIR I Authorised by: , A.2

1. STEERING BRIEF 1.1 Scene setter You are meeting Senior Research Scientist and member of the Google Brain Team. (Full bio and PAIR overview in background). will be accompanied by Relations Analyst at Google. , Public Policy and Government 1.1.1 Google's People + AI Research initiative (PAIR) The goal of PAIR is to focus on the "human side" of AI: the relationship between users and technology, the new applications it enables, and how to make it broadly inclusive. The project foresees three main lines of research: finding the educational materials and practical tools that engineers would need to build and understand machine learning systems, broadening the application of AI in augmented reality for professional experts, exploring the opportunities that design thinking could open to AI. The project shall also open-source new tools and educational material. 1.1.2 AI at the JRC Current areas in which the JRC makes use of AI are biometrics, fight against counterfeiting, agriculture, earth observation, epidemiology, migration, cyber- security, media monitoring and open source intelligence for law enforcement and tax authorities. The JRC has also made use of Neural Networks in the process of modelling and surveillance of Nuclear Fuel reprocessing, studying digital forensic investigation techniques for law enforcement, and exploring the interconnectedness and evolution of the cyberspace of spatial data infrastructures (INSPIRE, GEOSS) to improve data management, data access and usability. The JRC has now broadened its outlook on AI to study the impact of AI and of the Digital Transformation in general on economy, society and human/machine interactions. 1.1.3 EU Policy context EU policy on Artificial Intelligence and Robotics was given a sudden shake in February 2017 when the EP adopted in plenary a legislative initiative report initiated by the JURI committee on the Civil Law rules on Robotics. The EP requested COM to come forward with a legislative proposal on civil liability for damage caused by robots. It 2

also recommended to comply with the precautionary principle when moving technologies from research to real life testing, and then to the market. It requested COM to explore the suitability of a specific legal status for robots (at least for the most sophisticated autonomous ones). Directly addressed by the EP report, . DG JUST has launched a public consultation (in cooperation with DG GROW and DG CNECT). . For its part, DG CNECT has established in 2014 a Public Private Partnership for Robotics in Europe – SPARCS – with a combined funding of 2.1 billion euro. DG CNECT and DG JUST are jointly preparing a major initiative for 2018 in view of preparing a "European comprehensive strategy which will have to encompass three elements: 1) Building up Europe's technological and industrial capacity in AI and facilitating its uptake, 2) Addressing new ethical, societal and legal issues, 3) Tackling emerging socio-economic challenges". The conclusions from the Tallinn Digital Summit of last week, which are going to be discussed by the European Union leaders at the next European Council taking place on 19-20 October in Brussels, highlighted the necessity to prepare "agile and tech- friendly regulation [to] contribute to making Europe an attractive headquarters for new and growing companies to accelerate the digital transformation of industries through an uptake of the latest technologies, including artificial intelligence, big data processing and blockchain." The draft European Council conclusions for 19 October state: • "An openness to address emerging trends: this includes issues such as artificial intelligence or block chain technologies, while at the same time ensuring a high level of data protection, digital rights and ethical standards. The European Council invites the Commission to put forward a European approach to artificial intelligence by January 2018;" 1.2 Objectives • 3

• 1.3 Line to take • 4

2. OTHER QUESTIONS TO • 3. BACKGROUND INFORMATION 3.1 Biography 5

3.2 Google PAIR = People + Artificial Intelligence Research (Text copied from https://www.blog.google/topics/machine-learning/pair-people-ai- research-initiative/ ) Google has announced on 10 July 2017 the People + AI Research initiative (PAIR) which brings together researchers across Google to study and redesign the ways people interact with AI systems. The goal of PAIR is to focus on the "human side" of AI: the relationship between users and technology, the new applications it enables, and how to make it broadly inclusive. The goal isn’t just to publish research; we’re also releasing open source tools for researchers and other experts to use. PAIR's research is divided into three areas, based on different user needs: • Engineers and researchers: AI is built by people. How might we make it easier for engineers to build and understand machine learning systems? What educational materials and practical tools do they need? • Domain experts: How can AI aid and augment professionals in their work? How might we support doctors, technicians, designers, farmers, and musicians as they increasingly use AI? • Everyday users: How might we ensure machine learning is inclusive, so everyone can benefit from breakthroughs in AI? Can design thinking open up entirely new AI applications? Can we democratize the technology behind AI? One key to these puzzles is design thinking. Instead of viewing AI purely as a technology, what if we imagine it as a material to design with? Here history might serve as a guide: For instance, advances in computer graphics meant more than better ways of drawing pictures—and that led to completely new kinds of interfaces and applications. For example an effort called “human-centred machine learning” (HCML) recently initiated at Google looks across products to see how ML can stay grounded in human needs while solving them in unique ways only possible through ML. Google is open sourcing new tools, creating educational materials (such as guidelines for designing AI interfaces), and publishing research to answer these questions and spread the power of AI to as many people as possible. Many designers and academics have started exploring human/AI interaction. Google see community-building and research support as an essential part of our mission. 6

They are working with a pair of visiting academics— —who are focusing on education and science in the age of AI. Focusing on the human element in AI brings new possibilities into view. 3.3 AI in Europe – Global figures Europe is particularly strong in a number of AI technologies such as cooperating robots; speech and haptics-based human-machine interface, knowledge extraction and semantics, learning, safety, actuation (other than gears), grippers and dexterous hands, locomotion (except bipedal), navigation and collision avoidance, motion and task planning, modelling for control, bio-inspired systems, bionics, and cybernetics. Europe produces around a third of the world's industrial robots and half of the world's professional service robots (precision farming, security, health, logistics, etc.) which are all increasingly relying on AI. Europe is also home to three of the world's largest producer of industrial robots, which are KUKA, ABB and Comau. It has a long tail of mid-sized companies and SMEs that provide various components for robots from sophisticated control software to sensors, actuators and grippers. Europe's strength is also demonstrated by European companies being attractive takeover targets by worldwide competitors: • • • • • Google (US) bought Deepmind (UK) and Moodstocks (France) Teradyne (US) bought Universal Robots (DK) Intel (US) bought Ascending Technologies (Germany) Softbank (Japan) took over ARM (UK) and Aldebaran Robotics (France) Midea (China) took over majority of capital on KUKA (Germany) While Europe is strong in basic and robotics-related AI technology, one of Europe's key weaknesses is the dominance of non-European players in the burgeoning field of AI platforms. Their competitive advantage derives in no small part from the simultaneous access to large quantities of data that can be used to refine the quality of the results. Another of Europe's weakness in general is commercially exploiting research results. Europe is also less strong in creating critical mass, scaling up and pooling data resources the way US giants (e.g. Google) can do. Fragmented legal systems also create many obstacles, e.g. to free flow of data. 3.4 Centre for Advanced Studies Projects on Digital Transformation 3.4.1 Digital transformation and Machine intelligence and human behaviour (Lead scientist: ) 7

• Context: The interaction between infinitely scalable machine intelligence and limited human cognitive capacities results in a changing cognitive balance. Machine learning on the one hand can provide cognitive assistance in the assessment of the information, but on the other hand it may impact individual and personal decision making and behavioural autonomy and responsibility. In particular, the data and behavioural algorithms shaping the context within which we make choices are mostly owned by private companies whose behavioural motives may not necessarily coincide with individual motives and welfare. The prospect of overcoming the limitations of traditional sensory communication interfaces will push this frontier even further. The objectives of the project are to investigate (a) how and under which conditions this change in cognitive balance may improve or worsen private and societal welfare, and (b) how machine-intelligence driven changes in the gap between private and social welfare may require changes in policy. • Focus of project: to assess the social, economic, legal and regulatory impact of machine intelligence (MI) on society. The project does not seek to build MI algorithms (no coding) but to examine the implications of MI for individual and collective human behaviour. A good understanding of MI technology is required and should be combined with inputs from other social and cognitive sciences to grasp the implications for the welfare of consumers and firms and the impact on social and political institutions in society. Hence a multi-disciplinary approach is needed. The lead scientist will build up a multi-disciplinary team of computer scientists, social scientists and economists, cognitive scientists, etc., some of them on a contractual basis others as associates in a network of external researchers. The work is expected to start in January 2018. 3.4.2 Digital transformation and Governance of Human Societies (Experts: ) • Context: The project is going to look at the future governance of a digitally transformed European society in a global context and explore the transformations taking place in the adoption and adaptation of digital technologies in public institutions among others. The impacts of changes driven by the digital transformation stretch beyond the economy and include individual cognitive and psychological effects and changes in the economic, social and political institutions that govern collective behaviour. The evolution of digital technologies is expected to trigger deep transformations in public and private institutions at multiple levels, and in the organisational fabric of society. Research on these transformations so far has concentrated at the level of individual organisation or institution. However, a conceptual framework for understanding the interactions between digital technologies, new information flows and institutional change or changes in collective behaviour, is still lacking. 8

• Focus of project: to study how a deep understanding of digital transformation can help policy-makers address the challenges facing EU society over the next decades. Two external experts are going to lead the project with the contribution of social scientists, computers scientists and the relevant industry representatives. The project will contribute to the understanding these interactions and their implications for governance of human societies in the early stages of the digital transformation is of major social and political relevance in order to effectively manage both positive and negative impacts today and in the future. The aim of the project is to investigate how changes in information technology trigger major changes in political institutions and the organisation of society. In particular, it aims to understand how digital technologies and the shift in data collection, analysis, and knowledge about society from the public to the private sector affect institutional architectures and governance at all levels. 3.5 AI Techniques in use at the JRC . Here are some examples of how AI techniques are used in Big Data: • Decision trees • Neural networks • Support Vector Machines • Symbolic Machine Learning. AI techniques used in EMM (and TIM): • Statistical Machine Translation (using the third-party Moses software). • Neural Machine Translation, using Deep Learning techniques, is currently being explored. • Neural networks for the classification of Named Entities (using Perceptron and Xception). • Support Vector Machines (SVM) and Naïve Bayes for various classification purposes. • Grammar induction: Learning of rules for Information Extraction, used in the Event Extraction work, and more. • Knowledge representation and simple inference rules used in event extraction. • Learning of related words/dictionaries (with the JRC-developed OntoPopulis software) using distributional similarity and more, used as part of many EMM components. • Weighted edit distance in string distance similarity calculation. 9

• Distant learning, Hidden Markov models, neural networks, decision trees, logistic regression and clustering for various applications. AI techniques used in Cyber-Security, Privacy and Digital Identity: • Automatic behavioural analysis of software applications (included mobile apps) and IoT • Anomaly detection and malware characterisation • Contextual authentication and identification • Network analysis • Critical state analysis in Industrial Control Systems and Energy Smart-Grids For 2018 the JRC will explore the use of ML and AI also in the online privacy and data protection fields and in the distributed ledger domain. Within the same technological area, but on the law-enforcement side, the JRC Digital Forensic Investigation Techniques for Law Enforcement use of ML techniques to develop digital forensics tools to support law-enforcers for what concerns image recognition, biometrics and network analysis. 3.6 JRC and Google collaboration on the Global Human Settlement Layer There is a strong overlap between the GHSL activities and Google Earth Engine (GEE). Both are using big data and rely on advanced machine learning technology. There are already well established contacts between the teams: • • • , discussed the potential of porting the Symbolic Machine Learning classifier, developed by the GHSL project, into GEE and the possibility of adding it to the machine learning classifiers available in the GEE algorithms. The outcome of the discussion was that the porting requires some additional developments from GEE engineers in order to manage the specific encoded data structure generated by the SML. are discussing the upload of the GHSL data into GEE. Technical discussions related to the data structure are on-going. The data analysis for the 'Monitoring the Syrian Humanitarian Crisis with the JRC’s Global Human Settlement Layer and Night-Time Satellite Data' Report (JRC101733) was entirely implemented in GEE. 3.7 SciArt - Resonances III The next Resonances will be on the topic of Big Data. A first preparatory workshop takes place on 20th October. 10