Liability for a conduct of Artificial Intelligence VERONIKA ŽOLNERČÍKOVÁ INSTITUTE OF LAW AND TECHNOLOGY, FACULTY OF LAW, MASARYK UNIVERSITY Artificial Intelligence / Autonomous Machines uTERMINOLOGY uBASIC INFORMATION ON FUNCTION OF AI uEXAMPLES Characteristics of an AI driven machine Autonomous Agent uAutonomous behaviour – AI is further capable of gathering information through its own sensors or by data exchange and of making independent decisions based on that information, all without further human input uAutonomous agent - a software entity (goal-oriented) capable of executing actions in its environment: virtual, physical uExamples of autonomous agents: autonomous intelligent cars (AIC), drones/unmanned aircraft vehicles (UAV), satellites uSources of information: sensors, cameras, data connection, lasers u u Autonomous / Driverless Vehicles General Classification of Autonomous Agents I. uIs the classification used for AIC applicable for other autonomous agents? uDifferent level of: uconnectivity (interaction, user input) umobility (virtual x physical, means of movement) uintelligence (programming, machine learning) uVariability in data input usensors udata exchange uproducts u Level of Automation/Autonomy in Autonomous Intelligent Cars General Classification of Autonomous Agents II. uFailure of AIC on one level of autonomy will not be the same as the failure of another AIC on the same level uLevel of autonomy as a useful concept for classification but not for liability distribution uUAVs, planes, ships – similar autonomy level concepts (applicable), yet different purpose, users uPossible to compare liability concepts according to certain features used in autonomous agents u Overview of Legal Materials & Preparation Works uEUROPEAN UNION uCOUNCIL OF EUROPE uCZECH REPUBLIC European Union Council of Europe Czech Republic LIABILITY AS THE KEY ISSUE uWHY IS LIABILITY AN ISSUE IN AI uWHAT ARE THE POSSIBLE SOLUTIONS uLIABILITY CONCEPTS Liability as the Key Legal Issue in AI uActions are carried out by the machine independently (liability for damages) uSoftware x physical components (preventive liability) uSoftware service – level x Driving school for humans uChanges in the code, unpredictability u Liability for Emerging Digital Technologies uCommission Staff Working Document - Liability for emerging digital technologies uAccompanying the document Communication from the Commission to the European Parliament, the European Council, the Council, the European Economic and Social Committee and the Committee of the Regions - Artificial intelligence for Europe uIssued on 25th of April 2018 u uhttps://ec.europa.eu/digital-single-market/en/news/european-commission-staff-working-document-liab ility-emerging-digital-technologies Liability Principles uLiability - responsibility of one party for harm or damage caused to another party, may be a cause for compensation, financially or otherwise, by the former to the latter uCivil law x Administrative law x Criminal law uContractual x Extracontractual uFault-based (subjective) x Strict-based (objective) Principles of European Tort Law uNon – harmonized accross EU uThe European Group on Tort Law - a group of scholars in the area of tort law uPrinciples of European Tort Law (PETL) similar to the Principles of European Contract Law: http://civil.udg.edu/php/biblioteca/items/283/PETL.pdf uAttribution to a concrete person (PETL): uWhose conduct have caused the damage uWhose abnormally dangerous activity has caused the damage uWhose auxiliary has caused the damage within the scopes of its functions u Liability Principles in Civil Law u1) Illegal action (tort) u2) Damage u3) Causal Nexus u4) Fault (intent, negligence) u uIn strict – based liability fault is not one of the conditions – causal nexus – the element present in every liability principle uCausal nexus – causa / conditio sine qua non / but – for principle – would the negative effect appear without the conduct of the defendant? u Causal nexus for the conduct of AI u uLower ability to control an autonomous machine – does not follow the will of the driver, decision making is not fully predictable (even by the programmer) as the machine learns itself while in function uInherent level of uncertainty in software „The long-term operation of complex systems entails a fundamental uncertainty, especially in the context of complex environments, including new or unpredictable environments“ u Causal nexus for the conduct of distributed AI uIn complex processing environment, it is not possible to simply impose liability on any identifiable unit – multiple units means: multiple programmers, vast number of users, unpredictable number of interactions between the machines on number or platforms, operating systems, data exchanges, telecommunications programmes u Different Approaches to Causation uFull liability x Proportional liability uMulticausual damage: uAlternative causes uConcurrent causes uCumulative causes uThree types of approaches towards causation in Europe uOverarching uBounded uPragmatic u u Liability Concepts uOptions : keep, discard, amend uNo-fault/strict (product/design defect) liability uRisk based liability uQuick claim resolution in certain scenarios uLegal personhood (subjectivity) u Technical Standardisation uSTANDARDISATION AS A PREVENTIVE MEASURE uSTANDARDS VS. LEGAL REGULATION uHOMOLOGATION Technical Standards & Product Safety uPart of preventive liability concept uWay of ensuring product safety uDirective 2001/95/EC of the European Parliament and of the Council of 3 December 2001 on general product safety uSpecific product safety rules: dangerous goods, vulnerability of consumers, need for compatibility uInvolvement of all parties – manufacturers, distributors, users, law-makers uTechnical standards – description of a product from a technical perspective, construction, materials and other criteria uHomologation – when a product needs prior certification before it can be available on the market (motor and aerial vehicles) u Technical Standards uAlthough normative, do not have legal nature uRecommendation for manufacturers, best practice uCan be binding, if referenced in law – often published as subordinate legislation uAdopted by specialized authorities – ISO, NIST, ETSI… u uLegislation designates which products must be safe and technical standards determines how to achieve it. uTechnical standards as a necessary addendum to regulatory approach u u u u Missing Role Model uTechnical standards in the field of avionics – are they usable for other technologies, f.e. autonomous vehicles? uSoftware in autonomous vehicles has different challenges uPedestrian crossings, objects in the road, other vehicles uChanging traffic conditions u„Piloted“ by a citizen, consumer, not by a professional uMultitude of sensors – radar, lidar, camera uTechnical standards for autonomous trains – similar problems Testing / Homologation Methods Possible Solutions uPreventive measures – testing, standardisation, homologation uMandatory insurance (common insurance – predictability of risks) uExplainable AI uReallocation of the burden from the victim towards the person with most information uProximal causation - Harmonisation of criteria (at least doctrinally) for proving of causal nexus across EU uCompensation fund/ liability fund How to Regulate AI uUNANSWERED QUESTIONS uMETHODOLOGY New Technologies & the Law uChallenges of new technologies uMultidisciplinary approach is needed to comprehend how the technology works uMultidisciplinary approach is to be taken also in the field of law itself uRapid development, unpredictability uSolutions? uConcentration on the content and the purpose of legal regulation uThe use of analogy is needed uIt is impossible to regulate the current phenomena as well as the future ones Artificial Intelligence as the New Legal Challenge Questions? The Oatmeal Thank you for your attention! uzolnercv@mail.muni.cz