liability issues in autonomous and semi-autonomous systems
TRANSCRIPT
Paste title slide graphics over this grey box in slide deck
Liability Issues in Autonomous and Semi-Autonomous systems
John Buyers & Dr Sanjeev Ahuja – ITechlaw World Conference, Miami, May 2016
osborneclarke.com
Introduction – AI – expectations vs. reality
osborneclarke.com Private & Confidential
3
Introduction – AI – expectations vs. reality
osborneclarke.com
Introduction – AI – Distinguishing Features
• AI – Distinguishing Features
• Neural Networks • Learning• Recognition (processing of external stimuli)• Decision making capabilities• Autonomy (or semi-autonomy) to act (or react)
following decisions• Creativity ?• Ability to communicate ?• Self-Awareness ?
“Any sufficiently advanced technology is indistinguishable from magic”
– Arthur C. Clarke
osborneclarke.com
Introduction – AI – Liability – a Sliding Scale
Vs. Vs.
osborneclarke.com
Introduction – AI – Existing Liability Frameworks
CurrentAI Liability
Product Liability
Contract Negligence (Tort)
Consumer Protection
Legislation (UK CPA 1987) (Strict
Liability)
osborneclarke.com
Introduction – AI – Future Liability Frameworks
Future AI Liability
Insurance Funded (Strict Liability)
Turing Registries
Individuation or Machine
Personhood
Agency Crime
osborneclarke.com
AI – Intelligent Machines as Complex ProductsDriverless Car Example
Intelligent Machines as complex products
• Features
• Multiple manufacturers• Sensors• Software• Many systems• Multiple points of failure
• Existing liability models work well where machine functions (and hence responses) can be directly traced to human design, programming and knowledge
osborneclarke.com
AI – Intelligent Machines as Complex ProductsDriverless Car Example
Intelligent Machines as complex products
• Potential targets for liability
Vehicle manufacturer Reseller Importer Components manufacturers Computer software provider Machine designer Components designer Machine "operator"
osborneclarke.com
Product Liability - Contract
• Ensures that manufacturer/retailer sells products that meet contractual standards
• Predominantly aimed at Pure Economic Loss (ie. monetary value)• Contract liability created by:
• Express terms (as to defects and/or warranties)• Implied terms as to quality and fitness for purpose (see in UK
Consumer Rights Act 2015/Sale of Goods Act 1979) – not defects per se
• Fault based liability – Claimant must prove:• Breach of express/implied term• Breach caused loss - causation (Hadley v. Baxendale)
• Remedies: Damages (put party in position had contract been performed properly)
osborneclarke.com
Product Liability - Contract
• Advantages:
• Can define the scope/boundaries of the contract obligation and hence the liability
• Implied terms
• Disadvantages:
• Claims limited by Contract Privity – ie. between the contract counterparties (with limited exceptions: see Contract (Rights of Third Parties) Act 1999)
• Different standards apply to UK consumer contracts and B2B contracts
osborneclarke.com
Product Liability - Tort
• Underpins “duty of care”• Can run concurrently with contractual duties – see Donoghue v. Stevenson
[1932] AC 562• Predominantly used for real (ie non-pecuniary) loss/damage• Fault based liability - Claimant must prove:
• Defendant owes a duty of care• Defendant failed in that standard of care• In that failure damage was caused (which was reasonably foreseeable)
• Remedies: Damages (put party in position had the tort not occurred)
osborneclarke.com
Product Liability - Tort
• Advantages:
• Exists independent of contract relationship – but need to prove “duty of care” subsists
• Disadvantages:
• No recovery for pure economic (ie financial) loss• Contributory negligence can reduce recovery• as can: volenti non fit injuria (voluntary assumption of risk)• Evidence of “fault” normally subsists with Defendant
osborneclarke.com
Product Liability – Consumer Protection Act 1987
• Implements Directive 85/374/EC on Liability for Defective Products• Introduces “strict” product liability regime for defective products in UK• Does not affect availability of Contractual or Tort based remedies
Scope:• A person who is injured, or whose personal property is damaged by a
product can claim against the manufacturer or supplier if it can be shown if the product is defective
• No requirement to show fault but burden of proof on claimant to show defect existed
Defects• Defect exists where “the safety of the product is not such as persons
generally are entitled to expect” – Section 3(1) CPA 1987
osborneclarke.com
Product Liability – Consumer Protection Act 1987
• Advantages:
• No requirement to show fault• No privity• Wide potential selection of liability targets – suppliers/manufacturers
• Disadvantages:
• Still causation issues – need to show defect• No actions for pure economic loss (again)• Software/Intellectual products – unclear whether these are covered
by the Directive/CPA• Development Risks Defence
osborneclarke.com
AI – Intelligent Machines as Complex ProductsDriverless Car Example
Target Contract Tort CPA (Product Liability)
Vehicle Manufacturer Only if you have contracted directly
Only if you can prove negligence
Only if you can prove “fault”
Reseller Only if you have contracted directly
Unlikely Only if you can prove “fault”
Importer Only if you have contracted directly
Unlikely Only if you can prove “fault”
Components Manufacturers No – may be liable under indemnity to your counterparty
Maybe under contributory action
No
Computer Software Provider No (unless you have licensed software directly)
Only if you can prove negligence – otherwise possibly under contributory action
No
Machine Designer/Components Designer
No– may be liable under indemnity to your counterparty
Only if you can prove negligence – otherwise possibly under contributory action
Only if the designer is the manufacturer
Machine Operator No Only if you can prove negligence
No
osborneclarke.com
AI – A failure in causation ?
Intelligent Machines as complex products
• All of the following scenarios allow conventional product liability analysis:
• Traceable defects (in systems, software)• machine decisions that are based on defective programming• Failure to provide correct operating instructions • Incorrect operation (if relevant)
• But what about inexplicable failures ?
osborneclarke.com
AI – A failure in causation ?
Tort – A partial Solution:
• res ipsa loquitur – "the thing speaks for itself"• does not ordinarily occur without negligence• caused by an agency within the exclusive control of the defendant• not due to any voluntary action or contribution on part of plaintiff• Defendant's non-negligent explanation doesn’t completely explain injury
• Toyota Motor Corporation (2013) WL 5763178 (Texas)• multiple successive similar "inexplicable" failures
CPA – Another partial fix
• Consumer expectations test moderates causation
osborneclarke.com
Insurance model
But what about inexplicable isolated incidents?
• Existing legal models (including causation) fail• Argument for introduction of strict liability model
• See Accident Compensation Act 1973 – New Zealand• Increased complexity means increased costs• costs of litigating complex autonomous systems – could be
better spent on compensating victims rather than lawyers and expert witnesses
• Incentivises stable environment for these systems to be developed (rather than stifling innovation)
osborneclarke.com
Insurance model
Turing Registries as a basis for AI liability?
• Submit intelligent machines to certification process• Quantify "risk" certification based on spectrum:
• Higher intelligence, higher autonomy, greater consequences, higher premium
• Pay premium for certification• Premium funds risk "common pool"• No dealing with "uncertified" AI: system becomes self-fulfilling
• Similar, but not the same as traditional insurance• Removes causation, proximate cause and allows for wilful
acts of AI (usually excluded)
osborneclarke.com
Individuation – Intelligent Machines as “persons”
• Machines as “agents” of human principals• Criminal liability consequences?• Turing tests• How to define machine personhood: some contemporary and historical
analogues• Legal persons: Corporations • Animals • Children • Slaves ? “One who Is under the power of a master, and who belongs to
him; so that the master may sell and dispose of his person, of his industry, and of his labor, without his being able to do anything, have anything, or acquire anything, but what must belong to his master.” Civ. Code La. (1825) Art. 35
osborneclarke.com
Conclusion
“In law, a person is legally liable when he/she is financially and legally responsible for something.”
Wikipedia
“If we want to avoid the injustice of holding men responsible for actions of machines over which they could not have sufficient control, we must find a way to address the responsibility gap in moral practice and legislation”
The Responsibility Gap, A. Matthias
Paste end slide graphics over this grey box in slide deck
Thank you