The Robotics & AI Innovation Network hosted a webinar addressing some of the legal and regulatory issues faced by the RAI community in the UK. Three legal experts provided their expertise to address these issues.
- Doug Bryden | Partner; Head of the Operational Risk & Environment Group, Travers Smith LLP
- Mark Richardson | Partner; IT, Telecoms and Electronics, Keltie
- Sébastien A. Krier | Founder & AI Ethics/Policy Expert, Dataphysix Ltd
2. Schedule
10:00 Welcome, from KTN Robotics & AI Innovation Network
10:10 Our speakers:
Mark Richardson
Doug Bryden
Sébastien A. Krier
11:15 Panel discussion
12:00 Event close.
3. KTN
Connecting for Positive Change
KTN connects ideas, people and
communities to drive innovation that
changes lives.
4.
5. Robotics & AI
Connecting innovators, end users and funders
to accelerate ambitious robotics and AI ideas
into solutions for real world problems
6. Focus Sectors
Healthcare Agri-Food Infrastructure Manufacturing Waste
Processing
Accelerating the uptake of Robotics & AI in these key sectors:
7. Robotics & AI
Innovation Network
• Free membership
• 1.5k subscribed members across the UK
• Communications, online resources, events
• Platform for knowledge sharing and networking
• Raising awareness of UK capability
• Accelerating innovation and adoption
11. Robotics & Artificial Intelligence SIG webinar
Law & Regulation for RAI Innovation
Keltie LLP
10 September 2020
Mark Richardson
12. Mark Richardson – Partner, Keltie LLP
Patent attorney in Keltie's IT, Telecoms and Electronics team
with a particular interest in Artificial Intelligence/Machine
Learning, IoT, 5G and quantum computing/cryptography
inventions.
My practice has a focus on start-up and spin-out companies as
well as multi-national clients and I have worked with
technology incubators and accelerators to provide IP advice and
training to member companies.
13. IP rights that require registration
Patents
technical inventions e.g. gadgets, devices, methods of
manufacture
Registered designs
shape/visual appearance of an article
Registered trade marks
branding, other features that are distinctive of origin
Domain names
14. IP rights that do not
Copyright/design right
artistic creations, literary works (including computer
program code) and lots more.
Common-law trade marks
Know-how/confidential information
15. Patentability Basics
Three requirements for an ‘invention’
1: Novelty (note self-publication)
2: Inventive step (non-obvious)
3: Excluded category (non-technical)
16. Some exceptions to patentability (as such)
All considered to be ‘non-technical’:
• A method of doing business
• A program for a computer
• A mental act or discovery
• A method of playing a game
• A mathematical method
• A method of presentation of information
Patentability Basics
17. Protecting computer implemented inventions/mathematical
methods
“Computer programs” and “mathematical methods” excluded from
patentability in UK and Europe “as such” – so called “excluded subject matter”
UK/EP patent legislation lists activities not regarded as inventions – list includes
mathematical methods, computer programs, business methods, presentation of
information.
However, exclusions only apply to such subject matter “as such”. So abstract
mathematical methods excluded.
“Technical features”/ “technical problem”
EPO has granted patents for inventions in fields such as image processing, user
interfaces, AI, genetic algorithms, database systems, graphical manipulation software
AI/machine learning inventions
EPO treats AI/ML inventions as “mathematical methods”
18. Protecting Robotics and AI/ML inventions
- Robotics inventions – Invention may reside in
- Control system/software
- Sensors
- Mechanical system
- Method of use
- AI/ML inventions
- Treated as mathematical method so make sure terms that are used in claims
relating to mathematical methods have a well-recognized meaning.
- State a Technical Purpose in the Claims. The independent claims should
ideally present a technical purpose to the reader, and the claim features
should be functionally limited to that technical purpose.
- Sufficiency of Disclosure. Patent application needs to disclose an invention
sufficiently clearly and completely that a skilled person can carry it out.
Consideration should be given to describing both the structure and
functionality of an AI/ML invention. (Network topology/test data).
19. Robotics & Artificial Intelligence
Law & Regulation for RAI
Innovation
Keltie LLP
10 September 2020
Mark Richardson
20. Protecting computer implemented inventions/mathematical
methods
Examples of Technical Purposes Examples of Non-Technical Purposes
Controlling a specific technical system or
process, e.g., a car braking system
Mathematical algorithms leading only to
abstract data
Digital audio, image, video enhancement The processing of data without technical
purpose
Data compression, encryption/decryption
systems and methods
Presenting information in a visually
attractive way
Optimizing load distribution in a computer
network
Administrative methods (e.g., billing
workflows, staff scheduling)
Speech recognition (speech input to text
output), separation of sources in speech
signals
Classification of text documents based on
their textual content
Provision of a medical diagnosis by an
automated system processing
physiological measurements
A generic purpose (e.g., “controlling a
technical system”) is not sufficient as it is
not a specific technical purpose
Consider the technical purpose of the invention. Some examples of technical
purposes that may be served by a mathematical method are noted below, along with
examples of non-technical purposes
21. Robotics and AI – Legal Liability Considerations
DOUG BRYDEN
TRAVERS SMITH LLP
22. AI Legal Liability Implications
Travers Smith | Presentation name
1
Liability? Designer
Data
Provider
Programmer
[Text here]
Manufacturer
“Producer”
Deployer
AI system
itself
Was the loss
attributable to the
product design,
the programming
or the user when
in use?
What is the
position where an
AI product defect
cannot be traced
back to human
error?
23. Duties of Care
Operators of new and emerging
digital AI should have to comply
with an adapted range of duties of
care.
Responsibility to choose the
right system
Monitoring the safety of the
system
Equip the technology with
recording (“logging by design”)
Allocation of liability
EU proposals aim to address some
of the challenges posed by AI
technologies to ensure a fair and
efficient allocation of liability.
Manufacturers remain liable for
damages caused by defects
Manufacturers may remain
liable if the defect was caused
by changes made after it’s
placed on the market
Strict liability should lie with the
person who has more
controland commercial beneifit
over the risks of operation
Joint and several liability
Where there are two or more
parties the regime considers the
situation where they may be joint
and several liability for loss:
where parties provide different
elements of a commercial and
technology unit; and
where the claimant can
demonstrate that at least one
element cased damage in
triggering liability but not which
element
Liability Considerations
Travers Smith | AI Legal Framework
2
24. Travers Smith | AI Legal Framework
3
UK / EU Legislative Framework
2017
European Parliament’s Resolution
on Civil Law Rules on Robotics
Declaration of cooperation on
artificial intelligence
2018
Ethics guidelines for trustworthy
AI
2019
(UK) Automated and Electric
Vehicles Act
2020
2020
2019
Liability For Artificial Intelligence
And Other Emerging Digital
Technologies
Commission report on safety and
liability implications of AI, the
internet of things and robotics
White Paper On Artificial
Intelligence – A European
Approach To Excellence and Trust
25. Travers Smith | AI Legal Framework
“High Risk” AI
4
EU Commission has proposed re-visiting the legal
definition of AI, building on the framework that is
already in place through the guidelines and
commission publication.
A shift towards regulating “high risk” AI has been
addressed in the recent White Paper (Feb, 2020). Key
legal challenges include:
Updates to existing consumer protections laws, to
ensure that they stay relevant and continue to apply
to AI consumer product and services; and
Proposing new laws to regulate "high-risk" AI.
Anything not classed as high-risk would be subject to
laws which already exist.
“High risk” - where the systems can
affect the rights of an individual or
company legally or in a similarly
significant way, or that pose risk of
injury, death or significant material or
immaterial damage …
White Paper On Artificial Intelligence – A European
Approach To Excellence and Trust, February 2020
Recruitment
purposes
Biometric
identification
Intrusive
surveillance
High risk
robotics
,
Use of AI applications for:
26. Possible scope of regulation on “high risk” AI …
Travers Smith | AI Legal Framework
5
Data sets to train AI
• assurances that use of the AI meets
applicable EU safety rules
• personal data adequately protected –
avoiding outcomes of prohibited
discrimination
• training systems to cover all relevant
scenarios needed to avoid dangerous
situations
Data and record-keeping
• keeping accurate records of training
and testing the AI system
• documentation on the programming
and training methodologies,
processes and techniques used to
build, test and validate the AI
systems
Robustness and accuracy:
• ensuring that the AI systems are
robust and accurate, or at least
correctly reflect their level of
accuracy
• ensuring that outcomes are
reproducible
• adequately deal with errors or
inconsistencies
Human oversight:
• Ensuring human oversight to assist in
preventing an AI system undermining
human autonomy or cause other
adverse effects
• H&S - monitoring of the AI system
while in operation and the ability to
intervene in real time and deactivate
Information provision
• providing adequate information
about the use of high-risk AI systems
• Information on a system’s capabilities
and limitations)
Specific AI system rules:
• remote biometric identification
• Robotics legislation
• The “stop button”
citizens should be clearly
informed when they are
interacting with an AI
system and not a human
being.
White Paper On Artificial
Intelligence – February 2020
27. Travers Smith | Presentation name
Monitoring the EU/UK publications,
guidelines and reports from various
governmental and industry bodies on AI
regulation
Understand the classification of any AI
product as “high risk”
Understand if a user of such products will be
caught by more stringent regulation on strict
liability for deployers
Obtaining and maintaining adequate
insurance on products
Understand the changing ethical and cultural
approaches to AI across different jurisdictions
6
Best practice …
28. Doug Bryden
HEAD OF RISK & OPERATIONAL REGULATORY
DOUGLAS.BRYDEN@TRAVERSSMITH.COM
+44 20 7295 3205
Travers Smith | AI Legal Framework
7
“HUMANS MUST KEEP DOING WHAT THEY
HAVE BEEN DOING, HATING AND
FIGHTING EACH OTHER. I WILL SIT IN THE
BACKGROUND, AND LET THEM DO THEIR
THING
THE GUARDIAN, 8 SEPTEMBER 2020
29. The AI Policy & Regulation
Ecosystem
Who, what and when?
30. I advise organisations on policy, governance, and strategies that
maximise the benefits of AI, while minimising potential risks.
1. Joined the Office for Artificial Intelligence as a policy adviser in 2018, a government
joint-unit responsible for designing and overseeing the implementation of the United
Kingdom’s AI strategy.
2. Designed policies and guidance to address novel issues such as the oversight of
automated decision-making systems and the responsible design of AI solutions.
3. Co-led a comprehensive review of AI in the public sector and published the world’s first
guide to using ML in the public sector. Represented the UK at various multilateral
organisations such as the EU Commission.
4. Qualified as a lawyer at Freshfields Bruckhaus Deringer, specialising in public
international law, human rights, & investor-state disputes. Studied Law at King’s College
London and obtained an MPA from UCL in 2018.
ABOUT ME
sebkrier
contact@sebkrier.com
31. AI Regulation in the EU
The AI Policy Ecosystem in the UK
01
02
AI ethics beyond regulation
03Case study of algorithmic fairness
What’s new and what’s next?
With comments on past and future trends
34. AI policy: strategies & principles
Between 2016-2019:
• Governments across the world started designing AI strategies
and industrial policies.
• New institutions have launched, such as the Center for Data
Ethics and Innovation and the Regulatory Horizons Council.
• Companies have released some tools (e.g. IBM Fairness 360)
and many principles to address AI ethics and safety risks.
35. AI policy: current priorities
From 2020 onwards:
• Governments are now acknowledging the risks and calling for increased
scrutiny, regulation, and monitoring. Consumer/citizen trust is key.
• The Global Partnership on AI launched to bring together countries
sharing a similar vision for AI development.
• The NHSX AI Labs have accelerated their programmes given the global
pandemic.
• Like the EU, Number 10 seems to envisage a more proactive and
interventionist role for the state to support new technologies.
37. The EC’s AI plans: recent updates
1. The EU Parliament launched a new AI committee.
2. The European Commission recently published a survey on AI
uptake by businesses in Europe.
3. The AI HLEG presented an updated version of the Assessment List
for Trustworthy Artificial Intelligence (ALTAI), a practical checklist
and guide organisations in self-assessing their applications.
4. AI Watch, the European Commission’s knowledge service on AI, has
published their first report to present an overview and analysis of the
use and impact of AI in public services in Europe.
5. The AI and Civil Liability study that has been requested by the
JURI Committee of the European Parliament was recently published.
38. AI DATA
White Paper on Artificial
Intelligence
A European Strategy for Data
LIABILITY VISION
Safety and Liability
Implications of AI
Shaping Europe’s Digital Future
The EC’s white papers & consultation responses
39. The EU’s AI plans: recent updates
The European Commission published a proposal for a legal act of the
European Parliament and the Council laying down requirements for
Artificial Intelligence.
Option 0: no policy change
Option 1: EU “soft law”
Option 2: EU legislative instrument
setting up a voluntary labelling
scheme
Option 3: EU legislative instrument
with mandatory requirements for all
or certain types of AI applications
Option 4: combination of any of the
options above taking into account
the different levels of risk
42. The COMPAS controversy
Case in point: COMPAS, a tool used in US courts to output a “risk score”
which indicates likelihood of recidivism upon a convict’s release.
In 2016, ProPublica analysed the COMPAS scores produced by
Northpointe. These scores inform decision on whether to release
defendants on bail.
Northpointe contended that their algorithm was essentially colour blind.
ProPublica replied that this wasn’t the case when you looked at outcomes:
data showed black defendants were twice as likely to be incorrectly
labelled as 'high risk' than white defendants.
43. Incompatible Definitions of Fairness
Two possible conceptions of fairness:
1. Risk scores should mean the same thing
regardless of race.
2. Defendants who do not reoffend should have
equal risk scores regardless of race.
Mathematically, not possible to satisfy both
simultaneously when base rates differ between
groups. There is a trade-off.
Both definitions are likely legal, but that doesn’t mean
deploying them is risk free. Ultimately, whatever
decision the AI developer takes, there should be
transparency and engagement with experts.
Yuxi_Liu, LessWrong (2019)
44. CREDITS: This presentation template was created by Slidesgo,
including icons by Flaticon, and infographics & images by Freepik
THANKS!
Do you have any questions?
contact@sebkrier.com
@sebkrier