‘Healthy AI’ at MozFest 2019

HBP Ethics Support team members George Ogoh and Damian Eke at the MozFest

George Ogoh

The theme of the Mozila Festival (MozFest) for 2019 was ‘Healthy AI’. MozFest is an
eventful gathering of people interested in discussing topics related to a
healthier internet. This year was the 10th anniversary and the
events took place in London from 21 to 27 October 2019.

The theme of ‘healthy AI’ was chosen to highlight the ways algorithms
embedded in emerging and familiar technologies are making decisions for us. It enabled
the evaluation of AI advances and aimed to encourage brainstorming on ‘ways to
bring more social responsibility, ethics, and user agency to AI’. It was therefore
unsurprising to see a good number of presentations on artificial intelligence being
facilitated at the event. Examples include a talk on AI and Ethics by Cansu
Canca, and another on Trustworthy
AI: Imagining Better Machine Decision Making
by Irini Mirena.

 

Sharing health data
responsibly

As data is at the heart of artificial intelligence, it was interesting
to see that issues around data governance featured quite prominently at the
event, especially on day 6. One such session was titled ‘Health Data Sharing is
Caring’. This very interactive session began by drawing on the blue pill-red
pill metaphor of the Metrix movie to illustrate issues around machine learning
and data sharing. Recall that in the Matrix, Morpheus (leader of the resistance
against AI) held out two pills (a red and a blue one) to Neo, a computer
programmer. Taking the red pill will enable Neo to join the AI resistance, but with
the blue pill, Neo will to return to blissful ignorance about issues promoted
by AI.

An important point made by drawing on this metaphor was how technology
could very easily dominate the people for whom it was made. While so much good
may be drawn from sharing health data, some concerns have also been raised. Using
the blue pill-red pill analogy, it was pointed out that “AI has introduced the
possibility of using healthcare data to produce powerful models that can
automate diagnosis” (Blue Pill); and yet “scientists often lack access to
balanced datasets needed to optimally train algorithms and avoid replicating bias”
(Red Pill).

In light of such issues, one of the session participants pointed out
that “responsible data governance is needed” similar to the HBP’s Ethics
Support team’s proposal for a responsible
data governance of big neuroscience data
. Another participant explained the
concept of responsible
research and innovation (RRI)
to the group sparking a vibrant discussion
around the RRI framework approaches. Since 2011 this approach has been promoted
by major research funding bodies (e.g. the European
Commission’s Horizon 2020
and the UK’s Engineering
and Physical Science Research Council – EPSRC
). Following these
conversations, the facilitators of this session have begun seriously
considering RRI
as a suitable approach to healthcare data management
.

 

Data trusts

The idea of data trusts and data unions was also discussed at the event.
At one session titled ‘From data serfs to empowered citizens: the need for data
trusts’ it was suggested that data trusts could give ordinary citizens greater
control over how personal data is collected and used while also allowing everyone
to share in the value created from their personal data. Current business models
usually mean that only organisations derive value from sharing or sale of
personal data. However, data trusts could reverse this trend and enable individuals
to derive greater value from their personal data.     

A model was suggested where data trusts are controlled by a trustee that
holds fiduciary obligations and manages data according to terms of a trust. This
means that when data is assigned to the trust, the trustee cannot act according
to their interests but must base their actions on the interests of data
subjects. The trustee negotiates terms of access to data on behalf of those who
have provided their data with shorter, clearer terms that enhances informed
consent. In this model, data trusts leverage collective action and enable
people to re-balance the power dynamics between huge multinationals and
ordinary citizens.

Although ideas around data trusts are not entirely new, it has been
pointed out that the concept is still in its infancy (Kemp, 2019; Kirkwood, 2019). Yet, data trusts are
increasingly being seen as a way of unlocking
the value of data
while preventing harmful impacts. Another session that
helped to reinforce this was titled ‘Approaches to Data Governance’. During
this session, it was argued that there was need for governance systems like
data trust because they are easily able provide a clear definition of limits,
rely on a blend of legal rights, and create legally enforceable fiduciary
duties.

 

Protecting children
online

The importance of protecting children online cannot be overemphasised.
One interesting session that took on this topic was titled ‘Privacy, Ethics
& Security Framework for Children’s Data’. During
this highly interactive session, participants were urged to discuss key challenges they observed with children’s data
and to propose possible solutions. Among other things, participants suggested
they were concerned about issues around control and ownership of children’s
personal data, protection of children’s privacy, and how to obtain truly
informed consent when collecting and sharing children’s data. Some solutions
they suggested include creating a framework for privacy by default which for
example, allows people to opt out of data sharing and processing by default, greater
transparency for algorithms used in children’s software, greater fines for
negligence by companies, and simple quizzes to ensure that the individual
thinks about decisions being made when giving consent.

Overall, MozFest 2019 was an interesting
event that provided an opportunity for discussion and reflection on societal
concerns for AI, as well as issues around and data governance and social
responsibility when dealing personal data. It is interesting to see that Mozilla
has embraced the idea of data trusts and is building a tool for ethical
governance of AI through data trusts. However, like Croft (2019) points out, there remain
questions that still need to be answered about the legal structure that data
trusts would hold.

 

References

Croft, J.
(2019) Data trusts raise questions on privacy and governance. [Online]
Financial Times. Available from:
https://www.ft.com/content/a683b8e4-a3ef-11e9-a282-2df48f366f7d [Accessed 03/12/19].

Kemp, R.
(2019) Data trusts and frameworks are gaining traction and on the cusp of
widespread adoption
. [Online] IT Talk at the Apex. Available from:

Data trusts and frameworks are gaining traction and on the cusp of widespread adoption

[Accessed 03/12/19].

Kirkwood, I.
(2019) Element AI partners with Mozilla on R&D for data trusts, ethical
AI projects
. [Online] BetaKit. Available from:

Element AI partners with Mozilla on R&D for data trusts, ethical AI projects

[Accessed 03/12/19].

The post ‘Healthy AI’ at MozFest 2019 appeared first on Ethics Dialogues.


Source: New feed

Share This Post
Have your say!
00