SAFETY AND SECURITY track - Tuesday 28th
"While facial recognition technology is utilised increasingly across the globe, there are extending debates on the ethical aspects and acceptability of facial recognition. Such issues include e.g. that facial recognition is not an accurate tech, it is creating step by step everywhere reaching “surveillance state”, there are challenges with individual privacy and data security, as well as it may have distorting effects on democratic processes. It is suggested, among other things, that facial recognition technology needs to be well regulated, system needs to be transparent and include “bias checks” as well as there needs to be an administrational procedure for correcting technological and social biases and faults in the system."
MIKA NIEMINEN, Principal Scientist, VTT, Technical Research Centre of Finland
Smart City Mindtrek 2020 – conference
28th-29th January
Tampere, Finland
www.mindtrek.org/2020/
Apidays New York 2024 - The value of a flexible API Management solution for O...
Ethical Questions of Facial Recognition Technologies by Mika Nieminen
1. Ethical Questions of Facial
Recognition Technology
Adjunct professor,
Principal scientist
Mika Nieminen, VTT
mika.nieminen@vtt.fi
11.2.2020 VTT – beyond the obvious 1
2. 11/02/2020 VTT – beyond the obvious 2
There are number of identified
positive uses for facial recgonition
technology, including e.g. prevent
crime, find missing persons, aid
forensic investigations, diagnose
diseases, validate identity etc,
BUT, at the same time, it has been
stated that:
”Facial recognition, simply by being
designed and built, is intrinsically
socially toxic, reagrdless of the
intentions of its makers” (Stark 2019,
52)
3. 11.2.2020 VTT – beyond the obvious 3
Recent AI Global Surveillance (AIGS) Index by Carnegie
Endowment for International Peace compiles empirical
data on AI surveillance use for 176 countries (Feldstein
(2019): The Global Expansion of AI Surveillance) and
indicates constant expansion of the use of tech.Source: Feldstein (2019)
4. 11/02/2020 4
While the strong law traditions and independent institutions of accountability may protect us
against unlawful exploitation of AI tech, there are number of open questions relating to
potential abuse and unintended consequences of the use of AI tech
Such challenges include among others:
Facial recognition is not accurate tech and discriminates against non-whites, women, and
children (e.g. it has been estimated to produce errors up to 35% for non-white women)
(e.g. Kuflinski 2019)
Facial recognition can be claimed to be in any case a “racializing technology” per se
through “the classification and schematization of human facial features” and imposition of
racial categories on to human bodies even though it is scientifically unsound, and thus
reproducing and enhancing systemic inequality (Stark 2019)
If it is possible and easy to use, it is used “just for the case” expanding the extent of
surveillance, creating step by step everywhere reaching “surveillance state” even though
that is not intended by any actor (Kuflinski 2019)
Balancing pros and cons?
5. 11/02/2020 5
Balancing pros and cons?
The surveillance tech produces a lot of data which becomes useful only if connected to
other personal data: How much data there is from innocent people and does the storage
and combination of data breach the right for individual privacy and data security?
There is little publicly available information on the use of surveillance tech (Bode 2019);
Lack of transparency may erode societal trust on and acceptance of legitimate use of
surveillance tech
Surveillance tech may have distorting effect on public events and through that way on
democratic processes. For instance, “London policing ethics panel found that 38% of 16 to
24-year-olds would stay away from events using live facial recognition, with black and
Asian people roughly twice as likely to do so than white people”.(The Guardian 7.6. 2019)
7. 11/02/2020 7
The ethical use of AI has been considered also highly important in Europe: E.g. EU High Level
Experts Group’s recommendations for “Trustworthy AI” (April 2019):
“Individuals should not be subject to unjustified personal, physical or mental tracking or
identification, profiling and nudging through AI powered methods of biometric recognition
such as: emotional tracking, empathic media, DNA, iris, and behavioural identification,
affect recognition, voice, and facial recognition and the recognition of micro-expressions”
Currently the European Commission is exploring stricter rules for facial recognition technology
(EU Observer, 22.8.2019)
In the on-going debate has been discussed e.g. on the necessity of new rules due to existing
General Data Protection Regulation (GDPR)
It has been stated that the GDPR should cover essential questions and now its impacts
are monitored and evaluated (18.9.2019, Berlin; Joanna Goodey, Head of Freedoms and
Justice Department, European Union Agency for Fundamental Rights)
GDPR stipulates that any processing of "biometric data" (like facial recognition data)
requires explicit consent from the person who is concerned
However, data used for national security purposes is excluded from this
More strict regulation is under preparation according to the latest news
8. 11/02/2020 VTT – beyond the obvious 8
Besides number of general ethical guidelines for
the use of AI, there are also attempts to define
specific guidelines for non-biased and unethical
use of facial recognition technology
One recent attempt is put forward by U.K.
Biometrics and Forensics Ethics Group (February
2019) in response to field trials of South Wales
Police and Metropolitan Police Service (“Ethical
issues arising from the police use of live facial
recognition technology”)
Ethical principles include:
Public Interest: The use of this technology is
permissible only when it is being employed in
the public interest
Effectiveness: The use of this technology
can be justified only if it is an effective
tool for identifying people
9. 11/02/2020 9
The Avoidance of Bias and Algorithmic Injustice: For the use of the technology to be
legitimate it should not involve or exhibit undue bias
Impartiality and Deployment: If the technology is deployed for policing purposes it must
be used in an even handed way.
Necessity: Individuals normally have rights to conduct their lives without being monitored
and scrutinized
Proportionality: It can be permissible only if the benefits are proportionate to any
loss of liberty and privacy.
Impartiality, Accountability, Oversight and the Construction of Watch lists: (a) If
humans (or algorithms) are involved in the construction of watch lists for use with the
technology, it is essential that they be impartial and free from bias.(b)The construction of
‘watch lists’ needs to be subject to oversight by an independent body.
Public Trust: It is important that those using it (either in operational deployments or
trials) engage in public consultation and provide the rationale for its use
Cost-effectiveness: Any evaluation of the use of this technology needs to take into
account whether any resources it requires could be better used elsewhere
10. 11/02/2020 VTT – beyond the obvious 10
In order to be widely accepted and utilized facial recognition technology needs to be
well regulated (international and national standards & agreements)
The ethical governance needs to take place on institutional (formal regulation, “hard
governance”), organizational and individual level (self-regulation, ethical principles
e.g. organizational and professional codes)
The system needs to be transparent and include “bias checks” and a possibility for
citizens and organizations appeal against faults and participate in the development
of the system
There needs to be an administrational procedure for correcting technological and
social biases and faults in the system
We need to be aware of unintended negative consequences (like those for the
democratic processes and social activities) and avoid unnecessary use of
surveillance tech
Some preliminary concluding remarks