Instrumentation & Measurement Magazine 23-3 - 30

◗◗ Privacy and data governance, including respect for
privacy, quality and integrity of data, and access to data;
◗◗ Transparency, including traceability, explainability and
communication;
◗◗ Diversity, non-discrimination and fairness, including
the avoidance of unfair bias, accessibility and universal
design, and stakeholder participation;
◗◗ Societal and environmental well-being, including
sustainability and environmental friendliness, social
impact, society and democracy; and
◗◗ Accountability, including auditability, minimization and
reporting of negative impact, trade-offs and redress.
The aspects related to AI considered in this document
are certainly extremely relevant because of their suitability
to the goal of producing law-respectful AI; nevertheless, the
definition of AI seems to be quite reduced compared to the
real features of AI. Indeed, the definition of AI given by this
document:
Artificial Intelligence or AI systems Artificial intelligence
(AI) systems are software (and possibly also hardware) systems designed by humans that, given a complex goal, act
in the physical or digital dimension by perceiving their environment through data acquisition, interpreting the collected
structured or unstructured data, reasoning on the knowledge,
or processing the information, derived from this data and deciding the best action(s) to take to achieve the given goal. AI
systems can either use symbolic rules or learn a numeric
model, and they can also adapt their behaviour by analysing how the environment is affected by their previous actions
mentions only AI designed by humans, excluding other entities
that, potentially, could be designed by another AI.
This could-and should-require a deeper reflection on
the matter.

The Role of Metrology
The previous sections considered ethical and legal aspects of
AI, since they appear to be the most relevant ones raised by this
rapidly evolving technology. Another question may arise: can
metrology play a role in this context?
The answer is affirmative, of course, not only because of
the general importance of Instrumentation and Measurement (I&M) in providing quantitative knowledge of entities
and phenomena, but also because of very specific tasks that
do involve metrology and require a deep knowledge of its
fundamentals.
Both ethical and legal issues presented in the previous sections are implicitly or explicitly related to the decisions that an
AI entity takes, mostly in an autonomous way, based on the
output of sensors or on available data. The most debated point
is how to assess whether the decision was the correct one, or at
least the best one, according to the available data.
Metrology plays a critical role here. It is well-known that
data provided by measuring equipment are always affected
by uncertainty [7], and the risk of wrong decisions, when a
30

decision is made according to those data, is directly related to
uncertainty.
Therefore, any AI entity must be instructed to consider uncertainty on the used input data, or even evaluate it, if needed,
to ensure that it is suited for the intended purpose and, eventually, evaluate the consequent risk of wrong decision-making in
order to minimize this risk.
Failing to consider this may lead to dramatic consequences as, for instance, with the recent crashes of the two
Boeing 737 MAX jets, caused by the failure of the single angle-of-attack sensor considered by the automatic control
system. Embedding a bit of metrological competence in the
AI ruling the control system could have avoided the tragic
consequences of this failure, since the information about
the angle of attack could have been retrieved by combining
other available data, such as those coming from the artificial
horizon and airspeed; although less accurate than the information provided by a healthy angle-of-attack sensor, this
information was accurate enough to recognize the failure of
the sensor.
Moreover, we may expect that AI will refer more and more,
in the future, to Big Data and use information inferred from
those data to integrate that coming from its embedded sensors
and take a supposedly more appropriate decision. However,
this requires that the retrieved data have been validated and
their uncertainty considered: this is, once again, the critical
task of metrology and metrologists.
Not considering the relevant metrological issues might
make the designers of AI entities liable for the unfortunate
events that this may cause.

Conclusion
This brief overview on a new, exciting field that poses many
questions has shown the importance of finding an international common ground to establish the legal concept of AI
liability and define specific ethical requirements to ensure
protection for humans (fundamental rights) and avoid negative consequences coming from the complexity of these
products.
Probably our legal framework (referring only to legal orders of Western countries, considering both civil law systems
and common law systems), which has gone through centuries, needs only new interpretations of still suitable rules.
The first element we need is to change (or evolve) the way
we approach these products that are a mix of engineering
(production, etc.) processes and imagination that sometimes
frighten people, especially when they are not fully aware
of the possible effects. Pandora's jar has been opened once
again!

References
[1] L. Rosenberg, "Voices in AI: A Conversation with Dr. Rosenberg
about Swarm Intelligence," Unanimous AI Blog, Aug. 2018.
[Online]. Available: https://unanimous.ai/voices-in-ai/.
[2] "Research and consulting in AI ethics," AI Ethics Lab. [Online].
Available: http://aiethicslab.com.

IEEE Instrumentation & Measurement Magazine

May 2020

https://www.unanimous.ai/voices-in-ai/ http://www.aiethicslab.com

Instrumentation & Measurement Magazine 23-3

Table of Contents for the Digital Edition of Instrumentation & Measurement Magazine 23-3