Instrumentation & Measurement Magazine 23-2 - 84

them more widely in order to improve competitiveness in manufacturing companies. Access to them need not involve heavy and risky
investment, nor complicated IT tools. The chief obstacle to their utilization has been broadly identified as resistance to change, with that
classic excuse: "It's always been done this way, changing things
might be risky!" Perhaps managerial courage is also part of the
equation...
Measurement uncertainties do not concern trade and
industry only. They can also have an impact on research, a potential the research world needs to be armed against. We raise
this issue in our book, Smart Metrology: From the Metrology of Instrumentation to the Metrology of Decisions [1], where we quote
Physics Nobel prize winner Georges Charpak:
In his book Debunked (Devenez Sorciers, Devenez Savants),
co-authored by Henri Broch, Georges Charpak writes: 'It's
important to realize that the uncertainty associated with data
is every bit as important as the data themselves, because the
uncertainty determines how reliable the data are and thus
how believable the theories based on them...'. By reminding his peers of this fact, Georges Charpak indicates that the
reliability of data is not always as central as it should be to researchers' preoccupations.
Let us hope for some other way forward than Max Planck's
idea: A new scientific truth does not triumph by convincing its opponents [...] but rather because its opponents eventually die. True
as this may be, that particular route to salutary change may
prove to be very, perhaps tragically, long...

Artificial Intelligence: A Disruptive Force
that is Changing the World
The world we live in is aptly described by the English-speaking world as "opinion driven." It is this way of thinking, to go
back to the example of the pens given above, that tells us to expect the worst of our pen tops to match up with the worst of
our pens. This is a belief system that sits at the root of our reasoning, our demands and our practices.
In much the same way, an experimental design that models the behavior of a phenomenon by looking at input factors
is formed around opinion. The investigator, thinking (perhaps
believing) that the phenomenon of interest is influenced by
such and such a factor, constructs his or her experiment around
that belief, though without any certainty that one or more influential parameters have not been overlooked. In which case,
the model will not express reality. How many dysfunctional
situations originate from an incomplete belief? That is perhaps a question no one can answer, but all industrialists, on the
other hand, are at some time confronted with problems they do
not know how to solve. These situations really do exist, of that
there is no doubt!
Big Data, data storage capacities, Cloud and the tremendous calculating power of computers have opened up whole
new possibilities. Artificial Intelligence, which is the product of numerous calculation algorithms, offers us a very
different approach from previous ones. The term "naive" (or
84

agnostic) techniques is sometimes used to describe these "factdriven" algorithms, which require no other hypothesis than
the possession of data. That unique requirement is such an
opportunity for metrology! Because the fact is that data only
provides an accurate representation of reality if the measurements generating it are trustworthy! The collected data allows
us to decide upon an action (conformity, adjustment, etc.), yes,
but it also constitutes a capital that increases with time and,
when suitably exploited by a company, allows understanding and action.

Reliable Measurement Results: A
Categorical Imperative
In July 2017, the French magazine Science & Vie ran an article,
Une nouvelle intelligence est née (A New Intelligence is Born) [5].
An entire section was devoted to Artificial Intelligence and
the power of its algorithms. An example of this power was an
algorithm that had learned to "recognize" the presence of a
lion in a photo. It is worth remembering, here, that an image
seen by a computer is nothing more than a vast array of pixels!
Interestingly, the lion was recognized not by its color or the
shape of its mane, which would be the defining features for
most humans, but a peculiarity of its eyes. The algorithm discovered this signature feature when fed with a photographic
library where some photos did and other photos did not depict lions.
This example is very interesting for two reasons. Firstly,
it clearly demonstrates the ability of algorithms to find distinguishing factors that we humans do not naturally identify.
How many "weak signals" of this type remain hidden in industrial production processes and create problems that impact
negatively on company performance through scrap or altered
lifetimes? Secondly, it illustrates the need for greater measurement reliability than currently exists: if the photo is blurred,
how can the eyes be "seen?"
Data and Artificial Intelligence give us a new level of
control over phenomena that are difficult to grasp with our traditional tools. These new tools, however, bring with them new
requirements, not the least significant of which is the reliability of our data. The new "Smart" metrologist has an essential
role to play in a new world that cannot sit back and wait for
gradual transformation. There is no sense in replacing "worst
case" practices with statistical tolerancing. What is needed is
nothing short of a total disruption of the current way of thinking, setting aside an "opinion driven" mentality for one that is
"fact driven."
The pioneers of Big Data referred to three characteristics to
summarize Big Data: the 3 Vs of Volume, Variety and Velocity.
It probably did not take them long to identify the fourth that
was needed for their algorithms to function, namely Veracity!
Since the algorithms are agnostic, they can do no more than
"believe what they are told"... They can search out weak signals in a multitude of information, but they are incapable of
correcting the data fed to them, even were they supposed to,
which they are not. The saying "Garbage in, garbage out" really comes into its own, here. If you input rubbish, you output

IEEE Instrumentation & Measurement Magazine

April 2020

Instrumentation & Measurement Magazine 23-2

Table of Contents for the Digital Edition of Instrumentation & Measurement Magazine 23-2