IEEE Technology and Society Magazine - June 2018 - 84

information about an individual may confer power over
that individual, resulting in potential misuse by governments, corporations, or other individuals. For example,
households are being equipped with smart meters to
act as providers of temporally detailed energy consumption reports. The utility companies use the data to better estimate domestic power consumption leading to
optimized distribution. However, as shown in [18], several unintended and sensitive inferences such as occupancy and lifestyle patterns of the occupants can be made
from the data. Accordingly, we need configurable privacy-preserving tools that afford users fine-grained control
over how their personal information is shared.
From the algorithmic viewpoint, it is of fundamental
importance for users to understand the type of personal
information being used by algorithms to make decisions, and how such information is used. In addition, in
cases where the user understands that such information
is not correct, is biased, or is not appropriately used
(and thus leads to incorrect algorithmic behaviors), the
user can exploit the above-mentioned configurable privacy-preserving tools to adjust the usage of information
by algorithms. For instance, consider an automated
home heating system that self-regulates based on the
life patterns of inhabitants (and bills accordingly), and a
person who was constrained at home for 15 days due to
a bad winter flu. When such person eventually goes
back to work, he should be able to "see" if the heating
system is still acting on the basis of the wrong assumption that he is at home with a flu. The risk, otherwise, is
to lazily (and stupidly) accept paying more for heating.
For both concerns, pervasive computing technologies can potentially enable users to access the appropriate sensors (e.g., wirelessly via their mobile phones),
and see what data they have produced, and how they
have used what algorithms.

Algorithmic Guardians
If today algorithms affect what we can view online,
tomorrow they will modify our physical reality at home
or in a Smart City, and - as stated earlier - will do that
in personalized way. However, even if we are offered the
possibility of seeing what data is being used and by
what algorithms, the resulting personalization process
might not be transparent or comprehensible, and it
could be hard to understand and interpret, especially
for non-data-literate users.
Besides the problem of understanding personalization, another issue that may arise concerns the fact that
personalization might not always serve the users' interest, but rather the interests of the algorithms' creator.
There is usually an interest gap between you, theuser,
and the third party that paid for the algorithms to prioritize something for you. This can lead to conflicts and

obtrusive personalization. Different digital environments
serve different interests and thus capture different areas
of preferences. Algorithms generalize and simplify, as
they continuously filter out details that are considered
irrelevant or useless. In many cases, algorithms use
other people's data to fill in missing bits and pieces.
Today our algorithmic selves are beyond our control
and can leave us vulnerable. A possible solution could be
to have software tools and algorithms that are on our side
and under our control - algorithmic guardians - capable of somehow protecting us from undesirable behavior
on the part of third party algorithms. We envision algorithmic guardians as far more evolved instances of current
personal assistants such Siri, Watson, etc., that, thanks to
wearable and advanced human-computer interactions
models enabled by pervasive computing, will be always
and easily accessible for interaction. Our digital guardian
will protect us from algorithmic manipulation that
restricts personal freedom and will make sure that we are
not stuck on repeating behavioral loops or virtual echochambers. It will create an adaptive information interface
that is fresh and relevant. Furthermore, guardians will
support us in controlling our personal data flows and
deciding who can access our digital trails. For instance,
with reference to the home heating example in the Data
Access Control section above, our personal guardian
should be able to alert us that the heating systems is still
acting as if we still had the flu, should help us correct
such behavior.
Our digital guardian does not need to be intelligent
in the same way as humans. It needs to be smart in
relation to the environments it inhabits - in relation to
the other algorithms it encounters. In any case, even if
algorithmic guardians (unlike third party algorithms) are
user-owned and are totally under our own control, being
able to understand how they work will be a priority to
make them fully trustworthy. This issue, which applies
to all algorithms that will govern our life - is elaborated
in the next section.

Democracy Through Grey-Box Code for Algocracy
Algorithms are created by a handful of people or, in a
probable near future, by other algorithms that are created by an even smaller number of people. This raises an
enormous challenge for our democracies.
In the spirit of "freedom of information," publishing
the algorithms towards citizens seems like an obvious
thing to do. However, this doesn't make a lot of sense if
the overwhelming majority of the citizens cannot even
read them. In today's democracy, civil law is also produced by a handful of specialists and to use it, we rely
on lawyers. Nevertheless, the civil code can be consulted by anyone who is willing to make the mental effort of
digging through some heavy prose. With the current

IEEE Technology and Society Magazine

∕

JUNE 2018

Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - June 2018