IEEE Technology and Society Magazine - September 2020 - 39

(transient or permanent) sub-collectives and may be
part of super-collectives. The boundaries and containment of collectives essentially depend on the perspective taken for both analysis and synthesis.
As for MASs, CASs often consist of autonomous entities. Autonomy, generally conceived as the ability to
"self-govern," is actually a complex notion that can have
multiple characterizations and is not simply a black-orwhite feature. Importantly for systems designers, autonomy is often a source of unpredictability as behaviors
and responses vary depending on an individual agent's
self-determination or collective determination of the system. The challenge in CASs, as for MASs, is to promote
collaboration between components, such that together
they can carry out tasks that none of them, as an individual component, would be able to perform [9]. This is
especially hard in competitive settings, where each
component is fully self-interested, but has to interact
with other components to achieve its goal [38]; this may
involve negotiation and trust.
The other key source for unpredictability is the environment, because it generally has complex dynamics
that are at most partially observed. CASs are essentially situated: they are made up of components that
are immersed into some (logical and/or physical) environment and are engaged in non-trivial interactions
with it. Indeed, such systems are adaptive just
because they need to evolve in order to respond to
changes in the environment, the input patterns, or the
responses required.
From a structural point of view, a CAS may exhibit
various dynamic structures, often built and sustained
through self-organization: a robust, internal process by
which order is continually sought, often in an emergent
way [39]. The presence of several (autonomous) components requires an appropriate organization to be
enforced so as to assign roles and responsibilities to the
components themselves [40], [41], and promote the
ordered development of collective behavior.
This is closely related to the dual idea of co-ordination, where rules are enacted to constrain interaction.
Crucially, CASs often feature emergence, whereby
macro-level properties and behaviors spring from
decentralized, micro-level activity. The key point is that
the global properties arising from the interactions within
the system cannot easily be traced back to the properties and behaviors of the individual parts. This poses
significant challenges in the understanding, design, and
control of CASs, to the extent that many engineering
approaches attempt to eliminate the phenomenon. However, emergent behavior must often be expected in
CASs consisting of autonomous components and featuring decentralized control, since there is no single locus
of control over behavior. Moreover, decentralization is
SEPTEMBER 2020

∕

For the last decade the scientific
community has invested considerable
effort into making distributed
computing systems adaptive across
diverse scales.
fundamental in allowing systems to scale (adding more
components to avoid overloading) and to achieve
robustness (as functionality is not confined to few critical components).
CASs exist in the world, in our minds as mental models, in theory as mathematical models, and in software
[42]. Nature has been a great source of inspiration for
mechanisms used by engineers to endow artificial systems with features like self-organization and resilience
[43]. CAS-oriented features are especially useful in scenarios that are highly dynamic with complex dynamics,
that are highly open with a changing component population, and that operate at a very large scale.

CASs and Humans
Humans are prosumers of CASs. In general, humans
actively participate in the CAS by directly producing data
(for example, through handheld or wearable devices), by
monitoring and controlling components, or by providing
specific services in a larger workflow. At the same time,
they are consumers of the information and services
provided by the CASs, as the ultimate goal of the system is to provide some form of utility to human users.
Of course, humans are autonomous, generally unpredictable, have goals evolving with time and context,
and are definitely not directly controllable: humans are
therefore at the same time one of the most valuable
components, and one of the largest sources of design
challenges for CASs. The ultimate value of humans in
CASs is simply that software, whatever its intelligence,
cannot always substitute for human experience and
skill in evaluating situations.
Among the most relevant issues a CAS needs to tackle when leveraging users' data is privacy. Although the
work on privacy in computational systems is large and
well-developed, little is found specifically devoted to
CASs despite the fact that their peculiarities seem to
call for tailored practices. Due to the very nature of a
CAS being "collective" at its core, data may be processed by multiple agents in multiple locations, and it
can therefore be challenging to provide guarantees on
where and when data in CASs is being used and

IEEE TECHNOLOGY AND SOCIETY MAGAZINE

IEEE Technology and Society Magazine - September 2020

Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - September 2020