Self-organisation of Knowledge in Socio-technical Systems: A Coordination Perspective

Self-organisation of Knowledge
in Socio-technical Systems:
A Coordination Perspective
Stefano Mariani, Andrea Omicini
Universit`a di Bologna
ISTC CNR
Rome, Italy
November 6th, 2015
Mariani, Omicini (Universit`a di Bologna) Self-* Knowledge Coordination in STS CNR, Rome – 6/11/2015 1 / 48

Outline
1 Coordination Issues in STS
2 Tacit Messages and Perturbation Actions in Real-world STS
3 BIC, Stigmergy, and Smart Environments
4 Toward Self-organising Workspaces
5 M olecules of K nowledge
6 Conclusion
7 Outlook

Coordination Issues in STS
Outline
6 Conclusion
7 Outlook

Challenges of Socio-technical Systems
Socio-technical systems (STS) arise when cognitive and social
interaction is mediated by information technology, rather than by the
natural world (alone) [Whi06]
STS are heavily interaction-centred, thus need to deal with
coordination issues at the infrastructural level [MC94]
Among the many coordination issues in STS are:
unpredictability — “humans-in-the-loop” vs. software
programmability and predictability
⇒ coordination should account natively for unpredictability
and uncertainty
scale — large-scale distribution, openness, ever-increasing
number of users, devices, data
⇒ coordination should exploit decentralised mechanisms to
scale in/out upon need

Challenges of Knowledge-intensive Environments
Knowledge-intensive Environments (KIE) are workplaces in which
sustainability of the organisation’s long-term goals is influenced by the
evolution of the body of knowledge embodied within the organisation
itself [Bha01]
Usually, KIE are computationally supported by STS, thus they need
proper coordination too
Among the many coordination issues in KIE are:
size — massive amount of raw data, aggregated information,
reification of procedures and best-practices, and the like
⇒ coordination should minimise the overhead of
information needed for coordination-related (non-)
functional requirements
pace — high rate of information production and
consumption, huge frequency of interactions
⇒ coordination mechanisms should be as simple and
efficient as possible

A Path to Follow
Coordination models and technologies have already drawn inspiration from
natural systems, looking for mechanisms enabling and promoting
self-organising and adaptive coordination
[VPB12, MZ09, VC09, ZCF+11, MO13]
A novel perspective
Similarly, we focus on the human factor in STS, seeking novel coordination
approaches inspired by the latest cognitive and social sciences theories of
action and interaction

The Approach
1 We observed real-world STS/KIE, analysing their (implicit) models of
action and interaction
2 We generalised such models according to the theoretical framework of
Behavioural Implicit Communication [CPT10], devising out tacit
messages and implicit actions computationally exploited by such
STS/KIE
3 We conceived the M olecules of K nowledge model [MO13],
promoting self-organisation of knowledge in STS/KIE, inspired by the
above framework and geared toward the notion of self-organising
workspace [Omi11]

Tacit Messages and Perturbation Actions in Real-world STS
Outline
6 Conclusion
7 Outlook

Tacit Messages I
Tacit messages are introduced in [CPT10] to describe the kind of message
a practical action (and its traces) may implicitly send to its observers:
presence — “Agent A is here”. Any agent (as well as the environment
itself) observing any practical behaviour of A becomes aware of its
existence — and, possibly, of contextual information, e.g., its
location.
intention — “Agent A plans to do action β”. If the agents’ workﬂow
determines that action β follows action α, peers (as well as the
environment) observing A doing α may assume A next intention to
be “do β”.
ability — “A is able to do φi∈N”. Assuming actions φi∈N have similar
pre-conditions, agents (and the environment) observing A doing φi
may infer that A is also able to do φj=i∈N.

Tacit Messages II
opportunity — “pi∈N is the set of pre-conditions for doing α”. Agents
observing A doing α may infer that pi∈N hold, thus, they may take
the opportunity to do α as soon as possible.
accomplishment — “A achieved S”. If S is the “state of aﬀairs” reachable
through action α, agents observing A doing α may infer that A is
now in state S.
goal — “A has goal g”. By observing A doing action α, peers of A
may infer A goal to be g, e.g. because action α is part of a
workﬂow aimed at achieving g.
result — “Result R is available”. If peer agents know that action α
leads to result R, whenever agent A does α they can expect result
R to be soon available.

Tacit Messages in Real-world STS I
We identiﬁed a set of (virtual) practical actions, fairly common in
real-world STS despite the diversity in scope of each speciﬁc STS —
e.g. Facebook vs. Mendeley1 vs. Storify2
For each, we point to a few tacit messages they may convey:
quote/share — re-publishing or mentioning someone else’s
information can convey, e.g., tacit messages presence,
ability, accomplishment. If X shares Y ’s information
through action a, every other agent observing a becomes
aware of existence and location of both X and Y (presence).
The fact that X is sharing information I from source S lets
X’s peers infer X can manipulate S (ability). If X shared I
with Z, Z may infer that X expects Z to somehow use it
(accomplishment).

Tacit Messages in Real-world STS II
like/favourite — marking as relevant a piece of information can
convey tacit messages presence, opportunity. If the
socio-technical platform lets X be aware of Y marking
information I as relevant, X may infer that Y exists
(presence). If Y marks as relevant I belonging to X, X may
infer that Y is interested in her work, perhaps seeking for
collaborations (opportunity).
follow — subscribing for updates regarding a piece of
information or a user can convey tacit messages
intention, opportunity. Since X manifested interest in Y ’s
work through subscription, Y may infer X intention to use it
somehow (intention). Accordingly, Y may infer the
opportunity for collaboration (opportunity).

Tacit Messages in Real-world STS III
search — performing a search query to retrieve information can
convey, e.g., tacit messages presence, intention,
opportunity. If X search query is observable by peer agents,
they can infer X existence and location (presence). Also, they
can infer X goal to acquire knowledge related to its search
query (intention). Finally, along the same line, they can take
the chance to provide matching information (opportunity).
Now the question is
How to computationally exploit the envisioned mind-reading and
signiﬁcation abilities from a coordination perspective?
1
https://www.mendeley.com
2
https://storify.com

Perturbation Actions I
Perturbation actions are computational functions changing the state of a
STS, in response to users’ interactions, but transparently to them [MO15]
A possible answer is
Perturbation actions may then exploit the implicit information conveyed by
tacit messages to leverage mind-reading and signiﬁcation
for coordination purposes

Perturbation Actions II
Accordingly, perturbation actions may:
spread discovery messages informing agents about the presence and
location of another (tacit message presence)
establish privileged communication channels between frequently
interacting agents (opportunity)
undertake coordination actions enabling/hindering some
desirable/dangerous interaction protocol (intention, ability, goal)
autonomously notify users about availability of novel, potentially
interesting information (accomplishment, result)

Perturbation Actions in Real-world STS I
The virtual practical actions already identiﬁed are likely to (transparently)
cause perturbation actions under-the-hood:
quote/share — provided by Facebook, Twitter (retweet), G+, LinkedIN,
Mendeley (post), Academia.edu (publish), ResearchGate
(publish), Storify, etc. It is likely to help the STS platform,
underlying the social network application, in:
suggesting novel connections
ranking feeds in the newsfeed timeline
like/favourite — provided by Facebook, Twitter, G+ (+1), LinkedIN
(suggest), Mendeley, Academia.edu (bookmark),
ResearchGate (follow/download), Storify, etc. It is likely to
inﬂuence the STS as above.

Perturbation Actions in Real-world STS II
follow — provided by Facebook (add friend), Twitter, G+ (add),
LinkedIN (connect), Mendeley, Academia.edu, ResearchGate,
etc. It is likely to help the STS platform by:
suggesting further connections
activating/ranking feeds in the newsfeed timeline
search — provided by almost every social network, it is the
epistemic action by its very deﬁnition, thus may be exploited
by the STS platform in a number of ways:
re-organising the knowledge graph internally used by the
STS
tune the algorithm providing suggestions
improve personalised advertising policies
and many more. . .

A Systematic Analysis
. . . can we frame the above observations
within a
coherent computational framework
?
Of course we can.

BIC, Stigmergy, and Smart Environments
Outline
6 Conclusion
7 Outlook

BIC in a Nutshell
Implicit interaction
Behavioural implicit communication (BIC) is a form of implicit interaction
where no specialised signal conveys the message, since the message is the
practical behaviour itself — and possibly, its post hoc traces [CPT10]
BIC presupposes advanced observation capabilities: agents should be able to
observe others’ actions (and traces), as well as to mind-read the intentions
behind them, so as to leverage signiﬁcation
BIC applies to human beings, to both cognitive and non-cognitive agents,
and to computational environments as well [WOO07]
Through BIC, such environments can become smart environments, namely
pro-active, intelligent workplaces able to autonomously adapt their
conﬁguration and behaviour according to users’ interactions [CPT10]

Cognitive Stigmergy in a Nutshell
Trace-based Communication
The notion of stigmergy has been introduced in the biological study of
social insects [Gra59], to characterise how termites (unintentionally)
coordinate themselves during nest construction, with no need of
exchanging direct messages, but relying solely on local interactions instead
Stigmergy is a special form of BIC, where the addressee does not directly
perceive the behaviour, but just other post-hoc traces — in the form of
environment modiﬁcations
Such modiﬁcations are amenable of a symbolic interpretation, thus
exploitable by agents featuring cognitive abilities — either humans or
software
When traces become signs, stigmergy becomes cognitive stigmergy, which
involves agents able to correctly understand traces as signs intentionally left
in the environment [Omi12]

Computational Smart Environments in a Nutshell
In [TCR+05], an abstract model for computational smart
environments is proposed, which defines two types of environment
c-env — common environment, where agents can observe only the
state of the environment (including actions’ traces), not the
actions of their peers
s-env — shared environment, enabling different forms of
observability of actions, and awareness of this observability
Then, three requirements enabling them are devised
1 observability of agents’ actions and traces should be enabled by default
2 the environment should be able to understand actions and their traces,
possibly inferring intentions and goals motivating them
3 agents should be able to understand the effects of their activity on the
environment as well as on the other agents, so as to opportunistically
obtain a reaction

Toward Self-organising Workspaces
Outline
6 Conclusion
7 Outlook

Toward Self-organising Workspaces
Putting All Together
Quoting from [Omi11], a STS for working in knowledge-intensive
environments requires
“that all the relevant information sources are made available to
the user in a complete yet usable format, [. . . ] that the working
environment autonomously evolves and adapts to the individual
uses and work habits”
Furthermore,
“the main point here is the explicit representation, memorisation
and exploitation of user actions in the workspace”
Our master equations
BIC + (cognitive) stigmergy = Smart Environments
Smart Environments + Self-organisation = Self-organising Workspaces

M olecules of K nowledge
Outline
6 Conclusion
7 Outlook

M olecules of K nowledge Model
Outline
Model
Architecture
Early Results
6 Conclusion
7 Outlook

MoK in a Nutshell I
MoK
M olecules of K nowledge (MoK ) is a coordination model promoting
self-organisation of knowledge [MO13]
Inspired to biochemical tuple spaces [VC09], stigmergic coordination
[Par06], and BIC [CPT10]
Main goals
1 self-aggregation of information into more complex heaps, possibly
reifying useful knowledge previously hidden
2 autonomous diﬀusion of information toward the interested agents, that
is, those needing it to achieve their goals

MoK in a Nutshell II
A MoK -coordinated system is
a network of MoK compartments (tuple-space like information
repositories). . .
. . . in which MoK seeds (sources of information) autonomously inject
MoK atoms (information pieces)
atoms undergo autonomous and decentralised reactions
aggregate into molecules (composite information chunks)
diffuse to neighbourhoods
get reinforced and perturbed by users
decay as time flows
reactions are influenced by enzymes (reification of users’ epistemic
actions) and traces (their (side) effects). . .
. . . and scheduled according to Gillespie’s chemical dynamics simulation
algorithm [Gil77]

MoK Enzymes and Traces as BIC Enablers I
Model
enzyme(species, s, mol)c
enzyme(species, s, mol ) + molc
rreinf
−−−→ enzyme(species, s, mol ) + molc+s
trace(enzyme, p, mol)c
trace(enzyme, p, mol ) + molc
rpert
−−→ .exec (p, trace, mol)
enzyme
rdep
−−→ enzyme + trace(enzyme, p[species], mol)
species deﬁnes the epistemic nature of the action
s strength of reinforcement
p the perturbation the trace wants to perform
.exec starts execution of perturbation p 3

MoK Enzymes and Traces as BIC Enablers II
Reinforcement influences relevance of information according to the
(epistemic) nature and frequency of their actions and interactions
Enzymes situate actions, e.g., at a precise time as well as in a precise
space
Mind-reading and signification are enabled by assuming that users
manipulating a given corpus of information are interested in that
information more than other
Perturbation influences location, content, any domain-specific trait of
information, according to users’ inferred goals, with the goal of easing
and optimising their workflows
Traces enable the environment to exploit users’ actions (possibly,
inferred) side-effects for the profit of the coordination process —
promoting the distributed collective intelligence leading to
anticipatory coordination
3
Notice, p is implicitly defined by species, as highlighted by notation p[species].

M olecules of K nowledge Architecture
Outline
Model
Architecture
Early Results
6 Conclusion
7 Outlook

M olecules of K nowledge Architecture
MoK Ecosystem Architecture

M olecules of K nowledge Early Results
Outline
Model
Architecture
Early Results
6 Conclusion
7 Outlook

Simulated Scenario
Citizen journalism scenario:
users share a MoK -coordinated IT platform for retrieving and
publishing news stories
they have personal devices (smartphones, tablets, pcs, workstations),
running the MoK middleware, which they use to search within the IT
platform relevant information
search actions can spread up to a neighbourhood of compartments —
e.g., to limit bandwidth consumption, boost security, optimise
information location, etc.
search actions leave traces the MoK middleware exploits to attract
similar information

Anticipatory coordination
Figure: Whereas data is initially randomly scattered across workspaces, as soon as users
interact clusters appear by emergence thanks to BIC-driven self-organisation. Whenever new
actions are performed by catalysts, the MoK infrastructure adaptively re-organises the spatial
conﬁguration of molecules so as to better tackle the new coordination needs.

Discussion
MoK anticipates users’ needs, not based on behaviour prediction, but
on present actions and its mind-reading and signiﬁcation abilities
addressing unpredictability
MoK reactions act only locally, thus exploit local information solely
addressing scale
MoK decay destroys information as time passes — furthermore, the
overhead brought by MoK is minimal, since it exploits solely
information already in the system
addressing size
MoK reaction execution and BIC-related mechanisms are rather
eﬃcient, mostly due to their local nature and absence of complex
reasoning
addressing pace

Conclusion
Outline
6 Conclusion
7 Outlook

Conclusion
Summing Up
Engineering eﬀective coordination for large-scale, knowledge-intensive
STS is a diﬃcult task
Nature-inspired approaches proven successful in mitigating the issue,
by leveraging self-organisation and adaptiveness
We may further improve by shifting attention toward the social side of
STS, transparently exploiting the epistemic nature of users’ (inter-)
actions for coordination purposes
The tools in our hands
BIC, (cognitive) stigmergy, and biochemical coordination give us the right
models and approaches to do so

Outlook
Outline
6 Conclusion
7 Outlook

Outlook
A Bright and Exciting Future Awaits I
The world needs eﬃcient and smart ways of preserving, managing,
and analysing the astonishing amount of knowledge it produces and
consumes every day
Big data approaches are more or less the standard now, mostly
because they are good in ﬁnding patterns of knowledge, but:
they mostly fail in discovering anti-patterns, e.g., detecting outliers
they mostly fail in accommodating ever-changing, heterogeneous
knowledge discovery needs
they mostly neglect “humans-in-the-loop”, relying on algorithms and
measures (e.g. of similarity) which are completely user-neutral and
goal-independent
they won’t scale forever
they are not suitable for pervasive and privacy-demanding scenarios

Outlook
A Bright and Exciting Future Awaits II
We are in the perfect spot to start a paradigm shift toward
self-organising knowledge, where:
user-centric adaptiveness of knowledge discovery processes is the
foremost goal
measures and algorithms exploited for knowledge discovery, inference,
management and analysis natively account for users’ goals
seamlessly scale up/down/out/in naturally, being operating on the
assumption that only local-information is available consistently
As witnessed by the latest H2020 calls, increasingly demanding
user-inclusive policy making, governance participation, user-centric
knowledge sharing platforms, etc.
H2020-SC6-CO-CREATION-2016-2017
H2020-EINFRA-2016-2017
H2020-FETPROACT-2016-2017

References
References I
Ganesh D. Bhatt.
Knowledge management in organizations: Examining the interaction between technologies,
techniques, and people.
Journal of Knowledge Management, 5(1):68–75, 2001.
Cristiano Castelfranchi, Giovanni Pezzullo, and Luca Tummolini.
Behavioral implicit communication (BIC): Communicating with smart environments via our
practical behavior and its traces.
International Journal of Ambient Computing and Intelligence, 2(1):1–12, January–March
2010.
Daniel T. Gillespie.
Exact stochastic simulation of coupled chemical reactions.
The Journal of Physical Chemistry, 81(25):2340–2361, December 1977.
Pierre-Paul Grassé.
La reconstruction du nid et les coordinations interindividuelles chez Bellicositermes
natalensis et Cubitermes sp. la théorie de la stigmergie: Essai d’interprétation du
comportement des termites constructeurs.
Insectes Sociaux, 6(1):41–80, March 1959.

References
References II
Thomas W. Malone and Kevin Crowston.
The interdisciplinary study of coordination.
ACM Computing Surveys, 26(1):87–119, 1994.
Stefano Mariani and Andrea Omicini.
Molecules of Knowledge: Self-organisation in knowledge-intensive environments.
In Giancarlo Fortino, Costin B˘adic˘a, Michele Malgeri, and Rainer Unland, editors,
Intelligent Distributed Computing VI, volume 446 of Studies in Computational Intelligence,
pages 17–22. Springer, 2013.
Stefano Mariani and Andrea Omicini.
Anticipatory coordination in socio-technical knowledge-intensive environments:
Behavioural implicit communication in MoK.
In Marco Gavanelli, Evelina Lamma, and Fabrizio Riguzzi, editors, AI*IA 2015, Advances
in Artiﬁcial Intelligence, volume 9336 of Lecture Notes in Computer Science, chapter 8,
pages 102–115. Springer International Publishing, 23–25 September 2015.
XIVth International Conference of the Italian Association for Artiﬁcial Intelligence, Ferrara,
Italy, September 23–25, 2015, Proceedings.

References
References III
Marco Mamei and Franco Zambonelli.
Programming pervasive and mobile computing applications: The TOTA approach.
ACM Transactions on Software Engineering and Methodology (TOSEM), 18(4), July 2009.
Andrea Omicini.
Self-organising knowledge-intensive workspaces.
In Alois Ferscha, editor, Pervasive Adaptation. The Next Generation Pervasive Computing
Research Agenda, chapter VII: Human-Centric Adaptation, pages 71–72. Institute for
Pervasive Computing, Johannes Kepler University Linz, Austria, May 2011.
Andrea Omicini.
Agents writing on walls: Cognitive stigmergy and beyond.
In Fabio Paglieri, Luca Tummolini, Rino Falcone, and Maria Miceli, editors, The Goals of
Cognition. Essays in Honor of Cristiano Castelfranchi, volume 20 of Tributes, chapter 29,
pages 543–556. College Publications, London, December 2012.
H. Van Dyke Parunak.
A survey of environments and mechanisms for human-human stigmergy.
In Danny Weyns, H. Van Dyke Parunak, and Fabien Michel, editors, Environments for
Multi-Agent Systems II, volume 3830 of Lecture Notes in Computer Science, pages
163–186. Springer, 2006.

References
References IV
Luca Tummolini, Cristiano Castelfranchi, Alessandro Ricci, Mirko Viroli, and Andrea
Omicini.
“Exhibitionists” and “voyeurs” do it better: A shared environment approach for ﬂexible
coordination with tacit messages.
In Danny Weyns, H. Van Dyke Parunak, and Fabien Michel, editors, Environments for
Multi-Agent Systems, volume 3374 of Lecture Notes in Artiﬁcial Intelligence, pages
215–231. Springer, February 2005.
Mirko Viroli and Matteo Casadei.
Biochemical tuple spaces for self-organising coordination.
In John Field and Vasco T. Vasconcelos, editors, Coordination Languages and Models,
volume 5521 of Lecture Notes in Computer Science, pages 143–162. Springer, Lisbon,
Portugal, June 2009.
Mirko Viroli, Danilo Pianini, and Jacob Beal.
Linda in space-time: An adaptive coordination model for mobile ad-hoc environments.
In Marjan Sirjani, editor, Coordination Models and Languages, number 7274 in Lecture
Notes in Computer Science, pages 212–229. Springer, 2012.

References
References V
Brian Whitworth.
Socio-technical systems.
Encyclopedia of human computer interaction, pages 533–541, 2006.
Danny Weyns, Andrea Omicini, and James J. Odell.
Environment as a ﬁrst-class abstraction in multi-agent systems.
Autonomous Agents and Multi-Agent Systems, 14(1):5–30, February 2007.
Franco Zambonelli, Gabriella Castelli, Laura Ferrari, Marco Mamei, Alberto Rosi, Giovanna
Di Marzo, Matteo Risoldi, Akla-Esso Tchao, Simon Dobson, Graeme Stevenson, Yuan Ye,
Elena Nardini, Andrea Omicini, Sara Montagna, Mirko Viroli, Alois Ferscha, Sascha
Maschek, and Bernhard Wally.
Self-aware pervasive service ecosystems.
Procedia Computer Science, 7:197–199, December 2011.

Extras
URLs
Slides
On APICe
http://apice.unibo.it/xwiki/bin/view/Talks/MokCnr2015
On SlideShare
http://www.slideshare.net/andreaomicini/
selforganisation-of-knowledge-in-sociotechnical-systems-
a-coordination-perspective
MoK
On APICe
http://mok.apice.unibo.it/

Self-organisation of Knowledge
in Socio-technical Systems:
A Coordination Perspective
Stefano Mariani, Andrea Omicini
Universit`a di Bologna
ISTC CNR
Rome, Italy
November 6th, 2015

Self-organisation of Knowledge in Socio-technical Systems: A Coordination Perspective

More Related Content

Similar to Self-organisation of Knowledge in Socio-technical Systems: A Coordination Perspective (20)

More from Andrea Omicini (20)

Recently uploaded (20)

Self-organisation of Knowledge in Socio-technical Systems: A Coordination Perspective