Blockchain for Intelligent Systems: Research Perspectives

Blockchain for Intelligent Systems
Research Perspectives
Andrea Omicini
andrea.omicini@unibo.it
Giovanni Ciatto
giovanni.ciatto@unibo.it
Keynote @ BCT4MAS 2018
Santiago, Cile – 3 December 2018
Omicini with Ciatto (BCT4MAS 2018) Blockchain for Intelligent Systems Santiago, Cile, 3/12/2018 1 / 56

Outline
1 MAS vs. BCT: The Match
2 Take the Long Way Home

MAS vs. BCT: The Match
Next in Line. . .

Old News vs. New News
Blockchain Technologies (BCT)
BCT as the new kid on the block
everybody is doing BCT nowadays
even myself.
Multi-agent systems (MAS)
MAS as the perennial man-child
everybody (was and) is (still) doing MAS, too
yet, no longer as cool as it was, and already a bit awkward
exception made for myself :)

Is it Just BCT4MAS?
Do BCT and MAS match somehow?
is it a marriage made in (computational) heaven?
is it an arranged marriage for the sake of children (papers)?
do they love each other, actually?
What about MAS4BCT?
beneﬁts from the marriage are easy to see
some of them mutual
possibly not all of them so obvious

Why Intelligent Systems?
MAS for intelligent system engineering
abstractions for distributed systems
agent abstraction for integration of diﬀerent AI techniques
agent cognition & logics
agent society for social intelligence
MAS environment for situatedness
BCT against shortcomings of intelligent systems
accountability
trust & reputation
privacy & security
. . .
see e.g. [Calvaresi et al., 2018]

MAS vs. BCT: The Match The Match
Focus on. . .
The Match
Shake & Bake
The Misses
Logic-based Ledger
Logic-based SC
SC as Coordination Abstractions
Autonomous SC

Distributed Systems (DS)
BCT for distributed systems
BCT is a technology for DS
providing fundamental mechanisms for DS through a distributed
middleware
and features for DS such as transparency
MAS for distributed systems
agent-oriented computing is a paradigm for DS
provide fundamental abstractions for DS through agent middleware
and features for DS such as transparency

Autonomy & Decentralisation
Autonomy & decentralisation in BCT
BCT is a decentralised infrastructure
nodes are supposed to behave autonomously in a BCT
their inner computation is required to be neither predictable nor
observable
Autonomy & decentralisation in MAS
MAS are essentially decentralised in control
agents are assumed to behave autonomously in a MAS
their inner computation is required to be neither predictable nor
observable

Interaction
Interaction in BCT
interaction among distributed autonomous nodes is the key to BCT
behaviour
around simple notions such as value exchange or more complex
abstractions such as smart contracts (SC)
Interaction in MAS
interaction among distributed autonomous agents is the key to MAS
behaviour
around simple notions such as message exchange or more complex
abstractions such as agent coordination

Openness
Openness in BCT
no a priori assumption on the number of nodes—or components, in
general
Openness in MAS
no a priori assumption on the number of agents—or components, in
general

Environment
Environment in BCT
unpredictability and uncertainty are dealt with based on consensus
to build a common view of the environment
Environment in MAS
intelligent agents are assumed to work within a dynamic and
unpredictable environment
based on their autonomy and reactiveness

MAS vs. BCT: The Match Shake & Bake
Focus on. . .
The Match
Shake & Bake
The Misses
Logic-based Ledger
Logic-based SC
Autonomous SC

How MAS & BCT Complement Each Other?
Which are the features for intelligent systems. . .
missing from BCT, that MAS could provide or?
missing from MAS, that BCT could provide or?
The goal here. . .
is then to put everything together
possibly preserving all the combined beneﬁts

Technology vs. Model I
BCT technologies
+ BCT is a young yet (relatively) well-known technology
+ already tested in critical domains
– we cannot really talk of BCT models, for now
– no formal models, yet

Technology vs. Model II
MAS models
+ after years of research, the essence of agent-oriented architecture is
well-understood
+ captured by models, meta-models, architecture for MAS
+ formal models
+ framing and integrating AI results
– with some exceptions, agent technologies are somehow older than
solid, yet
– and tested in few domains

Technology vs. Model III
Engineering intelligent systems
MAS-based (formal) models and methodologies
BC-based technology and middleware

Useful Features for Intelligent Systems I
BCT features
trust & reputation
privacy & security
distributed consensus
accountability
. . .

Useful Features for Intelligent Systems II
MAS features
computational autonomy
goal-oriented behaviour
cognitive process
coordination as govern of interaction
knowledge representation & management, logics, planning, machine
learning, reasoning, . . .
social & situated intelligence
. . .

Glocal Knowledge
BCT global
consensus on distributed ledger provide all nodes with global, shared
view of the world
MAS local
situated agents have each its own fragmented, local knowledge of the
world

Embodied vs. Disembodied
MAS embodied computation
situated agents and MAS promote situated (embodied) computations
BCT disembodied computation
BCT distributed, replicated computations can occur anywhere, and
proceed as disembodied computations

Resources
MAS resources
environment, objects, artefacts, . . .
subject to and means for agent actions
REST and the like
resource as the general notion for acting on MAS environment and
mediating agent-environment interaction
BCT resources
BCT value exchange
PoW
Ethereum’s gas as cost of computation
resource as an economic value for systems

MAS vs. BCT: The Match The Misses
Focus on. . .
The Match
Shake & Bake
The Misses
Logic-based Ledger
Logic-based SC
Autonomous SC

Fragmented Technologies
Technologies for intelligent systems
Real-world intelligent systems mandate for a clear and mature
technological framework
BCT techs are tested yet immature and fragmented
MAS techs are hugely fragmented even though often interoperable at
some level

Weak Standards
Standards for intelligent systems
Integration of intelligent systems in human processes and activities calls
for standards
MAS standards are in place yet not covering the full ground
BCT standards are just non-existing

Take the Long Way Home
Next in Line. . .

Take the Long Way Home Logic-based Ledger
Focus on. . .
The Match
Shake & Bake
The Misses
Logic-based Ledger
Logic-based SC
Autonomous SC

Why LP for Intelligent Systems?
Logic programming (LP)
Programming as goal-oriented reasoning, on top of declarative language
sound and complete logic inference
over knowledge represented as ﬁrst-order logic theories
supports meta-programming
provides understandable outputs (proofs)
already largely adopted in MAS
in cognitive agents, e.g. AgentSpeak [Rao, 1996]
in agent coordination, e.g. ReSpecT [Omicini and Denti, 2001]

Agents & Knowledge
Omniscience
in the same way as as distributed algorithms, agents perform better
with complete knowledge
since MAS are dwelling mostly in uncertain distributed domains such
as the IoT, however, the very notion of agent omniscience have been
mostly banned from MAS research in the last decade at least
yet, there are distributed domains where supporting agent deliberation
with all available knowledge would greatly help
e.g., health care

BCT as a Logic Theory I
Making all knowledge available to every agent
BCT provide agent nodes with a distributed shared data space
? what if data were logic formulae?
! the whole BCT would then represent a logic theory
shared by all agent nodes
→ every agent node would have its own complete logic theory locally
available, consistently replicated

BCT as a Logic Theory II
Updating the distributed logic theory
BCT provide consistent read/write access to the distributed ledger
? if the whole BCT represents a logic theory, how would it evolve?
! agent nodes could issue assert / retract operations
suitable handled by a properly-deﬁned smart contracts
so the the replicated logic theory is updated by any node in a
consistent way
→ the evolution of the shared logic theory is overall consistent

BCT as a Logic Theory III
Using the distributed logic theory
Clients may issue goals to be proven on the BCT logic theory
miners may cooperatively participate to the resolution process
providing either partial or complete results
possibly publishing partial results on the BCT
economical incentives for a fair balancing of computational eﬀorts
e.g., discouraging already-proven goals to be proven again

Take the Long Way Home Logic-based SC
Focus on. . .
The Match
Shake & Bake
The Misses
Logic-based Ledger
Logic-based SC
Autonomous SC

Programming Languages for Smart Contracts (SC)
the language used to write SC affects the potential features of
SC—e.g, [Idelberger et al., 2016]
the language for programming SC is a typical line of research—e.g.,
[Seijas et al., 2017]
issues range from underlying mechanisms to high-level language
abstractions—including the programming paradigm, obviously
Language issues for SC
SC currently lack
high-level understandability of what a SC does and accomplishes
observability of the deployed source code
some degree of evolvability enabling them to be modified (or fixed)

Logic-based Smart Contracts I
Potential beneﬁts
Exploiting a logic language such as Prolog for SC brings some
beneﬁts—see, e.g., [Idelberger et al., 2016]
SC language becomes declarative and goal-oriented, improving
understandability—thus, explainability
built out of a declarative interpreted language, SC are inspectable
without disassemblers
separating static KB (for the immutable code) and dynamic KB (for
the mutable part) enables some sort of controlled evolvability via
assert and retract (on the mutable KB)
context-aware predicates for inspecting the current context
similarly to Solidity’s Globally Available Variables

Logic-based Smart Contracts II
Further issues
should the computational economic cost model be re-designed to
embrace logic programming basic mechanisms?
given the interactive nature of LP, how should logic SC interact?
LP formal semantics for BCT formal semantics?
is it technically feasible?
e.g., with existing languages such as tuProlog [Denti et al., 2005] on top
of some existing BCT

Take the Long Way Home SC as Coordination Abstractions
Focus on. . .
The Match
Shake & Bake
The Misses
Logic-based Ledger
Logic-based SC
Autonomous SC

Blackboard-based Approaches & Smart Contracts
Opportunity
Shared blackboard architectures may take advantage of the replication and
fault-tolerance features they would inherit if deployed on top of a BCT
layer. For instance
e.g. distributed logic programming—e.g., [Ciancarini, 1994, Calegari et al., 2018]
e.g. tuple-based coordination [Rossi et al., 2001]

SC for Coordination & Social Intelligence I
MAS coordination & BCT
generally speaking, SC are computational abstractions ruling the
interaction of nodes in BCT
analogously, coordination media [Ciancarini, 1996] are computational
abstractions ruling the interaction of nodes in BCT
→ SC are amenable of a re-interpretation as coordination abstractions
for BCT nodes

SC for Coordination & Social Intelligence II
Social Intelligence based on SC
as coordination abstractions, SC can be exploited as the core of social
intelligence [Omicini and Papadopoulos, 2001] for BCT-based MAS
as inspectable and malleable coordination artefacts [Omicini et al., 2006],
SC can in principle provide a sound foundation for adaptive,
self-organising BCT-based MAS [Omicini and Viroli, 2011]

Tuple-based Coordination upon the Blockchain
Can we build the archetypal Linda model [Gelernter, 1985] on top of BCTs?
if yes, tuple spaces would inherit
a lot of desirable properties
e.g., decentralisation & replication,
fault-tolerance, consistency, etc.
? which model for economical cost
model for Linda primitives?
? how to handle control
flow-related aspects?
e.g., suspensive semantics
? can we inject programmability
[Denti et al., 1997], too?
Networked hosts
The Blockchain
Communication 
& Coordination 
services
Application 
specific 
services
Workflow 
management
Service 
orchestration
Dependencies 
resolution
Data 
pipelines
Internet of Things
Business Intelligence Web Services
Vision: BCT as the backbone on top of
which communication and coordination
services are built

Take the Long Way Home Autonomous SC
Focus on. . .
The Match
Shake & Bake
The Misses
Logic-based Ledger
Logic-based SC
Autonomous SC

Languages Mechanisms for Smart Contracts
some language issues are related to their underlying operational
mechanisms
for instance
synchronous calls are usually hard coded by construction
poor care for what concerns inter-SC interaction
lack of control ﬂow encapsulation
lack of proactiveness
Basic idea
investigating the adoption of interaction-friendly paradigms such as
actors or agents
where some (all) of the above limitations are overcome at the
fundamental programming abstraction level

Smart Contracts & MAS I
MAS provide a rich (yet coherent) set of abstractions—richer than
actors
“weak” agent notion, basically exploiting agent computational
autonomy [Omicini et al., 2008]—e.g., Jade [Bellifemine et al., 2007]
“strong” agent notion, exploiting mentalistic notion for
cognition—e.g., BDI agents [Bordini et al., 2007, Rao and Georgeﬀ, 1995]
rich MAS meta-models—e.g., Agents & Artifacts (A&A) meta-model
[Omicini et al., 2008]
basic MAS notions



agent
environment
society
mapped upon BCT & SC
many diverse mappings are possible
beyond the straightforward agent/node, artefact/SC mappings

Smart Contracts & MAS II
The issue of autonomy
since they are implemented as threads, SC are said to be autonomous
full computational autonomy requires proactiveness—as in MAS
agents
SC are just reactive, they do not really encapsulate control [Odell, 2002]
so, SC are not really autonomous, they are not agents
? should they be?
? for what?

SC as Intelligent Agents I
SC reasoning
SC rule interaction in an automated way
however, not every interaction can be governed in an automated way
sometimes, it might be preferable to express the rules in terms of
goals to be achieved
sometimes, the the most eﬀective enactment of the rules may depend
on the environment

SC as Intelligent Agents II
SC as intelligent agents
Enhancing SC with
computational autonomy—so SC can act proactively
goal-oriented computations—so interaction rules can be implemented
more ﬂexibly
up-to-date knowledge about the world—using both BCT ledger and
(possibly) run.time oracles to rule interaction based on real-time
knowledge about the environment

SC as Intelligent Agents III
SC as intelligent agents: issues
what is the environment, here?
what can a SC-agent perceive?
how can goal-oriented reasoning be useful here?
should a SC reason about how its business logic?
what about epistemic actions?
should a SC ask for information to other (human) agents?
would multiple intentions – i.e., multiple control ﬂows – make sense
for an SC-agent?
in case, who is paying for them?
who is in charge for executing them?
and, using which concurrency model?

Conclusion
Run, Boy, Run
everybody (was and) is (still) doing MAS
everybody is doing BCT nowadays
more than everything else, everybody is doing AI nowadays
so, if you are doing both MAS and BCT, you have to rush towards
intelligent systems
before everyone else does that!

References
References I
Bellifemine, F. L., Caire, G., and Greenwood, D. (2007).
Developing Multi-Agent Systems with JADE.
Wiley.
Bordini, R. H., H¨ubner, J. F., and Wooldridge, M. J. (2007).
Programming Multi-Agent Systems in AgentSpeak using Jason.
John Wiley & Sons, Ltd.
Hardcover.
Calegari, R., Denti, E., Mariani, S., and Omicini, A. (2018).
Logic programming as a service.
Theory and Practice of Logic Programming, In press.

References
References II
Calvaresi, D., Dubovitskaya, A., Calbimonte, J. P., Taveter, K., and Schumacher, M.
(2018).
Multi-agent systems and blockchain: Results from a systematic literature review.
In Bajo, J., Corchado, J. M., Navarro Mart´ınez, E. M., Osaba Icedo, E., Mathieu, P.,
Hoffa-D¸abrowska, P., del Val, E., Giroux, S., Castro, A. J., Sánchez-Pi, N., Julián, V.,
Silveira, R. A., Fernández, A., Unland, R., and Fuentes-Fernández, R., editors, Advances in
Practical Applications of Agents, Multi-Agent Systems, and Complexity: The PAAMS
Collection 16th International Conference, PAAMS 2018, Toledo, Spain, June 20–22, 2018,
Proceedings, volume 887 of Communications in Computer and Information Science, pages
110–126. Springer International Publishing.
Ciancarini, P. (1994).
Distributed programming with logic tuple spaces.
New Generation Computing, 12(3):251–283.
Ciancarini, P. (1996).
Coordination models and languages as software integrators.
ACM Computing Surveys, 28(2):300–302.

References
References III
Denti, E., Natali, A., and Omicini, A. (1997).
Programmable coordination media.
In Garlan, D. and Le M´etayer, D., editors, Coordination Languages and Models, volume
1282 of LNCS, pages 274–288. Springer-Verlag.
2nd International Conference (COORDINATION’97), Berlin, Germany,
1–3 September 1997. Proceedings.
Denti, E., Omicini, A., and Ricci, A. (2005).
Multi-paradigm Java-Prolog integration in tuProlog.
Science of Computer Programming, 57(2):217–250.
Gelernter, D. (1985).
Generative communication in Linda.
ACM Transactions on Programming Languages and Systems, 7(1):80–112.
Idelberger, F., Governatori, G., Riveret, R., and Sartor, G. (2016).
Evaluation of logic-based smart contracts for blockchain systems.
In Alferes, J. J., Bertossi, L., Governatori, G., Fodor, P., and Roman, D., editors, Rule
Technologies. Research, Tools, and Applications, volume 9718 of Lecture Notes in
Computer Science, pages 167–183. Springer.

References
References IV
Odell, J. (2002).
Objects and agents compared.
Journal of Object Technologies, 1(1):41–53.
Omicini, A. and Denti, E. (2001).
From tuple spaces to tuple centres.
Science of Computer Programming, 41(3):277–294.
Omicini, A. and Papadopoulos, G. A. (2001).
Editorial: Why coordination models and languages in AI?
Applied Artiﬁcial Intelligence: An International Journal, 15(1):1–10.
Special Issue: Coordination Models and Languages in AI.
Omicini, A., Ricci, A., and Viroli, M. (2006).
Agens Faber: Toward a theory of artefacts for MAS.
Electronic Notes in Theoretical Computer Science, 150(3):21–36.
1st International Workshop “Coordination and Organization” (CoOrg 2005),
COORDINATION 2005, Namur, Belgium, 22 April 2005. Proceedings.

References
References V
Omicini, A., Ricci, A., and Viroli, M. (2008).
Artifacts in the A&A meta-model for multi-agent systems.
Autonomous Agents and Multi-Agent Systems, 17(3):432–456.
Special Issue on Foundations, Advanced Topics and Industrial Perspectives of Multi-Agent
Systems.
Omicini, A. and Viroli, M. (2011).
Coordination models and languages: From parallel computing to self-organisation.
The Knowledge Engineering Review, 26(1):53–59.
Special Issue 01 (25th Anniversary Issue).
Rao, A. S. (1996).
AgentSpeak(L): BDI agents speak out in a logical computable language.
In Van de Velde, W. and Perram, J. W., editors, Agents Breaking Away. 7th European
Workshop on Modelling Autonomous Agents in a Multi-Agent World, MAAMAW ’96
Eindhoven, The Netherlands, January 22–25, 1996 Proceedings, pages 42–55, Berlin,
Heidelberg. Springer.

References
References VI
Rao, A. S. and Georgeﬀ, M. P. (1995).
BDI agents: From theory to practice.
In Lesser, V. R. and Gasser, L., editors, 1st International Conference on Multi Agent
Systems (ICMAS 1995), pages 312–319, San Francisco, CA, USA. The MIT Press.
Rossi, D., Cabri, G., and Denti, E. (2001).
Tuple-based technologies for coordination.
In Omicini, A., Zambonelli, F., Klusch, M., and Tolksdorf, R., editors, Coordination of
Internet Agents. Models, Technologies, and Applications, chapter 4, pages 83–109.
Springer Berlin Heidelberg.
Seijas, P. L., Thompson, S., and McAdams, D. (2017).
Scripting smart contracts for distributed ledger technology.
Report 2017/1156, IACR Cryptology ePrint Archive.

Blockchain for Intelligent Systems
Research Perspectives
Andrea Omicini
andrea.omicini@unibo.it
Giovanni Ciatto
giovanni.ciatto@unibo.it
Keynote @ BCT4MAS 2018
Santiago, Cile – 3 December 2018

Blockchain for Intelligent Systems: Research Perspectives

Recommended

Recommended

More Related Content

Similar to Blockchain for Intelligent Systems: Research Perspectives

Similar to Blockchain for Intelligent Systems: Research Perspectives (20)

More from Andrea Omicini

More from Andrea Omicini (20)

Recently uploaded

Recently uploaded (20)

Blockchain for Intelligent Systems: Research Perspectives