Shall we Depend on Dependability for Industrial Uptake of Agent Technology

1
Shall we Depend on Dependability for Industrial
Uptake of Agent Technology?

• Rafael H. Bordini
• Department of Computer Science
• University of Durham, U.K.
• R.Bordini_at_durham.ac.uk

2
Dependable Systems

• As computer systems permeate further our daily
  life, mankind becomes increasingly dependant on
  them
• We need to make sure they do not ever crash, but
  also make sure they cannot be sabotaged
• The area of dependable systems brings together
  work on various aspects of computing such as
  fault tolerance, cryptography, and formal methods
• Ultimately, formal verification would give
  assurance that we can reliably depend on such
  systems
• Many of the typical areas of applications for MAS
  require dependable systems!

3
Formal Verification

• Two main approaches
• Theorem Proving
• Model Checking
• But also direct execution, and various formal
  approaches are also amenable to some form of
  verification (logic programming, Petri nets,
  etc.)
• Model Checking can be based on a variety of
  techniques
• Automata-Theoretic (explicit state)
• OBDDs (symbolic model checking)
• Reduction to SAT (to use existing SAT solvers)

4
Model Checking MAS

• Most of the (still very young!) work on
  verification of multi-agent systems has focused
  on Model Checking
• Some of the reasons for this might be
• Easier for end-users than theorem proving
• Counterexample can be quite useful for debugging
• Some interesting logics (e.g., ATL) for agent
  verification have simple (and tractable) model
  checking algorithms, but very hard to give sound
  and complete axiomatisations for
• Some of the researchers who have contributed to
  the area are M.Wooldridge, W.van der Hoek,
  M.Fisher, W.Visser, R.Bordini, A.Lomuscio,
  V.Panczek, M.Ryan, P.-Y.Schobbens, M.Singh, R.van
  den Mayden, J.Padget, A.Cimatti, M.Benerecetti,

5
TFG Q1 Is it Easier to Apply Formal Methods to
MAS?

• No! At the moment, probably more difficult (too
  recent, so not many specific/robust tools
  available)
• But arguably MAS techniques are appropriate for
  the development of certain types of distributed
  systems and verification techniques tailored
  specifically for this will be required or at
  least useful
• It is still an open question whether some
  characteristics of multi-agent systems (as a
  high-level modular approach to developing DS)
  will allow for particularly successful
  state-space reduction techniques (the use of such
  techniques are an essential part of practical
  model checking), e.g., compositional reasoning

6
TFG Q2 What are the Advantages of Applying FM to
MAS Developments?

• As for other systems, formal methods ensure that
  we can reliably depend on computer systems
• Verification techniques created specifically for
  MAS may help development itself (e.g., debugging)
  and it is possible they will be more efficient
  for these systems, but we are far from having
  enough evidence to claim that as yet
• Ideally, we would have off-the-shelf, verified
  MAS algorithms, which could significantly
  decrease development time and at the same
  facilitating the verification process

7
TFG Q3 Are Agents a Better Way of Introducing FM
to Industry?

• Conceivable
• If it turns out that MAS techniques indeed
  facilitate the development of certain kinds of
  industrial systems that require dependability
• If we deliver MAS techniques that have been
  verified or tools that facilitate verification of
  MAS in a way that is practical/cost-effective
  enough for Industry
• Can we be certain that will happen?
• Not as yet, I think!
• This is a good discussion point

8
TFG Q4 Will Technological Transfer be Easier due
to FM for MAS?

• Again, that is conceivable
• If we manage to deliver formal methods that help
  in the development process of MAS
• If it turns out that MAS techniques are
  particularly suitable for a class of applications
  which require dependability and we deliver
  suitable FM
• Can we be sure that will be the case?
• Not as yet, I think!
• In fact, MAS are already making it to Industry
  for ordinary applications see AgentLink News
  16
• This too is a good discussion point!

9
So, shall we depend on dependability for
Industrial Uptake?

• I do not think we have to depend on FM for MAS
  and the use of MAS techniques for dependable
  system for Industrial Uptake
• In a way, this is happening already, and MAS
  verification is just starting as a research area
• But my guess is (although guessing is never good
  for science)
• MAS will be particularly suitable for dependable
  systems, and
• FM for MAS will result in a major impact on the
  practical development of such systems