• System fault-tolerance has been addressed by means of replication of com-
ponents to avoid single critical failure points; but the determination of faulty
states to trigger re-configuration has been a problem of increasing importance
in correlation with increased system complexity. Fault detection and isolation
methods have developed sophisticated model-based reasoning mechanics to
do these tasks. The models used, however, are specifically tailored to the task
—a common problem elsewhere.
• Cognitive systems research has put consciousness back into the agenda after
many years of ostracism G.A.Mandler (1975) and hence it is addressing the
question of computer-based model building of this phenomenon.
• Information systems security —regarding human intrusion and the several
varieties of exo-code— has become concerned about the question of self/nonself
distinction in ICT systems Kennedy (2003).
• Information systems exploitation is fighting the scalability problem in main-
tenance tasks trying to mimic the scalable organisation of biological systems
Horn (2001)
In our context, control systems, our main concern is not of human mimicking or
reduction of cost of ownership. The question is more immediate and basic: system
robustness.
There are many technical systems that we depend upon: from the electrical net-
works, to the embodied pacemakers or ESPs in our cars. Dependability is a critical
issue that is being hampered by the increased complexity of individual systems
and from emergent phenomena in interconnected ones.
The justifiable quest for methods for managing reasoning about selves in this
context is driven by the desire of moving responsibility for system robustness from
the human engineering and operation team to the system itself. This is also the
rationale behind the autonomic computing movement but in our case the problem
is much harder as the bodies of our machines are deeply embedded in the physics
of the world.
But the rationale for having self models is even deeper than that: if model-based
control overpasses in capabilities to those of error-based control, the strategy to fol-
low in the global governing of a concrete embedded system is not just recognising
departure from setpoints but anticipating the behavior emerging from the interac-
tion of the system with it surrounding reality.
Hence the step from control systems that just exploit models of the object, to
control systems that exploit models of the pair system + object is a necessary one
in the ladder of increased performance and robustness. This step is also observable
in biological systems and while there are still loads of unsolved issues around, the
ASLab.org / Principles for Consciousness / A-2007-011 v 1.0 Final
13