RobustInstructionSet
'Definition:' A robust instruction set is one which is not broken by small changes but instead makes 'small' changes in the immediate outcome of its execution. The prime (and only?) example is DNA. Small changes have small effect locally. (Though a small change to a subsystem at a high meta level such as a system producing a growth hormone can have a large effect overall on the phenotype.
Discussion
An instruction set that uses error correcting code to recover in most cases is NOT robust by this definition. It either makes no change at all or breaks beyond a certain level of change.
A RobustInstructionSet probably has to be based around stochastic behaviour - that is its outcomes are the aggregate of many small events. Even without mutation to a program the output of a program will not be 100% reliably the same from one run to the next.
The Java machine nearly meets this requirement. It enforces object protocols ensuring correct behaivior of well defined objects. Exceptions are signalled allowing recovery. But a recovery methodology is needed to complete the picture. DNA employs numerous error correction stratgies at various levels of organization.
Moreover DNA is far less susceptible to single bit errors. Damage generally degrades behaviour gracefully - and may lead to interesting new behaviour. By contrast single bit errors in the encoded Java instructions will generally lead to broken code, e.g. code which loops forever.
A CommunicationMachineArchetechure, like the HumanBrain, DNA (EvolutionaryGameTheory), and the universe itself (InformationPhysics) could have these properties. It would need some organizing priciple, like Minsky's SocietyOfMind. It would also be easily scalable and would no doubt come up with FortyTwo :)