[SystemSafety] Does "reliable" mean "safe" and or "secure" or neither?

[Peter Bernard Ladkin]

On 2016-04-24 18:12 , paul_e.bennett at topmail.co.uk wrote:
> We do need to have some general agreement about certain terms. The
> three main words that we should be absolutely clear on are:-
> 
>   Reliability
>   Dependability
>   Integrity
> 
> ..........
> 
> I use the OED definitions for the most part (always one that seems 
> to fit the bill).

One would surely expect professional engineers to use the technical definitions of these terms
established in his area of expertise. In this case, that would be, say, IEC 60050 plus the
definitions in more application-specific standards that have not yet reached IEC 60050. Or the
definitions from IFIP WG 10.4 (the 2004 Laprie set) because we have observed that there are
differences in two of those terms (although I would argue they are compatible).

If you do, that means you think that the OED definitions are conformant with IEC 60050 and IFIP. I'm
not at all sure they are

If you don't, what's the reason you don't?

There is a reason to use technical vocabulary when it fulfils certain coherence conditions. Here it is.

First, a sample of ordinary-language definitions (Cambridge rather than OED because it's free on-line).

> Reliability:
>
> how well a ​machine, ​piece of ​equipment, or ​system ​works:
> [examples]
> how ​accurate or able to be ​trusted someone or something is considered to be:
> [examples]

Here's why this is not very good for electrotrechnology.

Suppose I have software S which fulfils its function brilliantly 99 times out of 100, but the 100th
time it kills 20 people.

And I have software S1 which fulfils its function only 80 times out of 100, but it never harms
anything when it fails.

Is S reliable? That means: how well does S work?
Is S1 reliable? That means: how well does S1 work?

Those questions seem to me irredeemably vague.

Here's a less-vague set of definitions:
Reliability: the extent to which a piece of software fulfils its specification
Safety: the extent to which a piece of software does not induce behaviour which harms or kills
people, animals, other things.

You can now say:
The reliability of S is 99%, but it's horrendously unsafe. Its safety quotient (by which I mean: how
many people it kills averaged over the runs) is 0.2.
The reliability of S1 is 80%, but its safety quotient is perfectly 0. You cannot get a safer piece
of software than S1 (although lots of other software is equally safe).

And you can thereby observe:

S is a lot more reliable than S1, but it is a lot more unsafe in operation

which conveys important information, to users as well as to system engineers, which could not be
conveyed using the Cambridge ordinary-language notions.

PBL

Prof. Peter Bernard Ladkin, Faculty of Technology, University of Bielefeld, 33594 Bielefeld, Germany
Je suis Charlie
Tel+msg +49 (0)521 880 7319  www.rvs.uni-bielefeld.de





-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 455 bytes
Desc: OpenPGP digital signature
URL: <https://lists.techfak.uni-bielefeld.de/mailman/private/systemsafety/attachments/20160425/15fe82ec/attachment.pgp>