[SystemSafety] AI and the virtuous test Oracle - action now!

[Peter Bernard Ladkin]

On 2023-07-03 03:44 , Les Chambers wrote:
>  > [PBL] Except for IMO, regulation occurs at the level of nation-states, or the EU
> ....
>>>>>>>>
> [LC] I am confused. Taking the case of IEC 61508, the IEC Mission Statement states:

You (and all system safety engineers) need to distinguish between regulation and standardisation. 
The IEC is not a regulatory body. Regulation is law; law is generally set by the law-giving bodies 
of nation-states (with the exception of the Law of the Sea, as I noted). A standard is an agreement 
between named persons and/or organisations to do things a particular way. That agreement may or may 
not be part of law in any given nation-state.

There is a British standard for a cup of tea, BS 6008:1980. However, AFAIK there is no regulation 
concerning cups of tea in England. Or Scotland, Wales or Northern Ireland for that matter.

Standards may become part of regulation. In Greece, the standard IEC 61508 is part of regulation 
because (I understand) the law of Greece establishes it explicitly as such.

Or not. In GB, the regulation is the requirement to reduce risks to be "as low as reasonably 
practicable" (following a judgement of Lord Asquith from 1949). Notice thereby that the regulation 
is thus ALARP in (many, most) common law domains such as Australia. ALARP is not a normative part of 
IEC 61508. So here the standard and the regulation are clearly different.

As they are different in Germany, where ALARP is not a regulation. The regulation is MGS, "minimally 
the same level of safety": the safety-related aspects of any system must provide at least the same 
level of safety in its operations as was achieved beforehand in those operations. MGS is not a 
normative principle of IEC 61508 either.

Indeed, one can expect that no country-specific regulations are going to make it into a standard. If 
country A has Reg.A and country B has Reg.B and they are not identical, and country A and country B 
are participating members of the standards committee in question, then country A's appointed experts 
on the standards committee are not going to agree with Reg.B being written into the standard, 
because they must use Reg.A and not Reg.B, and mutatis mutandis for country B's appointed experts. 
Since consensus is required on the contents of a standard, that will exclude Reg.A and Reg.B. That's 
why neither the GB reg nor the German reg governing the safety of engineered systems are a normative 
part of IEC 61508.

But that can also be turned around. In GB, the regulator HSE is said to use the IEC 61508 standard 
as a guide to whether risks have been reduced ALARP. As far as I know, that interpretative procedure 
has not been tested in law, so it is not completely clear that it holds.
> The terms “worldwide use” and “International standards” imply International  > scope. The term “Conformity Assessment” implies regulation.

No, it doesn't. You can conform or not conform to a standard, without that standard being part of 
any regulation. As with cups of tea.

Rather, the use of the term in IEC documents likely refers to the four associated Conformity 
Assessment Bodies. They have nothing to do with law (unless that is so specified in a specific 
nation-state). See https://www.iec.ch/conformity-assessment/how-global-iec-ca-systems-operate  See 
also IECEE https://www.iecee.org/who-we-are/about-us

Not keeping this straight does lead some people into further confusion. For example, AI safety is 
already regulated in GB (as well as common law countries which attend to GB rulings), indeed since 
1949, before the term was even coined. Risks are to be reduced ALARP.

This, however, is unlikely to suffice to regulate all aspects of the use of ML-based AI which have 
or may have deleterious social impact; just as ALARP has (so far) not been used to regulate harms to 
young people which are said to emanate from (mis)use of "social media". Kids have committed suicide 
in cases in which the English Coroner has determined were causally affected by the child's use of 
social media. AFAIK nobody has been prosecuted for it.

PBL

Prof. i.R. Dr. Peter Bernard Ladkin, Bielefeld, Germany
Tel+msg +49 (0)521 880 7319  www.rvs-bi.de




-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature
Type: application/pgp-signature
Size: 840 bytes
Desc: OpenPGP digital signature
URL: <https://lists.techfak.uni-bielefeld.de/pipermail/systemsafety/attachments/20230703/6badf15f/attachment.sig>