[SystemSafety] A Fire Code for Software?

[Peter Bernard Ladkin]

On 2018-03-18 12:13 , paul_e.bennett at topmail.co.uk wrote:
>  I know I have expounded this here before but all
> requirements should be Clear, Concise, Correct,  Coherent, Complete
> and Confirmable (Testable). Developers should accept nothing less, no
> matter what discipline they operate in.
It is a helpful list. But, looking at that last sentence, I suspect you will have trouble with it if
you are in a number of software-based fields, for example, in any field which uses deep-learning
neural networks to get results. Or the kinds of psychometric data gathering and analyses of the
Stillwell sort, which are hitting the papers this weekend because of claims of their use by
Cambridge Analytics.

A year ago I mentioned the work with NN machine learning in FBW, which is a couple decades old (for
example, there is a book by Johann Schumann of NASA Ames with Yan Liu, published by Springer in
2010). Someone installed an adaptive NN in an MD-11 to see if they could control the airplane using
the normal cockpit flight-control devices (stick, rudder and engines) but with physical aerodynamic
control suppressed and just using engine thrust, as Al Haynes managed to do in a DC-10 up to the
last moment at Sioux City. The answer was yes. Writing requirements for that application must have
been a challenge, and once you've done that there are essentially no techniques for assuring that
your adaptive NN satisfies/will satisfy them.

PBL

Prof. Peter Bernard Ladkin, Bielefeld, Germany
MoreInCommon
Je suis Charlie
Tel+msg +49 (0)521 880 7319  www.rvs-bi.de





-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: OpenPGP digital signature
URL: <https://lists.techfak.uni-bielefeld.de/mailman/private/systemsafety/attachments/20180318/7a1cfb88/attachment.sig>