Ai and 7671

The trouble with ai is that it states all as fact.
A colleague used ai to write a report on why solar and evse should not be on the same rcd.
The report looked to a layman as clear, factual and well explained 
A closer look revealed different editions of bs7671 had been quoted.

on one occasion it suggested safety maybe enhanced if a gfci was installed.

I suppose the way the ai is questioned will also influence the results.

Having said that I sometimes use ai and search engines to look up regulations but I always verify in the book after.

There is a general problem of creating a 'hollow professional', who does not know his/he speciality but is very good at looking stuff up quickly. 
To an extent this already happens, as even in the old days, for a particular cable type, we knew there might be a voltage drop problem, but had to either stand there and do a sum, or send 'the lad'  back to the van for the tables, if we want to be sure.
As things become less standard. more complicated electronics means more sorts of RCD, and and as per another post, switch-mode supplies mean that some loads are constant power, others are more like a resistor etc, there are more things to know, and then there are slightly arbitrary things that just have to be remembered (2.5m sockets in bathrooms, could have been 3m or two and my disappear one day), and eventually the cognitive burden becomes unreasonable for those without the memory of an elephant,  like myself. (that comma matters).

Then it becomes important to know enough of the fundamentals, to avoid foolish errors like the selection of unsuitable materials or cable routes, and then where to go to look up the final numbers. AI; and computers more generally certainly can have a big role in that, but moderated by training/knowledge and experience that has to rest between the ears of the candidate.

That legal link is quite amusing- the authentic looking 'hallucination' problem of AI has been described more pithily in other places,  including the following genuine academic reference ..

edit - the forum Software  won't let me post this link  directly so there is a tiny URL behind it.

https://research.kent.ac.uk/trust-moral-machines/learning-from-ai/

the problem is real, so perhaps we should call it what it is ;-)

(seems we cannot, at least on this forum)
Mike

mapj1 said:

There is a general problem of creating a 'hollow professional', who does not know his/he speciality but is very good at looking stuff up quickly.

Which is, in a sense, how open-book assessments (such as C&G on BS 7671) work.

I suppose that in any trade or profession, a worker knows how to do 90% of the work because it is the same day-in, day-out. Occasionally, reference materials have to be consulted. If it is an important matter, a competent and conscientious worker needs to be able to go back to the original (be it a standard, a legal case, or an article in a peer-reviewed journal) because secondary references (tutors remarks, course materials, text books, and above all, 'AI') can and do get it wrong.

mapj1 said:

There is a general problem of creating a 'hollow professional', who does not know his/he speciality but is very good at looking stuff up quickly.

Which is, in a sense, how open-book assessments (such as C&G on BS 7671) work.

I suppose that in any trade or profession, a worker knows how to do 90% of the work because it is the same day-in, day-out. Occasionally, reference materials have to be consulted. If it is an important matter, a competent and conscientious worker needs to be able to go back to the original (be it a standard, a legal case, or an article in a peer-reviewed journal) because secondary references (tutors remarks, course materials, text books, and above all, 'AI') can and do get it wrong.