I KNEW someone would point out the instructions [regarding ChatGPT] were rubbish but in fact could be fixed with a relatively simple edit by someone with knowledge, writes Ralph Freeman.
However, if that someone had been made redundant and replaced with an AI computer system, what happens then?
AI generated locking instructions
Many thanks to Keith who, as a boater, quickly spotted the sequence of events produced by ChatGPT for operating a lock were in fact a load of rowlocks! The point I was making in the original article was to someone new to the canals, the set of instructions produced by AI might appear valid?
I originally tested ChatGPT-4 on a problem relating to trains pulling loads up an incline. It was the sort of issue the likes of Stevenson, Brunel and Jessop solved everyday in the 1840's using Newtonian Mechanics from the 1600s!. ChatGPT chose the correct equation but applied it incorrectly due to a schoolboy error. Apparently it didn't know the difference between mass(M) & weight(W) and was in error by the factor of gravity(G) which is approx 9.8 (m/s) in metric units!
Predictive text?
So is this AI program really just a predictive text program on steroids? It is clear from simple tests it may know equations but it has no understanding what they mean! It recognises the words 'canal' and 'lock' etc but clearly does not understand the function of a lock. So as in my original article my advice is to use with extreme care.
If you are writing puff pieces for the national press it's fine, but for serious things where safety is involved —beware. It can speak with forked tongue.
Conclusion
If you want to know how to operate a lock ask a boater that has done a few (thousand?).