Are we preparing engineers well enough for the age of AI?

I feel that we should exercise great cautions, just as we would with any other 'latest and greatest', 'solves all problems' style solutions. Carbon fibre should have replaced steel everywhere by now. RAAC (Reinforced Autoclaved Aerated Concrete) should have replaced regular concrete every where.

AI will be great for the sales and marketing brochures. It may even improve them (have you seen most big engineering firm's web sites? - Dull as dishwater, on a good day!).

AI (LLMs) write better (more acclaimed average) essays, stories, narratives, and keeps on theme.

It will only help if it has suitable source material, and that's usually lacking. We are poor at writing down coherent rationales that pass as technically competent. Intellectual property is valuable, so we avoid actually writing down the hidden tricks, or partition the information to keep the tricks away from prying eyes. And often we don't realise that our systems are designed that way [e.g. customer drawings and corporate work instructions].

We are meant to learn from our mistakes but the write ups are often rather sparse, so AI will make the same mistakes with little to learn from. (see Sleipner A for one of the early advantages of CAD stress modelling).

This is an important question, especially for engineers working in power generation, industrial automation and other safety-critical environments.

AI can accelerate document reviews, fault analysis, requirements comparison, coding and troubleshooting, but it cannot fully understand the physical process, equipment condition, operating history or consequences of a wrong decision. A technically polished answer is not necessarily a safe or correct engineering solution.

I believe engineers should be prepared in three areas: strong engineering fundamentals, practical exposure to real systems, and the ability to critically validate AI-generated outputs. Training should include not only how to use AI, but also how to identify incorrect assumptions, incomplete data, cybersecurity risks and unsafe recommendations.

Professional accountability must remain with the engineer. AI should support engineering judgement, not replace it. The engineers who will add the greatest value are those who can combine digital tools with field experience, systems thinking and the confidence to challenge an answer that does not match the real process.

Thank you, Philip. Your point about source material is very relevant. Companies want faster delivery, but much of our engineering knowledge and past learning is still not properly documented. AI may save time, but its output will only be as reliable as the information available to it.

Thank you, Brian. I agree with the three areas you mentioned. With clients expecting faster service, AI can help engineers complete routine work more quickly. However, engineers still need practical experience and the confidence to question an answer that does not match the actual process.

In a way this isn't a new problem. If you think of AI as having human-like qualities, then it becomes rather like having human assistants. Very enthusiastic and perhaps generally well-read assistants, but ultimately ones that know little about your specific problem or situation and lacking in a lot of "common sense". Perhaps like a teenage work experience person who'd rather ask Google things on their phone rather than put in the wok to understand the fundamentals themselves, but still desperate to give a good impression and put a glossy spin on things to try and make themselves look good.

So why don't engineers have droves of human assistants? Generally it's down to costs - costs of employment of course, but costs of managing them as well - in some situations it's quite easy to reach the point where time spent managing an underling would have been more profitably spent just getting on and doing it yourself. AI seems cheap at the moment (the equivalent of "employment costs"), but mostly that's down to the providers 'loss leading' while they try to build a market - in the mean time they're spending literally billions on data centres. Once the real costs of using AI start to be passed onto the consumer, I suspect we'll be a lot more careful about the applications we put it to.- and as for the "management costs" they'll have to be taken into account as well.

   - Andy.

Your question: How should engineers prepare for the growing use of AI without losing the technical judgement, practical understanding and professional accountability that define engineering?

My contribution: I believe that AI has its own limitations as well which might not be resolved anytime soon. The only possible way in bridging the gap to some appreciable level is by incorporating automation and robots to support engineering activities together with AI.

However, it will still not rule out human interference completely. So, acquiring automation/robotics skills and knowledge could help in my thoughts.

Regards.

AJJewsbury said:

So why don't engineers have droves of human assistants?

Worth a quote from the Software folks: Brook's Law (Mythical Man Month - great book) 

"Adding manpower to a late software project makes it later."

There was an article in one of the IEE journals predicting how much extra communication time that adding extra people took. (If I find it I'll add the reference..)

I presume that the  spell check function when I am now using when preparing reports now has an AI element to it, rather than just highlighting or correcting misspelt words it is substituting them with completely random words, that often bear no resemblance whatsoever to what I actually wrote.

I am having to ensure that I read back through what I have written to check for accuracy, just typing this post I have had to retype several  words, for example “preparing” was altered to “perusing” and I am sure I spelt it correctly.

As for AI electrical certificates, they are going to require close scrutiny. Back when there was the long schedule of inspections, on NAPIT Desktop there was an option to tick all boxes, but then you had to go through and untick the boxes which did not apply. One year my NAPIT CPS assessor passed comment that I had tick the boxes which did to say there was an earth electrode, when there wasn’t one, I said “you know what I have done” he said “you used the option to tick all the boxes, then missed unticking that box. Just go through the schedule and tick them, it reduces mistakes”. AI pre-populating the certificates and EICRs is going to lead to the same happening, it will bulk out the paperwork, but then everything will need to be checked for accuracy and it gets to the point where you might as filled it all out yourself.

As a side note, there is still an article about that assessor, Paul Chambers who was my first NAPIT assessor and went on to be a regional manager, here on the IET website.

electrical.theiet.org/.../

There's a vast array of software algorithms for 'spell checking', along with alternate dictionaries (usually US) for getting the wrong alternate word..

Then there's the grammar checkers that enforce US Chicago style, much the the chagrin of older UK typists and grammar purists.

AI will help, like calculators vs slide rules.. [etc.]