Using AI: What You've Been Saying

Article by Adam Duckett

We asked members of TCE’s reader feedback panel to share their experiences of generative AI

Asked what specific applications they used generative AI for in their field of engineering, the most common answers include prompting AI to write and optimise code; to summarise technical topics and research findings; rephrase writing; and generate ideas...

“[To generate] an engineering project investment proposal and some stoichiometric calculations.”
Chartered member, director

“I use generative AI in helping me craft technical storylines…that make complex engineering concepts more accessible to a wider audience.”
IChemE Fellow, commercial head

“To generate a process safety checklist.”
IChemE Fellow, retired engineering professor

“[Asking] a general question like ‘what are typical engineering deliverables to support a class 3 cost estimate in projects?’ I restrict the use of AI to this type of question as I have discovered that AI still makes fundamental engineering mistakes with specific closed questions.”
Chartered member, projects engineering manager

The majority say they have seen improvements in efficiency since incorporating Generative AI into their workflow:

What you’ve been saying: About AI use policies

Those whose employers do have a policy said:

“General instruction is that it is not to be trusted for technical work.”
Chartered member, principal process engineer

“ I cannot use [it] for university work.”
Affiliate member

“It is up to each staff member to use it in teaching and research as they see fit. A staff member can put restrictions on its use in the units they teach. The general feeling for staff and students seems to be, learn how to use it well, but declare when and how it was used.”
Associate member, senior lecturer

“Use discretion and do not input confidential (client) information or organisational knowhow. Treat data sharing as you would outside an NDA – assume whatever you input will become public knowledge. Do not use to solve problems (who owns the IP generated?) and do not use it to write client facing reports.”
IChemE Fellow, special matter expert

“I know it has one, but I don’t think even the company knows what it says and it certainly hasn’t communicated it coherently.”
Chartered member, nuclear safety engineer

“Never read it.”
IChemE Fellow, consultant contract engineer

What you’ve been saying: About the accuracy of generative AI

How do you ensure that the Generative AI tools you use are reliable & produce accurate results?

“I take the approach that the GenAI tool is a bit like an uncooperative but enthusiastic team member. They contribute, but their contributions need to be validated!”
Associate member, academic

“The capability for generating code is absolutely spectacular – rarely have more than a few lines required tweaking before implementation.”
Chartered member, facilities engineer

“I think its reliability could be improved by [me]becoming proficient in prompt engineering, but I have not yet done that.”
Associate member, senior lecturer

“[Conduct a] comprehensive review of the resulting output. You still need to apply some engineering principles that perhaps are misrepresented or misconstrued.”
Chartered member, director

What you’ve been saying: About AI generating unreliable answers

“I was prompting ChatGPT to answer some technical questions, and it got some parts wrong. I corrected it, and it amended its responses as per my corrections. I then tried to deliberately trick it, which worked. The lesson I learned is that the user can convince the GenAI that it is wrong, even when it is not. This means that a non-expert validator is just as dangerous as the GenAI itself in terms of how easy it is to generate false data.”
Associate member, academic

“I tested ChatGPT on a PSV sizing calculation out of curiosity and it made significant errors, including ones that would be dangerous if it were used by someone who didn’t have the background to understand the output. The risk of ‘garbage in, garbage out’ is greatly increased and is a concern for the engineering discipline if not used correctly.”
Chartered member, projects engineering manager

“Incorrect data, non-working code, false statements. All of these have been sufficiently obvious that they are picked up easily on checking. It would be reckless to implement an AI-generated solution without verification.”
Chartered member, facilities engineer

An example of ChatGPT getting it wrong...

Johan Loedolff, a project process manager, replied to our survey saying he’d seen generative AI make calculation errors. He kindly sent us an annotated example to share:

“The test was to see if ChatGPT could provide a worked example of a blend calculation between two ingredients. The Pearson square is used to calculate the ratio of two blended ingredients. It was also very commonly used in the dairy industry to determine the ratio between skimmed milk and cream from whole milk separation. While the initial formula it gave was correct, it started generating errors when it substituted values into the formula. This is simple math.

“Care should be taken to validate ChatGPT’s execution of simple formula substitutions as there seems to be a misconnect between the variables known and unknown.”

We asked AI image generator Craiyon to draw “the future of AI in chemical engineering”. This is its answer. Visit Craiyon.com to generate your own images

What you’ve been saying: About the role you expect generative AI will play in the future of engineering

“It will begin to displace engineers’ input at detail level in many tasks, particularly design, including vendor liaison.”
IChemE Fellow, lead process engineer (independent contractor)

“By questioning the engineer and seeking to establish the scope [of the design project], the engineer will be able to ‘home in’ on the optimum solution by being guided by the AI model.”
Chartered member, retired

“It will be used to employ less able and cheaper people – like most technologies!”
Chartered member, process consultant

“It [can] make us more efficient...and possibly to help us achieve a four-day week rather than companies exploiting the benefits of AI to increase workload. Work is already making us sick.”
Chartered member, senior technical advisor

“Generative AI may produce more resource-efficient solutions to global challenges. From a ChemEng perspective, this may include efficient identification of chemicals to assist with carbon capture and storage, food...and better utilisation of human and industrial waste.”
Chartered member, senior inspector

“Another great tool to take away the mundane tasks (like Excel relieved us of hand calculations) so we can be more proactive and creative – why most of us got into ChemEng to start with.”
IChemE Fellow, special matter expert

“I think that the barrier will be ultimately legal: who takes responsibility for an AI-generated incident?”
Chartered member, Senior process engineering

Article by Adam Duckett

Editor, The Chemical Engineer

Recent Editions

Catch up on the latest news, views and jobs from The Chemical Engineer. Below are the four latest issues. View a wider selection of the archive from within the Magazine section of this site.