ChatGPT and Process Safety Truth
Vitrual assistants can be very helpful ― not because they are right, but because they make us think.
In the year and half since the release of ChatGPT 3.5 proccess safety professionals have asked, ‘Can I trust the answers and information that this virtual assistant provides?’ After all, if ChatGPT or one of its brethren, provides bad information or an incorrect analysis to do with the operation of a refinery or offshore platform then a serious accident may occur ― indeed, someone may die.
The ‘Most Important’ Management Element
I was prompted to think about these concerns on listening to Dr. Rainer Hoff’s paper AI-PSM: Where Are We Now?, presented at the Spring24 Meeting of the American Institute of Chemical Engineers. The paper referred to a question that I posed to ChatGPT 3.5 in the post ChatGPT: A Process Safety Expert. The question, which was dated January 3rd 2024, was,
What is the most important management element in a process safety management program?
The response was ‘Employee Participation’.
Five days later, Dr. Hoff posed the exact same question to ChatGPT. The answer it gave this time was ‘Process Hazards Analysis’.
I repeated the question on April 2nd 2024 and received yet another answer,
No single element is the “most important”.
(This, by the way, is what I would regard as being the best answer.)
So, here we have a dilemma:
When asked the same question on three different occasions, ChatGPT gave three different answers.
The obvious conclusion is that ChatGPT cannot be trusted, and process safety professionals should be very, very wary of its use.
Maybe.
Process Safety Truth
For millennia philosophers have struggled with the question, ‘What is truth?’ The answers that they have come up with may seem irrelevant to the modern, practical discipline of process safety. Yet these old-time philosophers can help us address the ChatGPT conundrum, ‘What is the most important PSM element’.
Broadly speaking, there are two ways of evaluating the truth of a statement. The first is what we now call Platonic, the second is Aristotlean. The picture at the head of this section is a detail from Raphael’s The School of Athens, painted between 1508 and 1511. Plato (on the left) is pointing upward, thus reflecting his focus on abstract ideas of truth and beauty. Aristotle is pointing down, reflecting his focus on observable reality.
Virtually all process safety professionals (most of whom have technical background) are of Aristotle’s way of thinking. The question that they usually ask is, ‘Does it work?’ They expect ChatGPT to provide objectively correct answers.
But what if there is not a correct answer? We have seen how the program gave three different answers to the same question. Rather than saying that the program is wrong, maybe we should recognize that there is not an correct answer to the question, ‘What is the most important management element in a process safety management program?’ Individuals have opinions, why should virtual assistants also not have opinions?
This is where Plato, who lived 2500 years ago, enters the discussion. He suggested that there are ideal ‘forms’ for every object and way of thinking. For example, all chairs are different from one another, but they share a ‘form’ that might be called ‘chairness’. Similarly with process safety management. There are many PSM programs, and they all differ from one another, but there is an ideal (unattainable) process safety program. Our aim as process safety professionals is to understand what that ideal may look like and to strive for its achievement.
All of this sounds like just so much fancy talk to those carrying out the day-to-day work of writing operating procedures, leading PHAs or managing a risk register. But the ChatGPT responses that started this post have profoundly practical implications.
Imagine three companies; the CEO of each company decides to organize their PSM program around those three different responses. The programs would look profoundly different from one another. The question then becomes, ‘Which program approaches the Platonic ideal?’ There is no objective (Aristotelian) correct answer to that question, but one of the three programs more closely approaches the process safety ideal.
Which brings us back to whether we can trust ChatGPT and other AI tools in process safety work. The answer, for now at least, seems to be that we must be very cautious when we are looking for objectively true/false responses such the selection of a material of construction. But, when it comes to management-related issues,
Vitrual assistants can be very helpful ― not because they are right, but because they make us think.
Here are examples of questions that address important process safety issues for which there is no ‘right answer’, but for which the use of ChatGPT is entirely valid.
What level of risk is acceptable?
How do I organize an incident investigation?
Which CCPS Process Safety Beacon reports should I share with my team?
There can only be opinions in response to these questions; ChatGPT’s opinion is likely to be useful and relevant.
Footnote
The above discussion puts one in mind of the famous saying of Alfred North Whitehead (1861-1947),
The safest general characterization of the European philosophical tradition is that it consists of a series of footnotes to Plato.
