OpenAI’s ChatGPT applied to tailings dams and associated risks
Jan 11th, 2023
As everyone else, we got excited about the new ChatGPT so we tried OpenAI’s ChatGPT applied to tailings dams and associated risks.
OpenAI is San Francisco-based research laboratory. Its founders and donors are, among others, Elon Musk and Peter Thiel, along with technology companies like Microsoft. ChatGPT is a publicly available free app that generates “human-like” written replies. The acronym GPT stands for “Generative Pre-trained Transformer” .
Note that our colleague Mike Porter started a thread on ChatGPT on LinkedIn. Numerous members participated and we particularly enjoyed the contribution by David Espinoza who, in our mind, nailed the subject 100%. Indeed he stated that:
“ChatGPT can only generate content based on existing information. If the concepts are wrong, it will amplify the incorrect ideas.
The fact that a theory is generally accepted doesn’t make it correct. The wildly inaccurate Ptolemaic system (a geocentric model of the universe) lasted well over a thousand years until Copernicus replaced it with a heliocentric model. If we had ChatGPT then, it may have lasted another millennium. “
Our trial started with the topic of “quantifying the likelihood of failure of a tailings dam”. This topic is very dear to Riskope as we have spent years of research and development advancing this question with ORE2_Tailings™. Remarkably, the question is even the subject of a Round Robin exercise our colleague Ryan Singh has promoted on LinkedIN. The idea behind this Round Robin exercise is indeed to:
“Assess if there is consistency in the quantification of likelihoods of two failure modes of confining structures of TSFs, when the assessment is carried out by various individuals or teams”.
We will soon discuss that in this blog.
Below a review of some of the questions we asked and the replies delivered by ChatGPT.
Q (Riskope) can you quantify the likelihood of failure of a tailings dam?
OpenAI answer: It is not possible to quantify the likelihood of failure of a tailings dam with an exact numerical value. However, the likelihood of failure can be estimated by evaluating the structural integrity of the tailings dam and its risk of failure, based on factors such as the materials used, the design of the dam, its existing condition, the geology of the area, rainfall, earthquake risk and any other external factors that may affect the stability of the dam.
Interesting answer indeed. OpenAI rightly states that uncertainty should be embedded in the quantification and therefore an exact number is preposterous. At Riskope we always deliver ranges for that very reason.
However OpenAI then confuses the term risk and probability/likelihood. Indeed confusing terms is a systemic issue in the risk management space. Let’s not forget that a risk is defined as a function of:
- consequence of a failure multiplied by
- it’s likelihood of failure.
and not as a mere probability. We are pushing to use proper glossary in all our report and paper as a large part of misunderstanding comes from confusion from the word used.
After this broad question on OpenAI’s ChatGPT applied to tailings dams and associated risks we asked a general and purposely open question to gain a more general feel.
Q (Riskope): are tailings risky?
OpenAI answer: Tailings can be risky depending on the type of mine, its location, and the type of material it’s processing. Potential risks associated with tailings include water contamination, air pollution, and potential health hazards. In addition, tailings dams can pose a safety risk due to their potential to fail or collapse.
This is a promising answer. However, we would add that ancillary water management, construction, and monitoring are also very important factors to the risks generated by a tailings dam. Indeed, they all alter the likelihood the probability of failure of a tailings dam, thus the risk.
Like others have done in the thread initiated by Mike Porter we then asked for a list of references OpenAI’s used to formulate the previous answers. ChatGPT is based on its last scan of the internet, thus it is blind to the latest news and may indicate out of date references.
Q (Riskope): Can you please prepare a list of references that helped shape your response?
Generally the OpenAI list four sources, but we don’t know which part of the answer above is answered by the four listed references. Indeed, this is an issue in itself. We note that the four references presented issues as discussed below.
ChatGPT answer:
- “What are Tailings Dams and What Risks Do They Pose?”. World Bank Group. Retrieved October 9, 2020.
Looks made up, doesn’t exist with this exact title.
- “Tailings Disposal Options Study for Sangan Iron Mine Project, Iran”. International Council on Mining and Metals. Retrieved October 9, 2020.
This one exists but from UBC from 2011 https://open.library.ubc.ca/soa/cIRcle/collections/59368/items/1.0107714 and not from ICMM.
- “Mining Waste Management Market by Mining Method (Surface and Underground). Metals/Minerals (Thermal Coal, Iron Ore, Gold, Copper, Nickel). Waste Type (Overburden/Waste Rock, Tailings, and Mine Water), and Region. Global Forecast to 2022”. Markets and Markets. Retrieved October 9, 2020.
This one is too generic … sure the reference exists but it doesn’t really help answering the initial question…https://www.marketsandmarkets.com/Market-Reports/mining-waste-management-market-104930371.html
- “Tailings dams: safety is no accident”. International Council on Mining and Metals. Retrieved October 9, 2020.
Correct reference: Roche C, Thygesen K, Baker, E (eds.). 2017. Mine Tailings Storage: Safety Is No Accident. A UNEP Rapid Response Assessment. United Nations Environment Programme and GRID-Arendal, Nairobi and Arendal, ISBN: 978-82-7701-170-7 http://www.grida.no/publications/383
Since the correct reference has a different title than the one listed by openAI the user may not find the correct material.
Conclusion on OpenAI’s ChatGPT applied to tailings dams and associated risks
Like a score of colleague and other professionals, we were excited to use OpenAI’s ChatGPT applied to tailings dams and associated risks.
The type and quality of answers we got varies greatly. ChatGPT is certainly a promising tool for certain use if its limitations are properly considered.
As a matter of fact, given that the system learns from information gathered in the web, the “knowledge” should be considered as generally accepted information at the date of the last scan, which is nether necessarily right or good.
When asked precise question OpenAI is afflicted by systemic issues similarly to those afflicting human beings. The answer may reflect popularity but be wrong. As a result, if an idea is commonly accepted but wrong OpenAI’s ChatGPT will promote. The result will be stagnation that will kill innovation, as pointed out by David Espinoza. Remarkably, an innovation tool like AI may have a completely opposite effect and “block” forward thinking ideas.
Furthermore, the reference function seems to “create” references and thus is difficult to use for in depth research or critical knowledge building.
ChatGPT is reportedly percolating into many areas such as legal, IT, science and education.
If one day such systems will be considered the “knowledge base”, or “common practice summary”, or worst “best practice” and they will find a way into universities and later courts of law… then we will have a serious problem!
Tagged with: decision, management, Risk Management
Category: Risk analysis, Risk management
Leave a Reply