Hello Christian, greetings,

I have been using chatGPT recently to rectify grammar, paraphrasing and many other things. Currently I am working on RS-GIS based hazard susceptibility mapping. During this process I found serious issues related to firmness of conceptual background in this open-AI.

In my study area there is strong positive correlation between hazard occurrence and drainage density. But somehow chatGPT took it totally opposite and came up with absurd explanation to the same steering it strangely.

My argument and feed to chatGPT was "The drainage density factor shows that both the total number of pixels and the number of pixels affected by landslides are highest in the class 0-0.348 km/sqkm, and lowest in the class 7.368-9.928 km/sqkm. These extreme classes also show the highest and lowest values of FR and MIV, indicating a strong correlation between drainage density and landslide occurrence."

Then chatGPT explained it as,

"It is common for areas with low drainage density (i.e., less interconnected streams and rivers) to have higher landslide occurrences, as there is more potential for water to accumulate and increase pore water pressure, which can lead to slope instability. This is reflected in the FR and MIV values, which are also highest in the class with the lowest drainage density. On the other hand, areas with high drainage density (i.e., more interconnected streams and rivers) tend to have lower landslide occurrences, as water is quickly drained away, reducing the likelihood of pore water pressure build-up. This is reflected in the lower FR and MIV values in the class with the highest drainage density."

The basic concept that chatGPT is taking into consideration is right but it seems there is profound gap in building interconnectedness throughout various concepts.

Probably I didn't supply enough argument in my statement so that chatGPT could understand fully!

However, I would like to say that, yes, chatGPT is a big support when used with proper scrutiny.

There seems to be a trend of asking too much from these tools (Which are fantastic). Our experience has been that it helps to fast-track some tasks. Data reduction or data processing are examples. This is a professional assessment of the topic, and I recommend looking at it.

https://www.viktor.ai/blog/91/ai-chatGRP-geotechnical-engineering

It seems that because most AI systems (including ChatGPT) are "black boxes" with unclear connections inside, their application in scientific research is limited. In addition, the results of AI application depend strongly on the initial database and, unfortunately, on probable errors in the database.

Thus, the application of AI is useful in some kinds of specific "diagnostics" tasks (classification, separation) or grammar/stylistic text correction, but not in solving scientific problems and generating scientific reports.

Dear researchers,

Please refer to a recent published paper on ChatGPT and Learning Motivation

Article Impact of ChatGPT on Learning Motivation: Teachers and Stude...

All my best!

Hello,

what bugs me about the term "artificial intelligence" is that it is in fact not intelligent. Rather, it is very advanced pattern recognition and imitation but without understanding in any meaningful sense. While that is a necessary condition for intelligence, it is not sufficient by itself. (But it can very convincingly present itself as intelligence to the layperson in any particular topic.) Because of this, the AI's answers are only ever going to be as good as the input data base, at least until there are some more major advances in AI. That means that AI is probably very good at explaining things that you would find in many textbooks but relatively untrustworthy for highly specialized, let alone controversial topics.

One use case I have recently been pondering, is literature research. If the AI has access to all research papers, it should be able to deliver decent review papers on many topics that would be very useful for teaching purposes but also to get a quick overview over a topic that a researcher is starting to venture into. For example: "AI, please write a review about the current state of the art on porphyry molybdenum deposits. Cite all sources properly and include hyperlinks to all sources. Include a table of the 20 largest molybdenum deposits with total locality, country, total reserves, annual production for the past 10 years, and average ore grade". Or "compile all available trace element data on peralkaline granites in a table". Or "list all localities of syenite in Canada with available age data". Or something like that, you get the picture.

I have not tried this myself and am not sure if it would work as expected at the moment but this is the kind of application, where I see the biggest potential for research "services" in the near future. These kinds of reviews would not be authoritative and would not replace review papers by actual experts but they may provide a good jumping off point for starting PhD students, science journalists, and even for experts who want to make sure that they do not miss recently published relevant works (in an increasingly dense forest of journals) when writing an introduction to their next paper. It may also be useful for bringing attention to topically related research from other fields (physics, chemistry, material science...). Cross-citation between disciplines is far rarer than it should be, mostly due to the fact that humans cannot keep track of the enormous amount of publications and need to focus on the outlets they perceive to be the most relevant. The same is true for certain methods that are historically popular in certain fields but may be very useful to other fields, if only the people working there knew about them. This leads to a similar kind of bubble formation as we observe in general public discourse on social media. AI-assisted literature search and condensation may help mitigate this, I hope.

Anselm

I recommend to read this manuscript.

ChatGPT in Academic Writing and Publishing: A Comprehensive Guide

Scientific writing is a difficult task that requires clarity, precision, and rigour. It also involves a large amount of research, analysis, and synthesis of information from various sources. However, scientific writing is also hard, time-consuming, and susceptible to errors. Advanced artificial intelligence (AI) models, such as ChatGPT, can simplify academic writing and publishing. ChatGPT has many applications and uses in academic and scientific writing and publishing such as hypothesis generation, literature review, safety recommendations, troubleshooting, tips, paraphrasing and summarising, editing, and proofreading, journal selection, journal style formatting, and other applications.

In this book chapter, we will discuss the main advantages, examples, and applications of ChatGPT in academic and scientific writing from research conception to publishing.

Read online.

https://achtago.com/publications/list/chatgpt-in-academic-writing-and-publishing/

Download full text

https://www.researchgate.net/publication/369817340_Chapter_2_ChatGPT_in_Academic_Writing_and_Publishing_A_Comprehensive_Guide

As an AI language model, I can provide general tips for writing scientific, geological, and geotechnical reports. Here are some guidelines that may be helpful:

Overall, writing a scientific, geological, or geotechnical report requires attention to detail, accuracy, and clear communication of technical information.

I attempted to see if ChatGPT was able to find any relevant information on my study area, and it was extremely basic, not detailed, almost irrelevant information. Therefore I would not recommend it. I have not tried entering information that I have collected and then turning it into a report. Which may work, but I just do not have faith in the system to retrieve enough relevant information on its own.

I have tested ChatGPT with a couple of very specific mineralogical questions.

1. It hallucinated blatantly imaginary compositional endmembers of a Ta-Nb mineral group when asked.

2. It made up incorrect P-T stability fields for metamorphic minerals. When asked again for a correction of this data, it came up with further wrong data.

The link to published literature did not seem to exist in either case.....!

What about using some natural intelligence instead?

Regardless of the quality of the geological reports, be 100% that higher managment of any company will involve AI and ChatGPT in daily routines! The motiv is the same as always - to reduce costs at any cost! So, prepare yourself for even more difficult future, as now there are tools nearly capable to replace human geoscientist (with exception of field work)! 🙃⛏

I tried ChatGPT for some suggestion regarding who wrote what on a specific topic, and to suggest some literature regarding that specific subject. As a bonus, it invented a lot of papers that I could not found as they were never written... I guess I might wait till they are also written by an AI Ph D. student.

In reference to my post of 22nd June, I tried the same 1st question (Ta-Nb mineral group endmembers) with the AI engine that goes with MS Bing. Surprisingly, the Bing generated AI answer was much more accurate and sensible than that of the earlier attempt with ChatGPT.