ASM Online Member Community

 View Only
  • 1.  ChatGPT and materials information

    Posted 01-25-2023 14:02

    The launch of the artificial intelligence chatbot ChatGPT in November has prompted a flurry of news items and opinion pieces about its potential use (and misuse) in a variety of domains. The Wikipedia article on ChatGPT provides an overview of the technology and summarizes the responses--both positive and negative--it has generated.

    My colleague @Michael Plishka, ASM's Business Development Manager for Education, decided to try out ChatGPT to see how well it responds to specific queries about materials science and engineering. Following are three of Mike's questions and the answers generated by ChatGPT:

    Which has higher strength: titanium or aluminum? 
    Titanium is generally stronger than aluminum, but it depends on the specific alloys being compared. Pure aluminum is quite soft and weak, while pure titanium is very strong. However, aluminum can be strengthened through various methods such as heat treatment and the addition of other elements, and it can be made into stronger alloys than titanium. Additionally, titanium has a higher density than aluminum, so a titanium object will be heavier than an aluminum object of the same size. 

    What is BCC vs. FCC?
    BCC (Body-Centered Cubic) and FCC (Face-Centered Cubic) are two different crystal structures that can be found in some metals and alloys.

    In a BCC structure, the atoms are arranged in a cubic lattice with an atom at the center of each cube. The atoms at the corners of the cube are shared by eight different cubes, and each atom touches four other atoms. The BCC structure is often found in metals that have a high melting point, such as chromium, tungsten, and iron.

    In an FCC structure, the atoms are also arranged in a cubic lattice, but this time there is an atom at the center of each face of the cube in addition to the atoms at the corners. The FCC structure is often found in metals that are more ductile and have a lower melting point, such as aluminum, copper, and gold.

    Both the BCC and FCC structures are common in metals, but the specific structure that a metal adopts depends on its atomic arrangement and the conditions under which it is cooled.

    What is the best etchant for brass?
    There are several etchants that can be used to etch brass. The best etchant to use will depend on the specific type of brass you are working with and the equipment you have available. Some common etchants for brass include: 
    1. Ferric chloride: This is a commonly used etchant for brass and other copper alloys. It is relatively cheap and easy to find, but it can be messy to work with and has a strong smell.
    2. Ammonium persulfate: This is another commonly used etchant for brass. It is relatively fast-acting and easy to work with, but it can be more expensive than ferric chloride.
    3. Nitric acid: This is a strong etchant that can be used to etch brass, but it can be dangerous to work with and can produce hazardous fumes.
    4. Hydrogen peroxide and sulfuric acid: This is a commonly used etchant for brass and other copper alloys. It is relatively fast-acting and easy to work with, but it can produce hazardous fumes and can be corrosive to some materials.
    It is important to follow all safety precautions when etching brass, as some etchants can be dangerous to work with. Be sure to wear protective gear, such as gloves and goggles, and work in a well-ventilated area. 

    Mike and I are very interested in receiving feedback from the ASM Connect community:

    • What do you think of these responses generated by ChatGPT?
    • Do you have ideas for potential applications of chat bot technology in materials science and engineering? What are some of the possible benefits and pitfalls of these uses?


    ------------------------------
    Scott Henry
    Senior Content Engineer
    ASM International
    ------------------------------
    Education courses


  • 2.  RE: ChatGPT and materials information

    Posted 01-26-2023 09:32
    ChatGPT always amazes me. I think Google is rightly concerned about how this can upend internet searching!

    Clearly, it can provide custom content from a direct natural language query, which would be very useful for visitors to ASM's website. What's not clear to me about ChatGPT is whether we can provide it custom content as part of the training. Since I assume that OpenAI wasn't able to scrape most of ASM's content for underlying training corpus, we could train a version of ChatGPT that could answer questions with the added information from ASM's content.

    However, I have concerns about putting this as chatbot on the website. Even giving it facts and such, ChatGPT can be factually wrong. From my understanding, the funny thing about the model is that it's designed to make things that appear to be right. It doesn't get extra marks for making content that actually is right. So asking it something like "what is the bulk modulus of Ni3Al" could be problematic. It also can't provide references. I read that it actually makes fake links to websites that appear like real links when you ask for references, haha. 

    ChatGPT is super exciting technology. I'm also eager to see how we can use it.

    ------------------------------
    James Saal
    Director - External Research Programs
    Citrine Informatics
    ------------------------------

    Education courses


  • 3.  RE: ChatGPT and materials information

    Posted 01-26-2023 10:58
    Actually, one interesting use case I stumbled upon is to convert from a natural language query to a SQL query. So it could be a good interface for searching for data in ASM's database?

    ------------------------------
    James Saal
    Director - External Research Programs
    Citrine Informatics
    ------------------------------

    Education courses


  • 4.  RE: ChatGPT and materials information

    Posted 01-27-2023 11:31
    Edited by Paul Mason 01-27-2023 11:32
    One of my colleagues asked ChatGPL to see if it could write an abstract a couple of weeks ago. Superficially it did a good job and seemed like a well organized review. However, in the body of the text it referred to various (experimental) studies.  When my colleague asked them it for the references cited, it responded, "I apologize, but I am unable to generate a full reference list for the literature review as the studies I mentioned are fictional and not exist in the real world." It then went on to suggest some real studies that had been performed on the material that we were referring to. 

    This is similar to what James is saying above. Based on this I think it should be used with caution and should always be double-checked for accuracy.

    ------------------------------
    Paul Mason
    Thermo Calc Software
    ------------------------------

    Education courses


  • 5.  RE: ChatGPT and materials information

    Posted 01-27-2023 16:46
    Edited by Scott Henry 01-27-2023 16:46
    James,

    I agree that the potential applications for search and discovery, query building, etc., are intriguing and exciting. I don't think it is currently possible to train ChatGPT using a specific body of content; however, we may not be far away from having similar tools available for such a use. ASM will definitely want to investigate AI tools alongside other methods to enhance search and discovery of ASM content. The ASM Digital Library added the ability to search for images in 2022. Artificial intelligence tools may have a role in enhancing the ability to find data in images as well as in text and tables.

    Scott

    ------------------------------
    Scott Henry
    Senior Content Engineer
    ASM International
    ------------------------------

    Education courses


  • 6.  RE: ChatGPT and materials information

    Posted 01-26-2023 15:53
    Scott,
    An interesting exercise and I was thinking about trying ChatGPT for my next editorial.  It would be interesting to compare it in a review article or paper. I think this is where it could have some applicability, since it is not 'new' research or finding but more data mining of already published papers. The key would be does it uncover the key and seminal papers. Also, does it just cite findings or can a more knowledge -based assessment be made. I do find that many review papers don't necessarily include 'critical thinking', thus could have been better written by a Chatbot.
    Tom
    Editorial-in-chief International Journal of Metalcasting
    fyi - This was not written by ChatGPT

    ------------------------------
    Thomas Prucha
    President
    MetalMorphasis LLC
    Rochester Hills MI
    (810) 423-7564
    ------------------------------

    Education courses


  • 7.  RE: ChatGPT and materials information

    Posted 01-27-2023 14:30
    Edited by Scott Henry 01-27-2023 14:31

    Tom,

    From what I have seen, ChatGPT was trained using a very broad range of sources, but probably not all of the relevant scientific literature related to a given review topic. It may do a better job with fundamental topics than with breakthrough research. As noted by @James Saal, its training seems to have favored generating plausible-sounding content, whether or not it is factually accurate. Anyone using it (ethically) to create content would need to scrutinize and verify the validity of the output--echoing the comment from @Paul Mason.

    The journals world is circling its wagons, with publishers defining limitations in-and in some cases, prohibitions against--using tools like ChatGPT in writing journal articles.

    One use that seems valid (to me) in content development would be for authors to compare their own (human-generated) unpublished drafts against output on the same topics from ChatGPT--to see if the tool turns up information or context the authors may have missed or possibly misconstrued.

    Scott

    ​​​

    ------------------------------
    Scott Henry
    Senior Content Engineer
    ASM International
    ------------------------------

    Education courses


  • 8.  RE: ChatGPT and materials information

    Posted 01-28-2023 15:56
    The answers provided as examples are not wrong, but they are not as accurate and helpful as should be expected from knowledgeable professionals.

    ------------------------------
    John Grubb
    consultant
    New Kensington PA
    (724) 448-5272
    ------------------------------

    Education courses


  • 9.  RE: ChatGPT and materials information

    Posted 02-07-2023 11:38
    I appreciate the conversation here.  Technology to pay attention to and approach with caution.  Another interesting use case: asking for a material property and then asking for it in a syntax for a simulation code's input file.  Here Ti64 Yield Strength for a COMOSL simulation.  Note that the AI "understands" the need for Young's Modulus and Poisson's Ratio, etc., to fully define the "material card".


    Chat GPT Example Text and Code


    ------------------------------
    Mike Plishka
    Business Development Director
    ASM International
    Novelty OH
    (440) 671-3819
    ------------------------------

    Education courses


  • 10.  RE: ChatGPT and materials information

    Posted 02-07-2023 15:54
    The ChatGPT responses are scarily good. AI certainly augments human capabilities. Subject matter expertise is still important to catch obviously bad responses from ChatGPT

    ---------------------------------
    Chandrasekhar Narayan
    Director, Science and Technology
    International Business Machine
    San Jose CA
    (408) 608-4964
    ---------------------------------



    Education courses