Monday, March 13, 2023

Uhhhhhhhhh

Also in this week's C&EN, this letter to the editor: 
What ChatGPT means for chemistry

Recent inroads made in chemistry seemingly have little to do with historical techniques that advanced the central science. The discovery of a new organic reaction or mechanism, invention of click chemistry, development of powerful light sources for X-rays, advent of 3D nuclear magnetic resonance and cryo-electron microscopy techniques, hyphenation of analytical techniques, and ab initio computation to solve chemical and biochemical problems seem “so yesterday.” Today, these rigorous and pejoratively “incremental” techniques have been broadsided by artificial intelligence, machine learning, and programs such as AlphaFold, which has predicted over 200 million structures to date.

AI’s rapid growth has predictably resulted in the creation of an interactive graphical platform called ChatGPT. Its ramifications for those in academia and the chemical craft cannot be ignored.

ChatGPT’s “dialogue format makes it possible for ChatGPT to answer followup questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests,” OpenAI, the company that made the program, says on its website. ChatGPT has revolutionized the search function to upend traditional engines such as Google. Its capacity to improve and accelerate the pace of research is significant. It paves the way for easy and rapid access to tomes of information.

Yet ChatGPT is not designed to substitute human intelligence. Nor is it capable of doing so. It ought to be regarded as a tool that supports, synergizes with, and amplifies researchers’ efforts. ChatGPT’s ability to write abstracts, rectify information, correct mistakes, and even coauthor publications is both welcome and worrisome. While there is no undoing this behemoth, it behooves the student of science and chemistry, be it a high schooler, undergraduate, graduate student, postdoctoral scholar, or academic or industrial researcher, to use it in a manner akin to the library and the CRC Handbook of Chemistry and Physics of the days of yore.

Establishing a culture of responsible use that educates individuals about the capabilities and limitations of ChatGPT is crucial. This can be accomplished by creating rules, best practices, and training programs on how to use ChatGPT responsibly rather than dismissing it, as might be the wont of a purist. Its application should not diminish the significance of contributions that human intelligence and creativity may provide to health, ecological well-being, and general good. While it is incumbent upon us not to subjugate ourselves to it, it is yet another inroad that can take chemistry skyward.

Disclaimer: This article was not written by ChatGPT.

Payam Kelich and Mahesh Narayan
El Paso, Texas
So who knows what ChatGPT will be like in ten years, but what turned me away from ChatGPT as a research tool was learning from Laura Howes that someone emailed her asking for an article that she didn't write because the AI suggested that she did. 

I would imagine that, over time, it would be easy enough to root this out. And, I'm broadly sure that there were errors in the CRC Handbook as well, but I am 100% positive that the rate of errors is far higher for ChatGPT than it is for any text in a typical chemistry library. 

4 comments:

  1. For those unaware, ChatGPT can write you python scripts essentially on demand, which is something I've used it for extensively. If you encounter an error running a script it can even explain the error code in plain english and offer fixes to code (either written by your or it) to rectify the error.

    For me it's lowered the barrier to writing some simple scripts that I'd wanted to develop for my research and I think it's a great tool for anyone looking for help to write something to streamline data processing.

    ReplyDelete
    Replies
    1. FWIW, ACS Nano also has a best practices article already out: https://pubs.acs.org/doi/10.1021/acsnano.3c01544

      Delete
  2. It's easy to root out issues with looking for references now. Namely, they are all wrong. ChatGPT in its current form should not be used for chemistry or referencing... end of.
    For Python, it's great

    ReplyDelete
  3. I can see lazy scientists using ChatGPT to write their papers and consequently introduce more errors into an ocean of literature already littered with errors.

    ReplyDelete

looks like Blogger doesn't work with anonymous comments from Chrome browsers at the moment - works in Microsoft Edge, or from Chrome with a Blogger account - sorry! CJ 3/21/20