From Black Mirror to The Matrix, warnings about the danger of AI taking over the world have overwhelmed popular culture. Especially since the release of ChatGPT, a large language model, we’ve seen several warnings about the potential threat of AI stealing human jobs. This principle reminds me of the introduction of Google before it was a verb. “How will kids effectively learn without using an encyclopaedia?” “Will people stop reading books altogether?” These seem silly now, as Googling the small problems gives us space and time to think about bigger tasks.
As a scientist, I value time to think about important questions, problems, and solutions. When I have to resort back to relatively thoughtless tasks, sometimes it gets frustrating. Over the last several months, I have used ChatGPT almost every day, because “he” makes a great research assistant. Here are five ways that I use ChatGPT on a daily basis to make myself a more efficient and effective scientist.
He performs simple calculations for dilutions, solutions, and reagents.
I’ve made more reagents in-house than I can count over the last 10 years, and all of them need to be accurate so that experimental outcomes can be trusted. It is very easy to move too quickly and miss a “carry the one” or divide incorrectly. Now, I never have to worry about making a human error when it comes to managing reagents and buffers. ChatGPT explains how he comes to the answer so checking his math is easy.
He writes code snippets, interprets errors, and edits your code to make it better.
In the near future, when AI is integrated across a wide range of platforms, software developers will think only at higher-orders of problem solving. Coding will not be as tedious as to write code and memorize syntax of various programming language, but rather on the synthesis and flow of logic and information. I never start my code from scratch anymore. Certainly, ChatGPT’s coding mind is shallow, and larger tasks are more likely to fail. But if you are building a pipeline, or if you run into a problem that you would Google, ask ChatGPT to start you off with some suggestions. Its effectiveness varies across systems (it is way better at Python than R, for example), but you no longer have to construct the logic of a for-loop the way you did in school. As with any other application of ChatGPT, the answers are only as good as your prompts.
He is a sounding board for brainstorming and articulating ideas.
As a scientist, most of my ideas are bad. A few of my ideas are good. It is actually quite difficult for me to distinguish the difference between a good idea and a bad idea just by allowing my inner dialogue to debate. Nearby colleagues often help with scientific banter, but usually they have other stuff to do. ChatGPT is excellent at helping to flesh out ideas for anything from an experiment to a grant. His ability to synthesise ideas is limited, so I would really just consider it a smart sounding board, it is certainly no grant-writing machine.
He retrieves information quickly and conveys it concisely.
Most of the little questions that I would directly type into Google, I instead type into ChatGPT. This applies to protocols, reformatting raw data, or small pipeline adjustments. In 2022, I would type a question into Google and browse ResearchGate or Quora for other scientists that have asked and answered this question before. There is usually two to three right answers among a sea of unhelpful tips. ChatGPT gives me all of them and saves me the mental fatigue to find them myself. It will also include limitations or variables for you to consider and decide for yourself what works best.
He can read a scientific article.
As I’ve started my postdoctoral work which is in the depths of cellular and molecular biology, I find myself needing to read papers that are extremely long. One way to gleam a summary is by reading the abstract, but I usually select papers by reading the abstract alone. As part of a new deep-dive into a particular paper, I will start by copy-paste the main text and asking ChatGPT to draft up a summary. ChatGPT also allows some room for a focused summary, if you have a particular interest in the results from one experiment or wondering what methods were performed in the study.
Kaitlin P. McCreery, Ph.D. 
Leave a comment