There’s also been much written about the drawbacks of the system including its production of outright falsehoods in published articles. The journal Nature reports that the program does not meet the criteria for authorship of scholarly articles “because they cannot take responsibility for the content and integrity of scientific papers,” meaning, the programs. Also: ChatGPT is ’not particularly innovative,’ and ’nothing revolutionary’, says Meta’s chief AI scientist There is another, more subtle limitation to ChatGPT that some scholars are discovering: its output can tend to have a banal, overly general, uninformative quality even when the program is not inaccurate.  That’s what editors Barbara Treutlein and Patrick Cahan discovered when they used ChatGPT to help them write an editorial for the scholarly journal Stem Cell Reports, a publication of the science publishing giant Cell Press. The editorial was posted online on January 10th. The editors posed to ChatGPT various basic questions about the field of computational systems biology. The point was to demonstrate “how advances in computation … can help practitioners across the broader stem cell research enterprise: by saving time and thus freeing us to do more research!” The authors reprinted, in edited form, their prompts to ChatGPT, and the program’s responses. Here’s one of the question and answer examples: Treutlein and Cahan observe that ChatGPT was “helpful in articulating how computational systems biology will aid in stem cell research and in identifying some obstacles,” but they also write that “overall, many of the responses lacked depth and insight.” “ChatGPT:> The need for large amounts of data: To accurately model stem cell behavior, researchers need large amounts of data about the behavior of individual cells and the interactions between different cell types. This can be challenging to collect and process, especially when dealing with large numbers of cells.” Also: Can AI detectors save us from ChatGPT? I tried 3 online tools to find out To dig deeper into the matter, ZDNET emailed the editors to ask for an explanation of how to distinguish meaningful answers.  Cahan replied in email with an annotated collection of prompts and responses from ChatGPT that, he wrote, demonstrated “a glaring lack of depth and insight” by the program.  Despite the shortcomings, Cahan told ZDNET that “In general, I am very impressed with this tool’s power.” Cahan teaches a computational stem cell biology course, and said he has tested the ability of ChatGPT to write code to answer some of his problem sets.  “It does well (not perfect), and it documents the code!” wrote Cahan. “I also think that it has an impressive ability to synthesize and summarize information, and so I am eager to see what will come from its customization for biomedical literature.” Also: People are already trying to get ChatGPT to write malware Following are the questions and answers that Cahan and Treutlein left out of their editorial because of the glaring insufficiency. Cahan annotated each with his own observations as to why ChatGPT’s responses are lacking. Response: There are several major challenges in using computational techniques to answer questions in the field of stem cell biology. Some of the key challenges include:

  • The complexity of stem cell behavior: Stem cells are highly versatile cells that can differentiate into a wide range of cell types. Understanding the mechanisms underlying their behavior and differentiation is extremely complex and can be difficult to study using computational methods. Patrick Cahan: This is actually precisely an area where computational models should shine because they can find patterns in complex data. [ChatGPT response continues]
  • The need for large amounts of data: To accurately model stem cell behavior, researchers need large amounts of data about the behavior of individual cells and the interactions between different cell types. This can be challenging to collect and process, especially when dealing with large numbers of cells. Patrick Cahan: Again, the large data size is an area where computational systems biology should outperform standard techniques. ChatGPT seems to be conflating data collection with data processing here. [ChatGPT response continues]
  • The limitations of current computational models: Despite advances in computational biology and systems biology, current models are still limited in their ability to accurately capture the complexity of stem cell behavior. This can make it difficult to generate reliable predictions and insights from computational studies. Patrick Cahan: This is an example of lack of depth. If we had prompted for specific examples, it might have been possible to elicit ChatGPT  1.) to give examples of specific biological phenomenon that elude current computational models and 2.) to speculate why this is the case. [ChatGPT response continues]
  • The difficulty of translating computational results into practical applications: Even if computational models accurately capture the behavior of stem cells, it can be difficult to translate these results into practical applications, such as the development of new therapies or treatments. Patrick Cahan: This is true, but it is not a challenge that is unique to computational applications. So, like the second point, ChatGPT seems to be slightly misunderstanding the question. Perhaps with further clarification or prompt engineering on our part, we could have elicited a more satisfying response. [ChatGPT response continues] Overall, these challenges highlight the need for continued research and development in the field of computational biology and systems biology to improve our ability to study and understand stem cell behavior. Patrick Cahan: We see lots of this type of generic summarization that adds no new information to the overall ChatGPT output.