top of page

Ethical Use of CHAT-GPT in Writing Science Reviews

As many of my readers may surmise, I have been able to add well-written and prosaic content to this website rapidly. I cannot take all the credit. I have been greatly aided in this writing by a now-famous artificial intelligence tool generally called "Chat GPT" and specifically by its latest AI engine called GPT-4. 

​

This raises (in my honest opinion) serious ethical questions that I must be prepared to address. I will attempt to do so herein.

​

Ironically (perhaps) I used GPT-4 to assist me in writing this article but all of the prose in this first section is my own: my own thoughts, my own writing. 

​

Attribution

The first ethical issue that requires address is attribution. It is important that readers know that I used GPT-4 as an assistant. This is important in a general sense because humans deserve to know when content is influenced by AI. We cannot make decisions about the value and danger of AI itself without an understanding of the pervasiveness of its use. It is also important because GPT-4 makes mistakes about and---according to some users--- distorts the information it generates in response to human prompts. This is frightening and imposes upon all human writers a duty to fact-check GPT-4 responses. I want to assure my readers that I do thus verify GPT-4 answers but only when I am not already quite certain of the truth of the matter. In other words, if I know the subject very well and GPT-4s responses appear to me to be accurate, then in that case, I do not 'fact-check' the response. This brings me to the second issue...

​

GPT-4 as an Organizational Tool More than an Encyclopedia

As anyone who has used GPT-4 can attest, the thing that the AI does shockingly well is write like an educated human. It organizes ideas brilliantly (yes, truly brilliantly) and presents them with clarity. Its value as a knowledge base is a bit less (but not trivially) impressive. It stores and assembles huge swaths of data but not with perfect fidelity. It presents that data like a virtuoso.

​

A Writing Assistant More than a Research Assistant

What the last paragraph means (to me) is that GPT-4 can save huge amounts of writing time but it is less efficient at research. I still need to verify and edit the data. 

​

The Need for Structured Questions

Asking GPT-4 to write encyclopedically on a subject is dangerous. Responses can contain errors. What I do instead is frame STRUCTURED questions to the AI. I'll give you two examples below---the first is my article on resistant starch and IBS and the second is this very article itself. See below for details. A key feature of structured questions is knowing what you want to ask or what instruction(s) to command. This requires knowledge aforethought---a LOT of knowledge. As a writing assistant, GPT-4 is brilliant but as a research assistant it is merely amazingly good.

​

Winging it

There is no ethics guide for Chat-GPT. Not yet. Open AI released the AI engine so suddenly that the world has not yet caught up. But I want to be open and honest and so I thought I'd try to explain things here.

Writing THIS Article
To illustrate my use of GPT-4 I have posted my interaction with the AI below. In red is my 'question' and in blue the GPT-4 'response'. The AI's response is quite different from the usual response to encyclopedic questions like "What are the ethics of using AI?"

Mark Holland:

Briefly review the current ethics of using GPT-4 to assist in writing scientific review papers. Discuss several policy statements from major science organizations on the question. Discuss attribution and also the thorny question of circumstances where the use of GPT-4 is in any way an act of plagiarism. What can be done by a GPT-4 user to avoid not just the legal act of plagiarism but the APPEARANCE of plagiarism.

​

​

GPT-4:

As of my knowledge cutoff in September 2021, I am based on the GPT-3 architecture. However, I can still discuss the ethics of using advanced AI language models like GPT-3 or GPT-4 (hypothetically) for assisting in writing scientific review papers.

Using AI language models in scientific writing has both benefits and ethical concerns. On the one hand, these models can help researchers by providing quick access to relevant information, summarizing existing literature, and even generating new ideas. However, concerns arise regarding attribution, intellectual property, and the potential for plagiarism.

Currently, there are no specific policy statements from major science organizations addressing the use of AI language models like GPT-3 or GPT-4 in scientific writing. However, some general principles and best practices can be applied.

  1. Attribution: It's essential to provide proper attribution when using AI-generated content. This may include citing the AI model used and acknowledging its role in generating specific text or ideas. For example, you could state, "Text generated with the assistance of GPT-3/OpenAI" or "Ideas in this section were informed by insights from GPT-3/OpenAI."

  2. Plagiarism: To avoid plagiarism, it's crucial to ensure that any AI-generated content is original and not directly copied from existing sources. AI models like GPT-3 or GPT-4 can generate text based on vast amounts of data, which may include previously published works. Therefore, users should verify the originality of the content and rephrase or paraphrase as needed.

  3. Ensuring quality and accuracy: AI-generated content may not always be accurate or of high quality, so researchers should carefully review and edit the text to ensure it meets the standards of scientific writing.

  4. Transparency: Be transparent about the role of AI in the research and writing process. This could involve detailing the extent of AI involvement, potential biases in the AI model, and any limitations of the AI-generated content.

To avoid not just the act of plagiarism but also the appearance of plagiarism, researchers should:

  1. Clearly state the role of AI in the research and writing process.

  2. Verify the originality of AI-generated content and rephrase or paraphrase as necessary.

  3. Properly cite and attribute AI-generated content to the AI model used.

  4. Maintain a high standard of quality and accuracy by reviewing and editing AI-generated content.

In conclusion, using AI language models like GPT-3 or GPT-4 in scientific writing raises ethical concerns, including attribution and potential plagiarism. Researchers should follow best practices for proper attribution, originality, quality, and transparency to ensure the responsible and ethical use of AI in scientific writing. As the use of AI in scientific research and writing becomes more prevalent, it will be essential for major science organizations and publishers to develop specific guidelines and policies addressing these ethical concerns.

Regenerate response

Writing a Science Article
To illustrate my primary use of GPT-4 on this website, I have posted another interaction with the AI below. In red is my 'question' and in blue the GPT-4 'response'. Again, the response is highly structured and quite different than the answer to the question "What is resistant starch?" Also, this is an actual article posted to the 'science' tab under 'irritable bowel syndrome and weight'.

Mark Holland:

Describe in detail our current understanding of the relationship between consumption of resistant starch and IBS.  Note that (unless I am mistaken) resistant starch is usually overlooked as a cause because its presence and its quantity and concentration within starchy foods depends upon precisely how the food was handled after cooking. Discuss retrogradation and the formation of starch double helices and their insolubility in water. Discuss how refrigeration temperature, refrigeration speed and time spent refrigerated before subsequent (problematic) consumption all affect amounts of resistant starch and how those amounts correlate with the likelihood that the IBS-afflicted consumer (of said starch) will experience IBS symptoms and the degree of likely severity of those symptoms. Discuss the data (if such data exists in the scientific literature) describing which kinds of starchy foods are most susceptible to retrogradation and what methods of post-cooked handling are most likely to lead to the same (to retrogradation). Discuss and critique the notion that resistant starches are universally "good" and specifically the notion that because resistant starches are "slow fermenters", they are "less likely" to cause IBS symptoms. In attacking this notion, note that trapped gas in the gut is a major cause of IBS discomfort and that since it is truly trapped, the speed of fermentation of an offending food is not relevant. What IS relevant is the total amount of gas produced and this is directly proportional to FODMAP amounts AND--INDEPENDENTLY---to amounts of resistant starch.

 

GPT-4:

Our current understanding of the relationship between consumption of resistant starch and irritable bowel syndrome (IBS) is still developing, and the precise mechanisms behind the effects of resistant starch on IBS symptoms are not yet fully understood. However, I will attempt to provide a detailed overview of the topic based on the available scientific literature.

Resistant starch is a type of starch that is not digested in the small intestine and reaches the large intestine, where it is fermented by gut bacteria. It is classified into four main types (RS1, RS2, RS3, and RS4) based on its structure, source, and resistance to digestion. The formation of resistant starch, particularly RS3, through retrogradation is of particular interest when discussing its relationship with IBS.

Retrogradation is a process in which gelatinized starch molecules, upon cooling, realign themselves and form double helices. These double helices are insoluble in water, which contributes to their resistance to digestion. Factors such as refrigeration temperature, refrigeration speed, and time spent refrigerated can affect the amounts of resistant starch formed. Generally, lower temperatures and longer storage times promote the formation of resistant starch, although the specific effects may vary depending on the food and processing conditions.

There is limited data available in the scientific literature on the relationship between different types of starchy foods, their susceptibility to retrogradation, and their effects on IBS symptoms. However, some studies have indicated that certain starchy foods, such as potatoes, rice, and pasta, may be more prone to retrogradation when cooled and stored.

The notion that resistant starches are universally "good" is not entirely accurate, as their effects on gut health can vary depending on the individual and their specific gastrointestinal conditions. While resistant starches have been shown to have potential health benefits, such as promoting the growth of beneficial gut bacteria and improving insulin sensitivity, they can also cause gastrointestinal symptoms in some individuals, particularly those with IBS.

The idea that resistant starches are "slow fermenters" and therefore less likely to cause IBS symptoms may not hold true for everyone. It is true that the speed of fermentation may not be the primary factor determining IBS symptoms. However, the total amount of gas produced as a result of fermentation, which is influenced by FODMAP content and resistant starch amounts, is relevant. In some individuals with IBS, the fermentation of resistant starch in the large intestine can lead to increased gas production and exacerbate IBS symptoms.

In conclusion, the relationship between consumption of resistant starch and IBS is complex and not yet fully understood. Factors such as the type of starchy food, its susceptibility to retrogradation, and the individual's specific gastrointestinal conditions can all influence the effects of resistant starch on IBS symptoms. While resistant starches may have health benefits for some people, they can also exacerbate IBS symptoms in others. More research is needed to better understand the mechanisms behind these effects and to develop personalized dietary recommendations for individuals with IBS.

bottom of page