Notes on the use of generative AI
You should make certain you have read, and understood, the university’s policy on the use of generative AI in coursework.
You may be tempted to use generative AI in writing your thesis, or in helping to generate ideas, polish drafts, perform data analysis tasks, etc. etc.
The purpose of this section is not to terrify you about the dangers of AI, or convince you that it is inherently evil (it is not), but to make sure that you are aware of the inherent limitations in AI relevant for the tasks you are doing in your project, and the potential dangers of AI use (and misuse).
Because generative AI draws on large language models, which essentially rely on statistical relationships between words in their training set, it has inherent limitations and biases which often result in poor performance for academic work.
For example, a query about the famous, classic “PaJaMo” paper1 (which our 3rd year students study in Fundamental Microbiology), produces the following AI overview2:
AI Overview
The PaJaMo paper, published by Arthur Pardee, François Jacob, and Jacques Monod, demonstrated that the lac operon is regulated by a repressor protein. This “repressor” binds to a specific region of the DNA (the operator) and prevents the expression of the lac genes, effectively turning them off. The paper also showed that the inducer (lactose) can bind to the repressor, causing it to release from the operator and allowing gene expression to proceed.
Here’s a more detailed explanation:
The lac operon: The lac operon is a group of genes in bacteria that are responsible for the metabolism of lactose.
Negative regulation: The PaJaMo experiment showed that the lac operon is negatively regulated, meaning that a repressor protein is responsible for turning the genes off.
The repressor: The lacI gene encodes a repressor protein that binds to the operator region of the lac operon.
The inducer (lactose): The inducer, lactose, binds to the repressor protein and causes it to release from the operator, allowing the lac genes to be expressed.
Diploid analysis: The PaJaMo experiment used diploid analysis, where cells with two copies of the chromosome were studied. This allowed them to determine the dominance and recessiveness of mutations in the lacI gene, leading to the discovery of the repressor.
At first glance, this might appear to be a plausible summary - at least, it contains content about the lac operon, and appears to be largely in agreement with modern knowledge of how the Lac repressor works.
However, closer reading will show that this text is actually blatantly and flagrantly incorrect.
Reread the passage: how many errors can you find? Then, click on the callouts below to expand them and see the errors that we’ve spotted.
AI Overview
The PaJaMo paper, published by Arthur Pardee, François Jacob, and Jacques Monod, demonstrated that the lac operon is regulated by a repressor protein. This “repressor” binds to a specific region of the DNA (the operator) and prevents the expression of the lac genes, effectively turning them off. The paper also showed that the inducer (lactose) can bind to the repressor, causing it to release from the operator and allowing gene expression to proceed.
The summary is completely and utterly wrong. Pardee, Jacob, and Monod, did not demonstrate that the lac operon was regulated by a repressor protein - in fact, in their famous paper, they explicitly ask “What is the chemical nature of the repressor?”. It was not shown to be a protein until much later3
Likewise, this paper did not demonstrate that the inducer binds to the repressor (how could it, when the nature of the repressor was unknown?)
Here’s a more detailed explanation:
The lac operon: The lac operon is a group of genes in bacteria that are responsible for the metabolism of lactose.
Here, the summary claims that the lac operon is “in bacteria” generally (whereas we now understand it to be specifically found only in certain species of bacteria, such as Escherichia coli). It is also quite vague/unclear - it doesn’t really explain what an operon is, or why the structure of the operon is important in this context.
Negative regulation: The PaJaMo experiment showed that the lac operon is negatively regulated, meaning that a repressor protein is responsible for turning the genes off.
Partial credit here, perhaps - the paper does propose a repressor model for lac regulation, but again, it doesn’t establish that the repressor is a protein.
The repressor: The lacI gene encodes a repressor protein that binds to the operator region of the lac operon.
The inducer (lactose): The inducer, lactose, binds to the repressor protein and causes it to release from the operator, allowing the lac genes to be expressed.
This is a superficial but not completely wrong summary of our modern knowledge of the how the lac operon is regulated in Escherichia coli, so partial credit here, perhaps; however, it’s really rather vague, and doesn’t explicitly make it clear that this is modern knowledge and not the PaJaMo paper.
It might be appropriate to put the PajaMo paper in context of our modern understanding of the lac operon, but it should be clearer, here, what is modern knowledge and what was done in the actual paper itself. And if the object is to describe the system according to our current knowledge of how lac regulation works in E. coli - a clearer explanation of what the operator region actually is would help, as well as more molecular details.
Note that a lot of the details are wrong or missing - the AI here isn’t even correctly formatting gene names (they should be italicized)!
Diploid analysis: The PaJaMo experiment used diploid analysis, where cells with two copies of the chromosome were studied. This allowed them to determine the dominance and recessiveness of mutations in the lacI gene, leading to the discovery of the repressor.
Not even partial credit here - this is just wrong. The paper did not use diploid analysis - it used merodiploid analysis. It doesn’t even require particularly close reading of the paper to spot this error - “heteromerozygotes” is in the abstract, even.
So what do you think - how well did AI do here? If you were going to mark it according to the University Marking Guide, what mark do you think it would deserve?
If I were marking the PaJaMo paper summary above, according to the University Marking Scheme, I would mark it as a 20:
Clear fail: Weak demonstration of the learning outcomes The work is very weak or shows a decided lack of effort. The work demonstrates - Very poor or confused knowledge and understanding, with reference to only a few key words, phrases or key ideas - No argument or one based on irrelevant and erroneous content - Irrelevant content and extensive omissions - Weaknesses of presentation and/or logic and/or evidence - Inadequate evidence of learning - Incomplete or inadequately presented references, if any]
If you do use AI tools during the course of your thesis project, you must ensure that you understand their limitations. You do not want to submit a thesis that contains blatantly incorrect information - firstly, because this is bad scientific/academic practice, but secondly, because your marks in the end will be consistent with this poor practice.
Footnotes
Pardee AB, Jacob F, Monod J. 1959. The genetic control and cytoplasmic expression of “inducibility” in the synthesis of β-galactosidase by E. coli. J Mol Biol 1:165–178.↩︎
on a Google query, “what did the PaJaMo paper show?”, 24 April 2025↩︎
Gilbert W, Müller-Hill B. Isolation of the lac repressor. Proc Natl Acad Sci U S A. 1966;56(6):1891-1898. doi:10.1073/pnas.56.6.1891↩︎