Unlock the Power of AI in Your Research Using ChatGPT as a Partner in Scientific Discovery

ChatGPT is a new chatbot that uses AI to allow you to kind of chat with it, interact with it, and ask it questions. And so recently, someone asked for me to write an article on how to use ChatGPT in scientific research. So that's what I'm going to be covering today. The different ways to use it. It's a little bit of a tutorial and a little bit of inspiration. Then, at the end, I'll discuss whether or not using it in these ways is ethical.

This is the website here, and to get access to it, all you have to do is scroll down and click Try ChatGPT. And if you don't have an open AI account, you are going to have to register and have an open AI account to be able to access it, but it is free to use right now in the research preview. You can use it as many times as you want. So it tells you a little bit about it here, but I'm specifically going to walk through a few examples of how to use it.

So the very first example I want to talk about is: what if you're trying to learn a new field? So for this example, I'm actually going to go on a very different topic than I did during my PhD, but I'm going to talk about obesity, reproduction, and lipid metabolism. So, if I just wanted to know what obesity is and how it affects reproduction, I'd ask. That's exactly what I'm going to ask. And so what it's going to do is process what I said and try to give me good information back on that topic.

And it was trained by data up until 2021. So you should be aware that this may not be the most accurate or up-to-date information available.Like if you're trying to figure out if something's novel or not, this might not be able to tell you because it could have been done in the last year, and then you would need to still go out and look for it. So that's one caveat to using it.

Okay, so it gave me kind of a small paragraph here. It says obesity is a medical condition in which excess body fat is accumulated.

And then, in terms of reproduction, obesity can affect both men and women. It reduces sperm count and quality, as well as X drive. And in women, obesity can cause irregular cycles and make it more difficult to become pregnant during pregnancy. There's a risk of complications as well. So this was a study that I did with animal science when I was in grad school, and so this would be really important if I didn't even know how it could affect it to go ahead and have that background information.

Another thing you can do is ask very specific questions. So for example, if I didn't know what preeclampsia was, I could just ask, "What is preeclampsia?" Okay, so it gave me a definition for what preeclampsia is. It's characterised by high blood pressure and protein in the urine, and it gives me some symptoms and all of this. So it's really helpful to really quickly get that information versus going and looking for literature reviews or Wikipedia or stuff like that to get really succinct information. That's similar to when you're first starting out in your field and learning what different words mean. So what if we take it a step further and say we're trying to come up with research topics and research questions to ask and we're having trouble?

So can we ask Chat or GPT for research questions? So what if I said, "What can you suggest?" Obesity and reproductive health-related research questions So it actually suggests, like, seven different research questions here. And they aren't all that bad.

They are a little broad, but I think it's really interesting in, like, this one specifically, like, how does obesity affect the outcomes of assisted reproductive techniques? I wouldn't have even thought of that one. And I don't know if there are papers out there on it or not. Now one thing is that because this was trained on data up until 2021, it can't tell if these are actually novel.

I can't tell if there are already research papers on it. Are they actually good research ideas to pursue or anything like that? And so if you want to use one of these and you're wondering, "Should I actually pursue this?" I would suggest Donna install her 30-day research jumpstart guide.

It outlines the steps you must take. What research questions are actually worth pursuing? But if you're really struggling with coming up with research ideas or research questions to ask, I think this is actually a really good method to kind of start thinking about it. Because sometimes once you start thinking about it, like, "What is the relationship in obesity between obesity, infertility, and women and men?" You might think, "Okay, what's that relationship?" And then you might start thinking about specific things as you learn about the field.

Like, oh, I wonder what the relationship is between obesity and hormone levels, or something like that. As a result, it can be a great place to start allowing yourself to get there.to your actual research question. Any of these are far too broad for a single study like this to be considered an overall; this is what our lab studies, but I believe this is a good place to start.

So, if we're doing this, we can chat on GT about it and turn it into a hypothesis. So I'm going to take this one because I just think this is a very interesting question to ask. So I'm just asking it: Can it write a hypothesis for this research question? because I know people. Ask me a lot. This is my research question.

Can you give me a hypothesis? And, like, I'm not the person to be giving you hypotheses in your field, but maybe you could potentially use something like Chat GPT for that. So it's giving us a hypothesis.

Obesity is associated with a decreased success rate for assisted reproduction techniques such as IVF due to its negative effects on reproductive health. So that would be the hypothesis, and then it actually goes one step further. So it tells us how we could test the hypothesis. This hypothesis could be tested by conducting a study that compares the outcomes of IVF in obese and non-obese individuals and examining factors such as pregnancy rate, live birth rate, and complications.

So, if you're having trouble turning your research question into a hypothesis or an approach, working through some of this with Chat GPT can either help inform the study you're trying to create or get you in the mindset of being able to think. because there are a lot of us. We do better when we think through examples.

So if you can create those examples using CPT for your specific field, then you can start thinking better, which might make you better at doing this research. So, what if we had another one, and we had a research question that we didn't know how to answer? Collecting data and, like analysing it, testing our hypotheses about it So let's say I wanted to test the hypothesis that lipid concentrations in obese mothers affect placental function. So I'm just asking how I would test this hypothesis. Okay, so it's given me a kind of approach here.

So it's basically saying that I should conduct a study to investigate the possible relationship between these two variables. done through a variety of different methods. So one is recruiting a sample of actually pregnant women, which is a lot of IRB work. Varying the levels with varying levels of obesity is impossible because you can't cause obesity in humans and measure the lipid concentrations at different points. and then use imaging techniques to assess placental function. So that would be a great human version of the study. And then it says, "Another approach would be to conduct experiments on animal models where you manipulate the lipid concentrations in obese mothers and assess the impact of placental function."

This can provide valuable insights. and help identify potential targets. So we actually did this study. This is what we did. We took mice, fed them specific diets, created obese mice, and then allowed them to mate.

and analyse their placental function. And so it's important to control for other factors such as age, diet, and other medical conditions. And this is exactly what we did in our study. We made sure the mice were the same age; we gave them the same diet, and they were controlled so that if they showed any other medical conditions, they were removed from the study.

So this is actually a really good initial approach. It's not giving you every single step; it's not giving you every test you need to perform, but if you're struggling and just trying to get there, this is actually a good way to start thinking about how you can get there. So now let's assume that we did the study and have a bunch of data.

How can we use chat GPT to actually help us analyse that data? So there are a few different things that you can do. The first way it can help you is actually to help you write code. So if you are someone who, like you, mainly uses Excel and you maybe wanna start working in R or something and you're trying to analyse something, how can it help you write code for that? Okay, so I'm asking it to write our code to create a grouped bar chart with the mean of the weights on the Y axis. X is on the X axis, followed by the various bars, the different groups being treatments, and then the standard error of the mean as error bars.

So it's going to say, "Here's an example, and it's going to say Library GiGi Plot Two." So that is the library I would use as well. It's reading in the data here as a csv, creating the mean weight with the aggregate weight by X and treatment, and setting it all here. The standard error is then calculated using the same type of function as before. The grouped bar chart is then created here. So the mean weight is

This is our aesthetic, as well as our X- and Y-access fill. It's adding a GM column, which is going to be your bars, and then it's adding an error bar of the weight minus the SM weight, and it's also dodging it. So this is really important to actually make sure your error bars line up whenever you're doing a grouped bar, and then it kind of tells you what's going on here.

So it's telling you what it's using to create this, and it says you may need to adjust this to reflect your data.

So then you can literally just copy this code, run it, put it in R, add it in, change whatever names you need to change, and run it. and it would actually generate your bar chart as you want it. So that can be really helpful. If you're just learning R and you want to do certain things to actually have chat, go ahead and create your code now for the long term. You can actually learn R by having it create your code and then messing with it.

Long term, this will not replace people who programme in R because I could have written this code in the same amount of time because you just get faster at it the more you write it.As a result, that can be extremely beneficial in that regard. Let's do another thing, though, where we're going to ask it to find the means and we're going to ask it to run a statistical test.

So I'll ask: can you compute the means of the weights of two treatment groups and tell us if they differ significantly in our case? So this is the example it gave; it told me what it was and what I would need to do, and then it gave me the code here. I don't know why it's telling me to do GG Plot 2, because we're not using it in this code. So that's a little odd.

It's loading the data for calculating the mean weights, and then it is. I'm also running a t-test here, which is exactly what I'd do if I were attempting to do this. It can actually debug your code as well.

So here's an example of code I wrote. I know what's wrong with it, and I'm telling you. I wrote it, but it isn't working. Can it tell me what's It basically told me that this one is "my filter," "my treatment," and "my treatment is not correct," so I don't need to use the equal sign. I need to be different way.This is the more efficient way to accomplish this using the code provided.

The other way is that I need to actually have two equal signs. So this is a single equal sign, and R is an assignment. A two-equal sign is a target. So that would be the other way to solve that code. And then down here, it's telling me that there's an issue with my selection because I need to. DF, and then select.

This is the same thing that I wrote up here. So there's not actually an issue with that, but it's being pulled in as an issue, which is interesting. So that is all coding; basically, we're going to help you write code and help you debug code.

The other thing it can also help you do is determine what kind of statistical test you're using. So I've created a table in which I have three treatment groups and 25 lipid concentrations for each sample in a treatment group.

How should I compare these to see if it's statistically different or higher in one group than  theother? She'sShe's pulling it in as an issue, which is interesting. So that is allcoding; basicallyodwe'recgoing to we're going to help you write code and help you debug code. The other thing it can also help you do is determine what kind of statistical test you're using. So I've created ainnwhichre in which I have three treatment groups and 25 lipid concentrations for each sample in a treatment group. How should I compare these to see if it's statistically different or higher in one group than the other?the other?'s pulling it in as an issue, which is interesting. So that is all for coding, basically gonna help you write code and help you debug code. The other thing it can also help you do is determine what kind of statistical test you're using. So I've created a scenario here that is just I have three treatment groups and 25 lipid concentrations for each sample in a treatment group. How should I compare these to see if it's statistically different or higher in one group than. So the answer it gave me is that I basically need to do a one-way analysis of variants or an Inova, which is what we did when we had this data. And then it tells me how to do it with my arm.

And it's telling me R because it already knows the answer to my question.So it knows that I'm asking for code "R." So it's going to give me the code "R" when it's trying to give me an example of that code. And then it's telling me what it did down here, which is basically what's going on. So those are some different ways you can use it for data analysis. So we've completed all of our data analysis. Now it's time to write our research paper.

How can we start using it to write our research paper? So the first way I would suggest using it is to help you get started, because if you're just staring at a blank page, how can it help you get started? I'm writing a research article, and I'm telling it what I'm writing it about.

And then I ask, "Can you write an introduction, an introduction paragraph for me?" So this was a pretty good introduction paragraph. The first sentence is all about significance, and if you're wondering how to write an entire research article, always start with "significance" as your first sentence. Then it gains importance in terms of how it affects people, and we move on to the implications of reproductive health and pregnancy.It actually flows very well. And then I wouldn't put this in my first introduction. I'd move that after I go through the background information, the information, and all of that. However, this provides you with a starting point.

Would you publish this? Maybe not, but at least it starts giving you something to edit and something to think about when you're trying to create this. And as we're going through this, I'll also leave my scientific research paper checklist in the description below as well. So then the next thing you can do is ask it to enhance you. So, for example, I have this abstract that was published, and I'm going to ask for it here.

Can you rewrite this abstract to make it easier for people to understand? Now, this is not a scientist, right? Like I would, I could use this as a structure, and then I would probably go back in and add these more specific things into this specific paragraph here. You can always hit "try again" if you don't like it, and it'll generate a new version.

So this time it gave me a new version of it that I could use instead. So it's a good way to generate ideas for how to do things. The final thing I'm going to ask it to do is suggest some titles. I know titling a research article can be really, really difficult, and I'm not even going to pretend that I'm really good at it. So I'm going to say, "Can you suggest some titles for this?" Abstract, so we can remember the previous abstract that I gave it, and now I can ask it to suggest some titles to get me started thinking about it. So they are; it generated a few different options that I could use or at least start thinking about to get started on it. So the final question is: Is any of this ethical? Like, is this the same as just hiring someone to write your thesis? and I would say no.

Honestly, all of the things that I'm asking it to do, I could easily ask a lab mate the same types of questions, and it would be considered completely ethical to do that. So, like, I could ask a lab mate, "What's a good research question?" Or do you think this is a good research question? Can you help me workshop something? All of these things are either enhancing what you've already done or giving you a starting point where you still need to do more research to complete the project. So in my view, the ways that I have mentioned today—none of them are unethical to do when you're in grad school or a researcher in general.

As technology gets better, it's good for us to allow it to help us become more efficient in our research, and it actually can help us develop better because it's giving us examples that other people would have to take time and effort to give us. Now, one use of Chat GP that I haven't mentioned is summarising research articles and the like.

And honestly, there's a reason, and that's because it won't do it. It actually will not do it because it's not hooked up in the right ways. And I've tried it before, and it's just like, "It won't do it unless you actually give it to it." So if you want to summarise research articles, I really suggest checking out either Paper Digest, which I will leave over here, or size-based co-pilot.