Hi everyone! Today’s blog post explored a slightly controversial topic- incorporating artificial intelligence (AI) into academia. In recent years, there has been quite a buzz around various AI interfaces, one, in particular, has been making the news very frequently. Some of you might have heard its name- ChatGPT. For this week’s lab meeting, we are all expressing our thoughts on the OpenAi platform- ChatGPT.
Honestly, I did not have a strong opinion on an AI’s involvement in academia before I learned of ChatGPT’s existence and even now, after doing some research on how this interface works, I am not fully sure what to think of it. My first instinct was to reject it, thinking of various ethical implications of it, as well as what this means for future job opportunities for a young-career person like me. However, after giving it a bit more thought I realize, perhaps I need to treat this as any other new technology being introduced to the market. Instead of my initial hostility (maybe that’s a strong word to use but I couldn’t think of an alternative) towards it, I should learn more about how this works and maybe there’s a productive way to incorporate it into my research. Maybe this is how our older generation felt about being introduced to the internet, or more recently to platforms like google and social media.
Anyways, with that goal in mind, I tried asking ChatGPT a bunch of random tasks (related to my research) and the results are… interesting to say the least! Out of multiple possibilities for its use in research/academia, I tried focusing on (1) writing a research proposal and (2) coming up with a new methodology – both of which are tasks I’m currently working on. Without further ado, let’s just get right into it. First up we have :
ChatGPT’s input on radar-dark halo craters on the Moon (because I had to haha)
I asked the AI to write a research proposal on RDHCs on the Moon and it gave me a detailed description which includes an introduction, objectives, proposed methodology, expected outcomes and conclusion. Now, that sounds like a legit research proposal format. So my initial reaction was definitely to be impressed (and slightly creeped out haha). However, after reading the actual content, I’m less impressed. Here’s a screenshot for example.
My first impression is that the wording sounds too redundant- almost as if some first-year student overused the thesaurus. But overall, the writing still feels like a legit research proposal (at least it would to an unseasoned person). However, the main issue is with its scientific accuracy. Now, I understand that my sample size is too small to come to that conclusion, and maybe others had more success getting an accurate description, but it’s true for this one case. I’ve learned a lot about these RDHCs over these past two years and there haven’t been many references to their volatile (e.g. water ice) content and its relation to the halo’s appearance. So, I’m confused, where is that coming from!? (Maybe there’s a paper out there I overlooked during my initial lit review.) My point is, to someone who is not too familiar with this research topic, ChatGPT gave a very ‘real-looking’ research proposal, but, in reality, it’s full of inaccuracies and repetitive sentences.
However, it’s not all bad. There are a couple points in this ‘research proposal’ which if nothing else, gave me something to think about. For example, possibilities for numerical modelling/simulating impact events to understand the halo’s mechanism. Now, currently, I don’t have enough knowledge (and more importantly, time) to actually implement these suggestions into my research but it’s an idea at least, maybe someone else in the future can explore that.
My verdict: ChatGPT can’t be the tool that writes your research proposal for you (at least not yet, who knows maybe in 5 years?), but one can definitely use it to get a quick idea about knowledge gaps and/or future implications of a research topic.
Matlab code for QUARTILE MAPPING & alternatively generating the map in ArcMap
Next, I asked the AI to come up with a MATLAB code for generating quartile maps of CPR images of the Moon. (sidenote: ChatGPT knew instantly what CPR I was referring to). It created a code for me alright, but due to my lack of coding knowledge, I’m unable to decide how accurate this is. My plan is to do a test run this weekend and see if it works. So far my experience tells me it’s very likely that the code is inaccurate but it sure has some steps that look very similar to an actual code for making quartile maps. I won’t lie, a part of me is rooting for ChatGPT for this one. Because I believe this would be an actually productive use of AI in research/academia. Even if the code isn’t perfect, if it could give a newbie like myself a sense of direction in the otherwise overwhelming coding world, I’d count that as a win.
I also asked it to tell me how I can make a quartile map in ArcMap and this is where things got weird. In my first attempt, I didn’t specify a software version and so some of the finishing steps didn’t match the features my ArcMap has (see pictures below). But then, I specified the version and asked ChatGPT two times to give me the steps. Interestingly, it gave me two different sets of steps/methods each time (I find it quite funny). One of the versions is more accurate than the other (i.e. similar to my current method). I guess it’s nice to have part of my current methodology be validated by an AI haha!
My verdict: Depending on how you word your request, you get a different answer each time. Meaning, this is probably not the most reliable way for trying to come up with a new methodology. Again, similar to my previous point, I don’t think this AI can be used as a sole resource/tool for any research-related task, at least not in its current state. But who knows, with time and enough feedback from its audience, this tool can actually learn to do some of these tasks more accurately.
So, my overall opinion is that it’s worth exploring more productive (and ethical) ways to make use of this technology. Literature review for example. In the coming days, I’m going to spend some more time with this tool, and see if it can actually be used in academia as a grad student.
Until next time!