Generative AI Brain Dump

 I've been trying to clarify my thinking around how I feel about GenAI (generative AI, specifically tools like ChatGPT in my case) and teaching and learning in higher ed (and in general), but haven't made much progress. So this won't be so much as a blog post as a brain dump of various things sloshing around in my head. 

 Do I think using GenAI is just like using a calculator? No. Calculators take away the drudgery of doing calculations yourself. You need to provide the raw materials as input (i.e., the numbers that you want to do the calculation on), which means that you have to have a reason for inputting those numbers. In other words, you are in control of the thought process and are just pulling in additional help for some of the grunt work. For the simpler calculations, you could have, in theory, worked out the answer yourself.  

It is still important for people to understand what the basic mathematical operations are even if they can use a calculator. Part of understanding the basics is to try to do it yourself, so it makes sense to still have basic math instruction in schools. 

Do I think using GenAI is just like using a stats program (SPSS, etc.) to analyze data? Not really, but this is a little more of a gray area than the first one. I don't think I could compute PCA by hand (and I certainly wouldn't want to). Some would also argue that there is a degree of "mindlessness" that creeps in when students learn which buttons to push and which numbers to report without really understanding what is going on "under the hood." 

This leads to the question of how much understanding is needed to use a tool, and the relationship between how the tool works and what the tool does. An automobile is a tool to get you from point A to point B. You have to know how to use the tool (i.e., how to drive) to make use of the tool, but how much do you need to know about the internal combustion engine (or the electric engine)? You need to know about filling the tank (charging the battery), but for anything over that, there are professionals who can help fix the tool if something goes wrong. And, of course, the mechanics who work on cars don't necessarily need to know how to design a car per se, though they will probably have some better intuitions than a layperson. 

Do I think that using GenAI is cheating? Well, this is a loaded question. Yes and no. I think this partially depends on what the expectation is. Riding the subway for a few stops in a city is depriving you of the exercise that you would have gotten had you walked, but it is not cheating unless you were a participant in that city's marathon where the expectation is that you are completing the course unaided. 

So part of the answer depends on what the impact of the discovery of GenAI use in a particular situation would be and whether or not you vouch for the product that you are putting forward. Deloitte was famously embarrassed by providing an GenAI written report to the Australian government. Had the report not been riddled with hallucinations, there would likely still have been the expectation that you were paying for the service of seasoned human researchers and not merely prompt engineers. Will this change when GenAI becomes even more common? 

I recall reading somewhere that GenAI use in STEM for writing up results is looked on much more favorably than in the humanities because the discovery of a new algorithm or methodology is the goal, not the research paper itself. Time "wasted" on writing up what you've already done could be better spent on making new discoveries. 

That sentiment is not shared widely in the humanities, from what I can tell. 

Do I think GenAI is a fad? Yes and no. The hype around AI will wane over time, but the technology is here to stay. Blockchain and crypto are still around even though the hype has died down from those early days about how they "will change everything." 

Hype or not, GenAI certainly does seem to be quite disruptive, and that is a bigger problem. Unlike the Covid19 disruption, the effects of which are still being felt, GenAI is ongoing. There is no telling how problematic this could become given some of the negative trajectories. It is true that there have been disruptive technologies before, but it seems that the rate of adoption was slower. The "brain rot" that started with social media could be hypercharged with GenAI. 

Is GenAI like Covid19? Perhaps "yes." What I mean by this is not the impact per se, but the fact that it doesn't matter whether I am a "fan" of GenAI or a critic, it is here doing what it does irrespective of my wishes. And, like Covid19, we are not prepared at the societal level for the disruption that it has and will cause.  

But there were some upsides to the pandemic (at least for some portions of the population). Although it has been reversing recently, the focus on work/life balance became clearer, and the ability to work from home at least occasionally became a reality for many. We realized that maybe we didn't need all of the meetings that we had, and that we could also work asynchronously and still collaborate. 

This was more of a stream of consciousness brain dump than a blog post. I don't think that I am any clearer in my thinking. So much is context dependent that it is hard to generalize. 

I don't like (and don't use) Gen AI for writing.  The only exception is when I need to create some kind of "boilerplate" language in which case I might ask ChatGPT for a suggestion. That usually gives me enough of an idea to get over my temporary block. I don't think I have ever copy/pasted something from ChatGPT directly into another document (unless I was intentionally creating a "this is what ChatGPT said" kind of document). 

I am much more likely to use ChatGPT for coding, but even that depends. Many times I will have it generate code, then ask it about the security implications of using that code (if there are any). If I don't feel comfortable that I know what the code is doing, I'll hold off going any further.