Once again, a blog post about ChatGPT. And while this blog mostly talks about the benefits that this technique brings, in my opinion, it also needs to be looked at critically.
What I’ve noticed recently is that a lot of students use ChatGPT to get answers to questions related to programming an application or website because they don’t have enough knowledge about it themselves. That sounds great; you ask a model to develop something for you and you barely need to know the programming language yourself. But what are the implications if this is done on a large scale?
I’ve noticed ChatGPT gives identical answers to programming questions. In addition, I see the same snippets of code popping up in multiple places. Often not even because they are written so concisely, but because they come from the same source (a language model such as ChatGPT). Because students are less inclined to think of and write their code, all the elaborations look more and more alike. I think this shows that this progression – which mainly provides convenience to the user – allows developments and innovations to diminish in the long run because the same structures and ways are mostly used.
Think of it like this: If I ask an AI model, which can generate something graphic instead of text, to design something in the style of a certain artist, it will indeed generate something non-existent, but in a pre-existing style. With this, I want to show that the current form of artificial intelligence helps us in a way that always uses pre-existing artefacts to arrive at something and does not arrive at something completely new. With that, AI is strong at reusing a particular technique but not at coming up with something completely new. Will that ultimately cause progress to stagnate?
What do you think?
(image https://bair.berkeley.edu/blog/2022/05/03/human-in-the-loop/)
Very interesting blog and great questions. I would say that it is good that students are using ChatGPT for coding. You might say that they get lazy and that ChatGPT is doing all the work, or you could say that ChatGPT is doing the initial easy part that, with a little time, everyone can do but does not want to do. Maybe I should phrase it a little bit differently: why should the students reinvent the wheel if it has already been invented?
On the second interesting aspect you raised about generative AI not being able to create something new, I completely agree that it is not yet possible to create something new. However, it is possible to create something that has never been created before. For example, imagine that a picture of a horse on the moon has never been created or made by an artist before. Generative AI is able to create that picture because it knows what a horse is and what the moon is. But the question is, “Is that something new?” or “Is anything not already done in some way?”