AI is overrated
Right now, AIs (specifically LLMs like ChatGPT) are a pretty big topic of discussion. Wherever you go, chances are, you‘re going to encounter the subject. So, naturally, I’ve developed my own opinions about these sorts of AIs that I’m going to share with you.
Right now, statements like “AI is going revolutionize [Insert Thing Here]” are all too common.
To be honest, I feel like the hype is pretty overblown, and there’s some ethical questions about AI most people weirdly just don’t seem to care about. For example, most big AIs, like ChatGPT, are trained using publicly available material from the internet. This is done using a program made by the developers known as a Webscraper, that just goes to a bunch of websites and just… downloads all the text/images/whatever the AI is supposed to be trained on.
Thing is: The people running those sites have absolutely no idea this is happening most of the time, and never gave consent for any of these companies to take their data to train their AI on.
This is not only questionable at best ethically, it also raises some questions about copyright. Are you infringing on an artist’s copyright by using their images to train your AI? We don’t know yet, and if the answer to questions like these turns out to be “Yes”, the whole field is basically dead. You simply can‘t collect the amount of data these neural networks need without basically scraping the entire internet for usable stuff.
Another pretty big ethical concern is dataset labeling: Once you have all this data to train your AI on, you need to categorize it, so the AI knows what it is. For example, if you want your AI to recognize pictures of dogs, you can‘t just give it a bunch of pictures of dogs and cats without telling it which is which. In order for the training to work, you need the AI to be able to check if the prediction it made was right, so it can learn. As you can imagine, labeling as much data as these LLMs need is very boring and time-consuming. You might think it would cost these companies a lot of money, which it would if they were doing it ethically. Instead, they are outsourcing this task to third world countries for literal cents per hour. This isn’t some grand conspiracy either. Just go to https://www.mturk.com and check for yourself. Its amazon’s own platform for selling these sorts of tasks to people. Ethical concerns aside, the applications for this technology that people are suggesting seem pretty dubious at best. LLMs are good at predicting text. That’s it. In essence, they’re just a really nice version of your phones autocomplete. Suggesting that they could replace programmers, artists or… any other profession, really is pretty absurd. Any task that requires logical thinking just doesn’t work. Because, as I said before, all they’re doing is predicting the word that comes next. That might work for simple math or writing essays, but for any complex problems, it fails spectacularly. Not to mention that, if it somehow managed to take over the jobs in a certain field, it would just make that field stagnate, because, guess what, if no one except ChatGPT/[insert other AI here] is working in a certain field, there‘s not going to be any more humans around to generate more data in that field that that AI can use to improve. And because real life, well… changes occasionally, those AIs would quickly become either useless or horribly inefficient because they simply wouldn’t be able to continue learning due to a lack of new data. Or there just wouldn’t be any new discoveries that get made in that field. Both aren’t really worth it though. It probably wouldn’t even save costs.
Currently ChatGPT costs 700,000 dollars to operate. Per day. I don‘t think you need to be very financially literate to know that isn’t very sustainable. So no, ChatGPT isn’t going to do your taxes anytime soon (or ever, for that matter), nor will it replace every programmer or take over the world like some sort of significantly more polite version of Skynet. It will just make writing things take slightly less effort. Whether or not that is worth the millions of dollars of VC investments and weird ethical practices is up to you to decide.