I haven’t addressed the phenomenon currently known as AI, except in passing. This is what I will be doing this time, although I won’t be doing a deep dive into the topic.
Why does artificial intelligence or, in short, AI matter? Well, there are multiple reasons for that. Firstly, it’s already everywhere and, in fact, it already was everywhere, before it became this thing that is, supposedly, somehow revolutionary, even though it has been around for a long time already. Secondly, it’s a business. It’s all about the money or, to be more accurate, all about the flow of capital. Thirdly, it’s a social concern. Fourthly, it’s an environmental concern.
I’ll unpack the first reason in this essay. So, to start with, it’s worth noting that AI is not any more artificial than anything that has been made by humans and non-humans alike. It is equally artificial as a fork or a bird’s nest. There’s also nothing intelligent about AI. It’s a program, among other programs. It’s certainly different from other kinds of programs, sure, but it is only a program. It does what it is programmed to do. It works, but it only works as intended. Even when it works in ways that appear unintended, it is working as intended. That’s not a feature. That’s a bug. There’s a difference between the two, sure, but it is still a program. It’s been programmed to do something, in a certain way, and the way it does it depends on the programming, which may be this or that good. Think of a car. Some are better. Some are worse. The difference is not up to the car. It is up to its makers. There is, of course, wear and tear, but that’s also the case with computers.
Artificial intelligence or AI is just a buzzword. It’s just software, among other software. It has its uses and it already has had its uses for decades by now. It’s all about machine learning. You feed the program data, marking it as such and such, so that when you test more data on it, it can mark it as such and such. It’s that simple and very useful. This has been in industry use for ages already. For example, if you mass produce something, let’s say microchips, you are working with something tiny products. It would take ages to go through them, to see which ones work and which ones do not work. So, you opt for a visual inspection. You photograph each chip and then compare the photos with existing photos of faulty chips. This would also take ages to do. So, you develop a program to do this for you. It has been fed with a lot of data, in this case photos of circuitry. Some photos have been tagged as faulty, because they were indeed faulty. Others have been tagged as not faulty, because they were not faulty. Then you task the computer to compare the data and flag any chips that match the ones flagged in the data that was fed to it. Now, there can, of course, be fully functional chips that end up being flagged as faulty, but that’s a minor issue. That can be fixed by setting all the ones flagged as faulty to the side and going through them again, checking whether they actually work. The photos of the ones that do work are then flagged as working, while the rest remain flagged as faulty. This data can then be used as more data to improve the process. The whole point of this is to make the process efficient.
To be clear, this also means that what’s labeled as AI does not think, nor sense anything. So, when a computer is tasked to check if a chip is faulty, it does not look at a photo and then think that it must be faulty because the photo of the chip looks a lot like a photo of another chip. It only ever compares the pixels, whether each pixel has a certain value, like each pixel does, and then compares those values with the pixel values of other photos from the so called training data. That’s it. That’s what happens. No seeing. No thinking.
Okay, to be fair, the process is actually a bit more complex than that, considering how compute intensive it is to check each pixel individually, but that is beside the point. They actually work on a number of pixels, this by that many, which are then the tokens that it works with, but that’s a matter of efficiency.
This does not mean that this is not highly useful. It is. It is very, very useful. It’s all about pattern recognition. Humans do this all the time. It’s just that humans have their limits when it comes to this and by this I do not mean that computers are somehow superior in this regard, but rather that humans tend to get bored by repetition. If you had to manually check, to visually inspect photos of circuitry, comparing them to see if the chips that are produced are faulty, you’d get sloppy, sooner or later, assuming you hadn’t quit already.
This also applies to writing and speech. It works exactly the same way. There’s data and then that data is matched with more data. It’s easy to be fooled to think that, ah, a computer program is thinking when it provides you with something in some form, such as writing, speech, images or moving images (which are, in fact, images that appear after images, one after another). The thing is, however, that the program does not know how to write, to speak, or to create images or moving images. It doesn’t know how to do anything. What it provides you with is something that is only ever a regurgitation. It taps into existing data and then provides you something that is based on that existing data. The reason the program manages to fool anyone is because it appears to know how to write, to speak or to create images or moving images. However, it only ever provides you with something that is based on existing data. It checks that, okay, there are these words or, to be more accurate, tokens, which can be words, parts of them and/or punctuation marks (depends on how it is all tokenized, of course), and they tend to appear in certain order, one after another, so it goes with that. With images, it’s the same thing, just with bundles of pixels that get tokenized and it goes with that.
What interests me, in particular, is not that people get upset by this, but why they get upset by this. For example, when someone, let’s say a student, has used one of these programs to provide them with writing, it makes it appear that the person wrote what the program provided them with, even though the person did not write it. Others are then fooled by this. They are impressed by how well someone writes. There are no poorly constructed sentences, no small errors like wrong verb forms, not even typos. They are upset when they come to realize that the person who claims to have written it has not, in fact, written it. It might be that they come to read something else written by the person and realize that the person cannot be this good a writer. It might also be that this is revealed by some plagiarism checker tool. The thing is, however, that they get upset by how a program can provide them with something that is equally good as what the best of writers can write. This is, however, only the case because what’s considered good writing is highly formulaic. There are, of course, different genres, but a program can be tailored to take this into account. The data is simply flagged accordingly.
What I’m after here is that people, especially academics, get upset by this because a program can provide them with exemplary writing, such as academic writing. There is this idea that it takes years of practice to do this. The irony is that it is very easy to produce academic writing, for the simple reason that it is highly formulaic. For example, you have all these linking words that often appear in academic texts, such as ‘however’, ‘nevertheless’ and ‘therefore’, but do not appear that often in non-academic texts, instead of ‘but’, ‘yet’ and ‘that’s why’. There is also this expectation to avoid contracted forms, such as ‘it’s’ and ‘don’t’, and use the full, non-contracted forms, such as ‘it is’ and ‘do not’. There is this formality to the genre. This is, of course, not to say that all academic texts are this rigid, but the vast majority of them are. There’s this vast pool of data that are flagged as belonging to this genre. It is then also flagged by when it was published, in what kind of publication it was published etc. This makes it extreme easy for a program to simply provide you with exemplary writing in this genre.
For me, this is highly ironic. There is this gatekeeping, that one ought to write in a certain way, just as one ought to speak in a certain way, or to draw, paint or photograph in a certain way, etc., and when this expectation is met, with ease, people get upset by it. The problem really is that the criteria for what is considered good when it comes to writing, speaking or expressing something in some other semiotic mode is flawed. If it is the form that matters, then, yes, exemplary form you shall have. These programs make mockery of that, really. What they are capable of providing people with are so exemplary that they end up being parodies, which is what really upsets people, especially academics.
The good thing is that this is an easy problem to fix as there’s nothing intelligent about these problems. They can’t do anything. There’s nothing creative about them. It’s all just regurgitation, like vomiting what’s been digested, which is why whatever they provide people with often appear so unappetizing. Anyway, this simply shifts what matters away from form. It’s now all about the content. Is it new? Is it interesting? This forced people to reconsider what it is to write or to speak and to think of language more as an art than as merely stating the facts, as people often mistaking think what language is all about. This, of course, also applies to semiotic modes of expression.
The bad thing is that people might not want to fix this problem. The problem is that there’s no going back to a time when it was not possible for programs to do all this. That time is gone. There’s no way to stop people from using these programs to make sure that they have written is as perfect as it can be, form fitting, really. Plus, the problem here is that they could learn to write like that by themselves, according to the norms, what fits this or that genre, but you could say and the companies that develop these programs would certainly say that this is a waste of effort. If a computer can do all that for you, why would you? It’s exactly like with how it is with producing microchips. Why would you make humans do something so repetitive? Why wouldn’t you task the human to focus on creating something new? The programs cannot do that, because it’s not intelligent, but humans can do that, because they are intelligent.
The reason why people might not want to fix this problem is that they are used to working the way they do, which amounts to learning a certain pattern and then repeating that pattern. This also applies to academics. It is rare, extremely rare that you find any journal article, book chapter or a book that takes any chances. It’s usually the same stuff, over and over again, year after year, not because they are not intelligent, nor capable of creating something new, but because they’ve grown up in education systems that do not value creativity. They are expected to do things a certain way, or else, and they therefore do those things that way. They are also incentivized to do things that way, so it’s unlikely that they’ll ever do things some other way.
There are many reasons to oppose what’s dubbed as artificial intelligence or AI, but this is not it. This is actually what it does best and serves people the best. It may upset people, that’s for sure, as I’ve outlined here, in this essay, but that’s more telling of their insecurities than anything else. I actually see this as a wake-up call. It ought to make us rethink why we write, speak, or express something in some other semiotic mode and reconsider our priorities.
The other reasons that I mentioned are far more important than this reason to oppose these kinds of computer programs. I initially thought I would do that in this essay, but I think they warrant their own essay or essays.