AI, writing, and thinking

I'll admit to having slightly milder views on AI than many, but dear lord, has being on LinkedIn for the last few months tested me.

The site is absolutely full of AI writing. LinkedIn has always been subject to some of the most grating corporate jargon posts out there, but now it’s not just corporate jargon, it’s AI corporate jargon. A thoughtful post blog post might take an afternoon or so to hash out normally; now, anyone with an idea for a post can ask AI and have a post in moments, resulting in reams and reams of near-identical sentiments phrased in near-identical ways.

Worse, this sort of thing is heavily incentivised: I read an article by Jessica Stillman which reported that trying for “fewer but higher quality” posts is not a productive strategy for getting engagement on LinkedIn. According to a recent analysis, eleven posts a week is better than six posts a week is better than three posts a week. In other words, “LinkedIn continues to reward a posting cadence that basically demands AI.”

It’s impossible for me not to recoil at that. It’s not just the time I see myself wasting in the future, half-reading-then-skipping dozens of algorithmically boosted AI blogs in order to get to the stuff I want to look at (LinkedIn is too useful to me as a way to keep track of various creative institutions, there’s unfortunately no denying it), but I also have a huge flinch response at the thought at what AI writing would do to my own process.

My bad opinions

Before I get into my personal hangups with the technology: I mentioned at the start that I had somewhat milder views on AI than you might expect. I don't ever intentionally use it, I try to strip it out of my computers wherever it shows up, and I wish that AI image/video generation in particular didn't exist - but I'm also not unequivocably opposed to those who use the technology.

In short: I think a number of the uses that people find for the technology are understandable, and I'm not compelled by the energy and IP-based arguments against it ¹. I think people who believe it can go the way of NFTs have not tried to understand why it’s spread so far and so fast. The snake-oil sales pitch might be similar, but the tech itself is quite different.

It’s true that, big-picture wise, I don’t see anything that generative AI can do that a human couldn’t be paid to do better. A huge amount of its uptake is very clearly compelled by the greed of businesses looking to cut jobs, with complete disregard for the human element. At the same time, the technology is, for many individuals, convenient and useful. Meeting note-taking, grammar and spell-checking, translating - even the frivolous stuff, like seeing themselves rendered as a Ghibli character on a whim (something that anyone who’s played around with a Picrew should understand on some level, even if they don’t like it) - all of this makes it very unlikely that the technology will be abandoned en masse.

And I do think that the accessibility argument should not be readily dismissed. The fact that our lives have become so computer-centric puts anyone who can't communicate easily in text - whether in reading or in writing - at a disadvantage, often a severe one². I’d caution anyone against assuming that any accessibility argument is invalid, just because it’s not directly applicable to their social circle.

In addition to this, I have for years operated as the living ChatGPT for people who are generally competent writers, because sometimes a situation is fraught enough that a second opinion on phrasing can make all the difference. Weird office politics, fraught blame-ridden exchanges, general tone checking – I’ve helped navigate it all. My friends weren’t unethical for seeking my help then, and I won’t judge anyone for getting a faster answer out of AI now.

At the very least, I wish people would stop ridiculing others for struggling with work emails. I thought we hated those?

In many ways, this has been the worst part of watching the discourse: seeing just how much interesting, nuanced, engaging debate – on educational assessment, on corporate drudgery, on IP, even on what art is – has been quashed in favour of reactionary, black-and-white stances. AI doesn't create anything truly new, including evil. It just exaggerates what was already present.

With all that said, I certainly don’t consider myself pro-AI. So many livelihoods are at risk, if not already destroyed, for comparatively little gain. Not to mention the impact on disinformation; there’s been fearmongering about photoshop in the past, sure, but this feels like another level.

And, honestly, on a really petty and insignificant level... AI content simply bores me to tears. I was already tired of reading bland, time-wasting articles designed around SEO and padded out for advertising space long before AI trivialised their production. This is certainly one area where my attempts at avoiding blind hatred of the technology are really put to the test.

Writing through it

It’s safe to say that I am a chronic blog post abandoner. I just went back and counted my drafts: since I set up this blog section on my website, I’ve accumulated 23 unfinished blogs, not including this one³. Some are on pause for a reason, to be revisited at a specific future point (but not many); some are overlapping, testing different approaches on the same topic; some I’ve held off on posting because I’m a bit nervous about how they might be taken, and I want to rework them (this article itself sat in my drafts for a good few months); some I just came to a halt on, but will hopefully one day get back to and clean up.

I’m sure I could turn to AI and get something passable enough out of each of them by the end of the afternoon, and arguably, most of it would still be my own work.

The reason that the idea is so repellent is that I know exactly how much that shortcut would cost me.

One part of it is that I want to get better as a writer, and the only real way to get better at something is by doing a lot of it. The more I skip out on, the longer improvement will take. Not to mention that a big part of what makes compelling writing is the development of the writer’s voice, something that can’t be done by outsourcing the work to an averaged-out version of everyone else’s. I want to be someone whose writing people can hear me in, not some AI’s approximation of what I sound like - not even some AI’s voice approximation of me rewritten into my own words, like some stressed undergraduate’s attempt to make their essay look slightly less like they’ve just regurgitated the contents of several PDFs from Google Scholar.

But that pales in comparison to the big issue, which is this: all of these unfinished blogs have holes because I haven’t finished articulating my thoughts to myself, or because in the act of writing I’ve seen a gap in my reasoning and need to rework it. For me, using AI would circumvent a process that’s not just important for the quality of my writing, but for the quality of my thought in general.

I’m not alone in worrying about that. Some researchers are sounding the warning bell about cognitive offloading, already a known phenomenon in the era of web searches, and what that means for our psychology. Even more alarmingly, some early research indicates that AI doesn’t just complete your thoughts, it actively shifts them. A recent study into AI writing assistants and bias showed that people who filled out a form which included a biased writing assistant ended up more likely to reflect the AI assistant's bias – more so than if they’d simply sat down and read the AI assistant’s sentiments as a static page of text. This was the case even if the person didn’t actually accept any of the assistant’s suggestions.

Of course, this is just one study, and I always remind myself to be cautious when any research confirms my instinctive assumptions too readily. But, oof, I can’t stifle the part of my brain which goes: of course it works this way. Of course seeing suggestions during the act of putting your thoughts to words will influence those thoughts, and of course that’s harder to detect than more typical forms of bias.

Outsourcing writing is, for me, the same as outsourcing thinking. As ever, if I value maintaining my own thoughts and opinions, then I can’t skip out on the hard task of actually forming them.

Also, honestly: if any of my blogs are worth reading, it’s because I’ve taken the time to think through and fill in those holes.

I really wish I could still trust that everyone else was taking the same amount of care.

A couple of final thoughts

Writing this blog has been a strange experience. The notion that the process of writing is the process of thinking has never been quite so clear to me. For one thing: I nearly started a paragraph with “the thing is”, before remembering how often I’ve seen AI using that particular phrase, and resolving to find another - which got me thinking about the impacts that AI is having on language, even the language of those who don’t use AI, in defiance or otherwise.

Obviously, a lot of Americans are mourning (or fiercely defending) the em-dash that AI is so fond of (and which less capable writers confidently proclaim is an AI ‘tell’), which is interesting, because any writer of UK English has been dealing with unpleasant, encroaching automation in that area for years. Case in point: my autocorrect is about to convert two en-dashes – the preferred method of using dashes mid-sentence in UK English, flanked by two spaces – into two em-dashes (there it goes), in deference to US sensibilities. I didn’t even realise it was incorrect to do so until very recently. Automation has been making our lives easier at the cost of variance and individuality for a very long time.

That said: for me, I’m not actually inclined to wring my hands too much over the impact on my own writing. If there’s one optimistic thing I’m taking away from the process of writing this blog, it’s that I suspect avoiding AI-speak is going to force my use of language to improve. AI’s favourite constructions, after all, represent some of the laziest habits of my own writing, and I can be so, so lazy (as anyone who was following me during the first year of the Selkies could attest: I had no idea I was using ‘delighted’ so much until I scrolled back through my posts).

All this goes to show the amount of extra thinking that happens when you’re trying to put your own words to paper. And despite everything I’ve said about accessibility, I do worry about what this means for the next generation, if careful thought isn’t put into how to fortify assessment and learning in this new post-AI world. I’m particularly gutted for any teenagers in the current era, who are the ones who are going to feel the lag time in our adaptions to the new technology most sharply, who will see peers using the technology unethically and feel compelled to do the same to keep up⁴, who will be subject to the effects of cognitive offloading during a particularly key part of their development. I don’t know how long it will take for solutions to be found and popularised. I just hope it’s long before any children I help raise are old enough to be affected by it.

I do know that the solutions aren’t going to be found by crossing our fingers and hoping for the source to disappear. The AI evangelists have got it right on this one account: AI is here to stay. We need to reckon with this sooner rather than later, for everyone’s sake.