AI and the End of Thinking

I hypothesize that the topic that comes in second after the US election in the news is AI. So, this post is about me….cuz in most fonts, Al and AI are indistinguishable.

This was prompted by a long article in last Saturday’s Globe and Mail Opinion section, titled

The Automation of Writing is Almost Here

Followed by the tag line:

‘But what will happen to us, Michael Harris asks, if we cede our written language to AI?’

It is a long article, and Harris makes a number of points, all of which I basically agree with.

As to what is wrong with AI writing, he says “There is the shape of meaning and yet nothing solid.”

I would only add that this can also be said about most bureaucratic writing from universities, government, corporations. Indeed, AI makes all writing sound bureaucratic – only the shape of meaning.

He typifies the coming of AI as the third great revolution in communication, the first being writing itself, the second the invention of the printing press.

Another quote: “Passing around pre-fabricated blocks of text is efficient, to be sure, but it also makes a mockery of the word ‘communicate’….”

Again, I cannot disagree, except perhaps to wonder at what ‘efficient’ means in that context. It is easy, yes, but efficient?

As someone who taught a lot of what my employer deigned to call ‘Essay courses’ (simply meaning student grades were determined by my evaluation of a minimum number of their written words), I said to those students often that ‘writing is thinking’. I meant it, even though most of them didn’t like what that implied. Thinking is hard, so writing well is also hard. Students….no, humans….are pretty ingenious at avoiding things that are hard. That’s not necessarily a moral failing. Well, it is when it gets people hurt or killed, but it is also probably part of our evolutionary heritage. Struggling along on the savannah, smaller. slower and weaker than most of what we wanted to eat, our ancestors absolutely had to conserve their energy in any way they could, so as to have a hope of winning a battle against a tougher, stronger and faster opposing animal.

The instinct to conserve energy seems to have survived, the need to best superior predators, not so much.

There is another vein of thought about AI and language that I have read about but which Harris ignores. So far as I understand it, it is that AI generated language is destined to turn into meaningless mush, eventually. It has something to do with the fact that AI will increasingly find itself training itself on a mountain of verbiage out there on the web which it has increasingly generated itself, and this process is inherently unstable. Here’s two sentences from the Abstract of a paper about this, titled ‘The Curse of Recursion’

“What will happen to GPT-{n} once LLMs contribute much of the language found online? We find that use of model-generated content in training causes irreversible defects in the resulting models, where tails of the original content distribution disappear.”

It doesn’t say what those ‘irreversible defects’ are but some simple examples in the paper suggest that as a LLM (Large Language Model) is trained successively on text that it generated itself, the output it produces becomes what we would now call gibberish.

Given what I write below, I would call that good news.

My own concern with these LLMs follows from what I see their owners selling them for already. It is an extension of the same concern I had the few times students used AI for their writing in the last years of my career. I came upon these recent AI sales pitches because I watch sports online, and the commercials are mostly those that are broadcast on US TV. The ones I am going to talk about below are all, so far as I can remember, from Apple Intelligence, the cleverly named AI wing of Apple.

I recall three such TV ads.

1. A guy enters a business meeting, sits down, and quickly realizes he has no idea what is going on. He quietly scoots his wheeled chair out the door, checks on his phone for what I assume was the original emailed materials for the meeting, and hits the ‘AI’ button on his program. The algorithm immediately gives him a ‘summary’ of said material. He smiles, slides himself slowly back into the meeting, and says ‘OK, let’s get into the prospectus’.

Lesson: You don’t have to prepare for anything, and you certainly don’t have to study. AI will prepare you in seconds.

2. A guy is sitting at his desk, thinking out loud about something that is never clearly described. He speaks some garbled thoughts into his phone, hits the AI button, and his gibberish is turned into a paragraph of clear prose, which he sends to someone (his boss, it appears). Boss reads it, looks up and says something like ‘This is from Gibberish-guy?’.

Lesson: You don’t have to write or even think clearly, just say something, anything, and AI will make it brilliant.

3. A guy (natch) is incensed about someone stealing his pudding cup from the office fridge, and he is writing an inflammatory email to the entire office, castigating whoever was asshole enough to steal it, and promising dire consequences if it is not returned. Before hitting Send he looks at the Teddy Bear sitting across the office from him (No, there were no teddy bears in the office when I worked, either) and hits the ‘Kindness’ button on his AI program. The email is duly transformed, he sends it, and a young lady walks over to his desk, says ‘Such beautiful words.’ and returns his stolen pudding cup.

Lesson: You don’t have to emotionally regulate or understand proportionality, AI will make you seem like a nice (and articulate) person.

If AI is going to self-referentially crash some day, I say bravo. I fear however that the message to most of my fellow humans prior to that is that it will eliminate the need for them to think, be articulate, prepared, or an adult, and that will be too attractive to resist. However, I will have to interact with those people in situations in which they cannot rely on AI. In line at the grocery store, out for a walk, sitting at the bar. Doing so will be extremely unpleasant, because without their AI they will be ignorant, inarticulate and angry.  Unless they let their phone do their talking for them.

That’ll be great.