Update on this article (5/1). I had two different things on my mind and did not get them separated! Let’s try again.
There’s been a lot of talk in the news about new Artificial Intelligence entities that do pictures or words. People worry about whether they will take over the world or cause all artists to lose their jobs, etc, etc, etc. Is this likely to happen in the real world?
Rachel shared an article with me some time ago from the Washington Post in 1965, byline Rosemary Donihi.
“Equate the child to a miraculous computer, self-coding, self-feeding, self-propelled. There can be no excuse for not programming him with maximum respect.” That’s the opener from Father.
What is in his [the child’s] mind at any given time is what he is.Mom, 1965
Mom says, “A computer has to have a memory. … You can’t give it problems without adequate memory. The child is like a camera, he takes pictures with indifference. What is in his mind at any given time is what he is. This is the mother’s business — to stay with the child, to interest him, to give him opportunity, to help him grow to his optimum self.”
Both of them are talking, in the end, about the words that they will use to children and that the children will have in their own minds (self-coding).
Father used to say explicitly that you don’t know what you think until you write it down, that is put it into words that anyone else can see. As long as it is all in your own head you can fool yourself about what you actually think.
I’ve gotten spam offering to write content for this blog using AI, every three or four days. Obviously if I let AI write my content I’ll never figure out what I think. And then the question arises! Will only an AI *read* my content as well? Could you really have a world where AIs wrote stuff and other AIs read it and responded? Where AI starts to take over the world? No, I don’t think so, for several reasons.
It’s clear from the descriptions of AI that the actual programs are run by people who have some pretty specific ideas of their own and the AIs are programmed to stay within the parameters of those ideas. There’s a remarkable article on this blog (ACOUP) from a University professor who has received assignments written using AI. He wrote
It is crucial to note, however, what the data is that is being collected and refined in the training system here: it is purely information about how words appear in relation to each other. That is, how often words occur together, how closely, in what relative positions and so on. It is not, as we do, storing definitions or associations between those words and their real world referents, nor is it storing a perfect copy of the training material for future reference. ChatGPT does not sit atop a great library it can peer through at will; it has read every book in the library once and distilled the statistical relationships between the words in that library and then burned the library.https://acoup.blog/2023/02/17/collections-on-chatgpt/
The current AI makes things up, invents sources when it writes ‘scholarly’ articles, and cannot grasp things that are obvious to almost any human reader. The example Acoup.blog gives is asking the AI to write about two books that discuss some Roman happening. The AI writes an essay explaining similarities in how the books discuss the event. BUT it misses completely the fact that book #2 was written to discredit book #1 even though that’s how book #2 begins.
AI is not doing analysis. It isn’t even doing a good job of facts. It is associating words that are found constantly associated in whatever it was given for training materials. That’s one reason it will fail to take over the world. It can’t keep up with even one reasonable grownup when it comes to analysis.
It is also incredibly expensive to run.
I remember my daughter talking about artificial kidneys. They are the size of a refrigerator and need to be plugged in and don’t filter as well as the real organ. Kidneys are about the size of a fist. AI feels like that. It’s huge compared to a human and doesn’t do as well at taking over the world.
There are lots of smaller jobs that AI can and does do well.
I’m including the picture that went with the WaPo article because I love it even though it isn’t really relevant.
One thought on “Cameras that click”
This very encouraging. I was figuring we could all just learn to speak in poetry and cultural allusions, slightly misspelled, which AI would not get. But this makes it look even easier. We can dodge this thing, also we can deceive it. It is essentially stupid and does not get jokes.