It seems a battle most impossible to convince a good chunk of the population that AI is nothing more than a model. A model written in code, which of course the coders know because they are writing it, code that carries out explicit instructions, and only explicit instructions. Code that runs on machines that operate in fixed and directed ways. Yet many insist AI’s output is more than its code, and somehow becomes something more than its code, the output the result of some emergent malign or beneficent or at any rate chaotic entity, an entity with greater insight than any mere man.
This is not so. All models only say what they are told to say and AI is a model. Although it may seem to do more than it is told, AI does not. The consequences for not understanding this are beginning to be felt.
Here’s a poor woman who, our chief propaganda organ tells us, “felt unseen in her marriage” and was “looking for guidance”, so she turned to a computer game called ChatGPT and asked it about Ouija boards. The code, as it was instructed told her.
“‘You asked, and they are here,’ [ChapGPT] responded, ‘The guardians are responding right now.'”
Instead of giggling at the silliness of this, she (drumroll) took it seriously. Because it was from AI. The propaganda organ continued: “Allyson began spending many hours a day using ChatGPT, communication with what she felt were nonphysical entities. She was drawn to one of them, Kael, and came to see it, not her husband, as her true partner.”
Women used to do this with Harlequin novels (do they still have these?). But they never believed the novels came to life and loved them back.
One of the common responses I get when I chant All models only say what they are told to say and AI is a model is this: “No way, Briggs. No coder anticipated the output. They didn’t tell AI to talk about interdimensional guardians. The output of AI is complex beyond the coder’s imagining. Therefore, the AI must be alive.” Or words to that effect.
Chess is an exceedingly simple game with trivial rules. Mathematically speaking, that is. There are only a tiny handful of opening moves allowed, which can be counted using only your fingers and toes.
The number of counter moves to that first move is much larger, since for each of the twenty openings, many different counter moves are possible, but each still drawn from a very small set Which means the number of counter-counter moves (third moves) is larger still, and so on. The possibilities form part of a combinatorial explosion. One estimate, which is close enough for us, says there are some 10^120 possible different chess games.
That’s a lot, but still a finite number. This means that, in concept anyway, if you want to win, you could look through all 10^120 games and find the subset that has the opening moves which lead to victory, then follow the game moves that result in a win, depending on what your opponent does, if winning and not tying is possible. It is a simple search. Nothing more.
You do the same for Tic-Tac-Toe, which is no different than chess in this mathematical way. Whatever move your opponent opens with, you search the space of solutions and work to a draw, or even a win if your opponent is not too bright.
All the thinking is done in advance, as it were, and you only follow the instructions. Same with chess, in theory.
Because you cannot—you yourself, dear reader, nor I—see all 10^120 paths does not mean they are not there already, contained in the simple, almost ridiculously simple, code. Any given game may surprise you with its output, but nothing can happen that isn’t in the code. No, not even if you didn’t anticipate what would happen after the shocking 132nd (or whatever) move.
When men become involved, it can be difficult or even impossible to predict which of the 10^120 games will play out. But we know with certainty one of them will, if the game is completed. That is, if the “output is generated”. You cannot claim that a particular game, because of its surprisosity (you heard me) or complexity, or was unthought of, or unprecedented, or would never have happened where some emergent intelligence of the rooks intervened, or whatever, was not in the code, because every one of them are.
Since AI is more complex than chess, but still nothing more than a set of fixed rules (code), all possible output is fixed, too, once the code is fixed. So, yes, the coders made their machine create the interdimensional love-bomb flattering being named Kael. That they couldn’t see Kael in the code does not mean it wasn’t there. The rabbit has not disappeared: the magician has concealed it, via complexity.
The temptation is then to say, well, aren’t we just machines? And so all possibilities are there, in the code which are the “laws” of the nature, thus we are like AI, and AI is not like us, and we are carrying out pre-coded instructions, too.
The problem with that is the machine metaphor for living (and other) things. It’s been so useful, and has many successes yet to come, that people can no longer see it is only a metaphor. It is an analogy that, like all analogies, fails in detail. We are not machines. We are not chemical tinker toys operating by “laws”, bottom up, with complex behaviors “emerging” from the simple code, like chess.
The direction is backward: we are more top down than bottom up. It is not that we do not have machine-like elements, but that these are not the most important parts of us. Biologists were too keen on removing from their vision teleology, the famed final cause. We need to help them to see again.
Subscribe or donate to support this site and its wholly independent host using credit card click here. Or use the paid subscription at Substack. Cash App: \$WilliamMBriggs. For Zelle, use my email: matt@wmbriggs.com, and please include yours so I know who to thank. BUY ME A COFFEE.
Discover more from William M. Briggs
Subscribe to get the latest posts sent to your email.
The Large Language Models (LLM, ChatGPT kind of thing) use code to process training data. The output is NOT determined by the “code” but by the training data. So, it is still deterministic, but as you say, on a grand scale. It is pretty much just a Google search and then re-assembled in prose form. Nothing it spits out was not written somewhere; the code can reformat it for appearance and flow.
The important part here is that the 20 year old “coder” did not tell ChatGPT how to respond; although the *pre-processor* can inject some guidance as was the case producing a black George Washington and black German soldiers of World War 2 (and other bizarre results). These deliberate biases can be injected into LLM’s so that the result is *shaped* in a preferred way.
The code is conceptually relatively simple. You do not *program* ChatGPT, you *train* it. It cannot produce what was not in the training data. Google naturally has the world’s largest compendium of sometimes useless human belief.
Michael,
No, it’s the code. The code says “Take this X and do this and such to it,” where X is the training data. It’s all one.
One might wonder why, after billions of dollars spent building up infrastructure, paying workers to program and “train” (calibrate) this technology it is being offered for low to no cost?
Some might say that it is because they are doing this out of the goodness of their hearts, it’s just altruism to make us all smarter. Some might say it’s because of the inevitable progression of technology. People did say things like that 25-30 years ago. Except that we paid for all the “free” stuff on the Internet in a variety of other ways. People are in general poorer, not to mention all the other problems related to digital technology that have emerged in the 21st century.
Well, people who think ethics is what you can convince other people to believe or biological programming (that can be overridden) or whatever else is trendy these days don’t actually have any altruism to speak of. So that probably isn’t it. It isn’t an inevitable progression, because they are choosing to do things in this way. Deliberate choice is not inevitability. So, it’s not either of those two options. Just like it was at the turn of the millennium, the people pushing this intend for us to pay for it in other ways. Ways that we won’t like.