AI: Uses and Abuses

There's a lot of talk about AI, which is supposed to stand for "Artificial Intelligence", but could equally well stand for "Artificial Idiocy", when one considers what the term is most frequently applied to -- LLMs (Large Language Models). 

LLMs are computer programs that are fed large quantities of text from various sources, and are programmed to handle the text in various ways, including generating new text based on the patterns they detect in the existing text they have been fed. This enables them to do things like produce summaries of documents, and to gather and present information gathered from a wide variety of sources.

Some of these publicly-accessible LLM bots are programmed to present a quasi-human user interface, which has led some people to regard them as persons and to treat them as if they were human beings.

Wisdom of the chip, compassion of the code

The Barna research group has found, for example, that 48% of practising US Christians say they would trust AI with their spiritual growth, and 34% say AI's spiritual guidance is as trustworthy as a human pastor's. Other research has found that some people even fall in love with chatbots, and develop a quasi-romantic love relationship with them. 

The problem with all this is not with the LLM bots themselves; it is with the human beings who make, promote and use them. 

In all this we need to remember that LLM bots are inanimate machines; they are not AI and cannot think. 

It is important to understand this to know what they are good at, and what they are not good at. One of the things they are not good at is knowing what they are not good at. or "knowing" anything, for that matter.

Some things LLM bots are not good at 

Some companies that have adopted AI agents have discovered that they have botched important tasks. The US military has used AI agents in its war against Iran, and some have suggested that a school full of children that was hit by a missile was a target selected by AI.

In these and similar cases the main problem is that a situation arose that was not foreseen by the programmers of the AI bot. A human being faced with something unforeseen, can make a decision based on what is actually happening. They may make a wrong decision, and often do. But they also can and often do make decisions that mitigate rather than exacerbate a bad situation. 

 Some things that AI bots can be useful for

One of the things that AI bots can be useful for is research, though some caveats are needed.

When I was working on my doctoral thesis some 30 years ago, long before LLM bots, I went to the periodicals section of the university library and walked along the shelves looking for journals with likely titles. Having found one, I would begin with the most recent and go through the indexes looking for articles with likely titles. When I found one I would read the abstract. If it looked relevant, I would put that journal in my pile to have the article photocopied. That is what is called a literature search. It was also very time consuming.

An LLM bot could search the same number of journals, and many more that weren't in the library, and come up with a list of articles in less than a minute. That could save a lot of time, time that could be better used in reading the greater number of articles the bot could (theoretically) find. 

The caveat, in this case, is that the bot may not have been trained on all the articles or journals that dealt with the subject, and, since it works by pattern matching, it could make up article and journal titles similar to those of the ones it found, which could be (and sometimes are) completely spurious. That is why LLM bots should be supplementary to, and not replacements for, a competent and trained subject librarian.

Creativity

The tendency of AI bots to make things up like this should not be mistaken for creativity. But we should be careful about the complaint that what they produce is simply piracy of what human beings have produced. This is especially noticeable in graphic art.  A bot that has been trained on hundreds of thousands of paintings mixes those elements and produces something different, and people claim that it is derivative. But a human artist has usually also been trained in like manner, by seeing pictures produced by thousands of artists before them.

Donkey-centaur (Isaiah 34:14)

I was once intrigued by the biblical description of a desolate place from which the human inhabitants had fled, leaving it to be occupied by various monsters (Isaiah 34:14). I asked a friend who had access to a graphic AI bot to get it to produce a picture of a donkey-centaur, and this is what it produced. I thought it was rather nice. All the elements are derivative: the style was learnt from hundreds of artists who painted in that style, but those artists themselves learnt from the other artists who painted in that style, often as semi-robots themselves, as apprentices painting a small part of a picture by an acknowledged master.

If a human artist sees a lot of paintings in a similar style, and then paints one in the same style, it isn't piracy, unless they try to pass it off as the work of someone else. Similarly, an author who reads a lot of books will be influenced in what they write by the books they have read. That is not piracy. Nor if a bot produces a document in a certain style. But if a bot produces it, though it is not necessarily piracy, it's not really creative either.  

I see nothing wrong with getting a bot to draw such a thing to illustrate an article like this. But I would not present such a thing to an art gallery, as a replacement for the work of human artists. Nor would I ask a bot to produce an ikon of a saint -- that would be analogous to asking a bot for pastoral advice, as described above. And in doing such things, bots also make mistakes, like putting right hands on left arms, adding fingers and toes etc. 

 Using LLM bots for creative writing

A couple of months ago a friend and I heard a talk on  Christian Nationalism. It is a topic at the intersection of theology, politics and history. My friend is a physicist, so it was a bit out of his field, so he asked Claude AI to produce an academic article on Christian Nationalism, which it did. Since that topic is more in my field, he sent it to me and I read it as though I were doing a peer review for a journal in my field.

The article contained quite a lot of information, but it was poorly arranged and poorly presented. It lacked a clear and consistent thread of thought. To put it briefly, Claude AI could not see the wood for the trees. It could connect one idea to the next, but it could not connect a string of five ideas to the next five. 

I had written an article on a related topic 25 years ago, and I was interested to see that Claude had used some sources that had not been available then, and it looked as though that might be more interesting than the article itself. My friend then gave Claude the feedback -- my "peer review" and the bibliography for my article, with the addition of a reference to my article as well. 

It spat out a new article, which was a little bit better than the original, but still suffered from the same faults. But this time my friend asked it to produce a bibliography on the subject, and that was far more useful. It had added mine to its original one, a dredged up a whole lot more. Some of the items, including some of the URLs, proved to be hallucinations, but it was useful nonetheless.

I would not use AI bots for writing stuff, but as a  kind of research assistant, they look useful. 

Claude AI tries fiction

I then asked my friend if he would try Claude on fiction.

I had seen somewhere that the name Tiffany was a short form of Theophania, which gave me the idea of writing a novel on the theme of "Tiffany had an Epiphany". So I wrote one.

My original (complete) MS was The Venn Conspiracy, a children's novel which I have written but not published.

My friend submitted the first two chapters to ClaudeAI. These chapters established the setting, the main characters and the beginnings of the plot. He asked Claude to complete the story. To distinguish it from my original, it was given the title Tiffany the Spy. Since the original is unpublished, Claude AI would be unlikely to find it on the internet and draw on it for the plot, but having a human-written original would make a comparison easier.

I made some comments on what I saw as the main shortcomings in Claude's version of the story. My friend submitted my comments to Claude AI which produced a revised version, on which I have made more detailed comments, though I haven't asked my friend to submit them to Claude. There is a limit to how many iterations one can go through.

I think the experiment shows that "AI" is not good at creative writing. 

One problem that stands out is that the AI bot lacks empathy (though sometimes the programmers try to give it the appearance of empathy) and so does not know what makes a good story. The lack of empathy may seem like a good thing to people like Elon Musk, who thinks that tech must rule, but I believe that tech, like the Sabbath, was made for man, and not man  for tech. 

In fiction, as well as non-fiction, Claude can't see the wood for the trees, and it can't even do trees very well. It can describe trees lyrically but cannot express their significance to human beings because it isn't one. It came up with a plot that was complex, but trivial and meaningless.

For an example of the lack of empathy, consider the following passage:

David looked at her sideways. "You sound like you've been talking to lawyers.",
"I have been," said Tiffany. "Mark's father. And a woman called Miriam Goldstein who's an advocate."
"What's an advocate?"
"A special kind of lawyer who goes to court."
David was quiet for a moment.
"That's brave," he said, in a tone that made it clear the observation was genuine. "Talking to lawyers about this."
"I don't feel brave," said Tiffany.
"Well you don't look it either," said David, which was as close as he could get to saying the same thing that Mark and Asha had both said, and which somehow, coming from David, meant a great deal.

Put yourself in the position of Tiffany in that conversation.

"I don't feel brave," said Tiffany.
"Well you don't look it either," said David.

How would you, or any other English-speaking reader, take that? I would see David's remark as rather offensive, wouldn't you?

The bot has no empathy, no feel for human reactions to things said. And that's because it has no feel for anything at all. It is an inanimate machine. It has a tin ear. All it can do is to spit out "rite words in rote order", to quote Marshall McLuhan. It does that rather well, but we must not mistake that for human intelligence.

So I repeat: It is important to know what AI bots are good at, and what they are not good at. 

One of the things they are not good at is knowing what they are not good at. or "knowing" anything at all. It's a tool; we should neither accept it uncritically nor reject it without qualification, but rather heed the advice of the apostle: test everything, hold fast to what is good (I Thess 5:21).