AI

AI. That pair of letters, when I first encountered it, was a Bronze Age city.

It was, according to the Book of Joshua, a stronghold of the Canaanites, the Semitic-speaking population of Palestine before the invasion from across the Jordan by their cousins the Hebrews. The name, ha-`ay, “the heap of ruins,” must have been applied to it after Joshua, on the second and more deviously cunning attempt, conquered and destroyed it, killing twelve thousand men and women. Unknown is what it was called by the peoples who had inhabited it since before 3000 BC.

The site’s name in Arabic is al-Tell, which means exactly the same. All over the Middle East are the tells, precious to archaeologists: mounds built up layer upon layer by the debris of successive human cultures. It is as if this relatively small one is the prototype: THE ai, THE tell, THE ruin.

 

AI 2

The acronym for Amnesty International, useful for many decades, is now in the past. Hijacked. No longer can we conveniently refer to an “AI Urgent Action” or “AI’s recent policy discussion about the human rights implications of artificial intelligence.”..

 

AI 3

There is much about the distinction between artificial intelligence and natural intelligence, that is, the capability and speed of the brain in solving problems and comprehending vast assemblages of data such as the scene before your eyes.

But I’m not sure I’ve seen anything convincing about its distinctness from other kinds of machine program. They, too, perform operations in a microsecond that would take a human several hours. The complexity of any program can be increased with further thought by the programmer.

The distinction suggested is that an AI program can improve its own capability without human involvement. It would have to have been initially designed that way by a human programmer. It can do so by assessing the results of different ways of doing things, and selecting the steps that led to better results.

If so, AI is not fundamentally different from other modes of programming. I, who have no professional training in either, could build such a feedback loop into my own programming.

Here is part of one of my sphere-type illustrations, from 2020, intended to show how Saturn’s oppositions progress along the ecliptic. On the scale that can include Saturn’s orbit, Earth’s orbit is a small ellipse. But I wanted to mark at least some of the month names, so as to show the way Earth is facing at those oppositions – the Gemini opposition has to be near the beginning of the calendar year, an opposition in the Leo or Virgo direction happens in spring. But to show all twelve names would be a mess. A line in my instruction file for the program reads a number for the monthly interval, and another number for the size of the characters. So I set the first as “3,” meaning each 3rd month. That didn’t achieve quite what I wanted, so I used the mouse to nudge the labels a bit, copied one of them, moved the copy to the place appropriate for January and typed over its name.

I could devise feedback loops to do all that. The subroutine would keep running through combinations of numbers and character sizes until two conditions were met: five labels, for the first month and the four containing the quarter dates; and no labels overlapping or too close to each other, as measured by calculating distances between the rectangles containing the labels.

That’s the algorithm, the outline of the idea. To translate it into Fortran statements migjt take me some hours of thinking. Thenceforth, it would take the computer a microsecond to do that particular bit of fiddling with labels. Would it be worth doing?

Yes, if it were likely that I have to tackle the same situation very many times in the future. No, since it is not. For a task that costs the vast human eye-brain-hand system little time, nothing is gained by devising a machine equivalent.

This is not to say whether AI is inevitable or good or bad, whether it will reduce us to a heap of ruins or be as worthy of our blind trust as the arrow of William Tell; only to question whether it is new or mysterious.

 

__________

This weblog maintains its right to be about astronomy or anything under the sun.

ILLUSTRATIONS in these posts are made with precision but have to be inserted in another format. You may be able to enlarge them on your monitor.

One way: right-click, and choose ”View image” or ”Open image in new tab”, then enlarge. Or choose ”Copy image”, then put it on your desktop, then open it. On an iPad or phone, use the finger gesture that enlarges (spreading with two fingers, or tapping and dragging with three fingers). Other methods have been suggested, such as dragging the image to the desktop and opening it in other ways.

Sometimes I make improvements or corrections to a post after publishing it.  If you click on the title, rather than on ‘Read more’, I think you are sure to see the latest version. Or, if you click ‘Refresh’ or press function key 5, you’ll see the latest version.