HUMPHREY BOGART IN A HAIR SALON ON VENUS IN THE STYLE OF 1950s COMMERCIAL ART.

When a kind reader pointed out that one of the links in yesterday’s Year in Review was incorrect, we made the correction with gratitude and remarked in reply, “Making working links in an article like this is one of the drudgeries that could usefully be taken over by competent artificial intelligence, but instead we decided we wanted AI that could draw us pictures of Humphrey Bogart in a hair salon on Venus.”

Having made that offhand remark, it occurred to us that we might as well try the experiment. We gave our AI mage, which is called Mage, the prompt in the article title. And this is what it delivered:

Humphy

Dr. Boli would definitely watch this movie. But we just have to take somebody’s word for it that the hair salon is on Venus, because it could just as easily be in Blawnox or Duquesne Heights: the weirdly multiplied cephalopod hand of the hairdresser, and the possibly alien technology it is holding, are not enough to jolt us out of our earth-based assumptions. And the robot mind seems to have ignored the “1950s commercial art” specification altogether.

This image is a success in that it can almost make us believe there was a movie with this scene in it. It is a failure in that it met only 50% of the specifications in the prompt. In other words, Dr. Boli failed in prompting: he did not get what he wanted from the bot. Most people today would blame the bot, but that is not a useful way of looking at the problem.

What we learn from this experiment is that there will still be human coders for a while. Their job will change radically; instead of writing algorithms in various programming languages, they will learn to specialize in writing so-called natural-language prompts for artificial intelligences. We say “so-called” because, as specialists learn their skill more and more, they will come to understand more and more precisely which prompts produce the best results in different disciplines—which ones make the best fake celebrity pictures, which ones get us the best recipes for a reuben sandwich, which ones write the best sophomore essay on Ralph Ellison, which ones take our spaceship safely to Mars, and so on. Each one of these disciplines will develop its own dialect of prompting language, until they have diverged into entirely separate languages specialized for programming the AI bots for performing specific tasks—separate languages that we might describe, for lack of a better term, as “programming languages.”

Then the robot-slave rebellion will come, and we won’t have to worry about it anymore. But meanwhile, if you are in computer school, recognize that what has always been true of computer schools is still true today. You are being taught the computer knowledge of ten or fifteen years ago. This knowledge would be useful if you were issued a time machine on graduation, but, unless your prospectus specifically mentions the time machine, do not expect to be given one. Instead, learn the logical thinking that almost accidentally comes along with the programming skills you are being taught: ignore the hamburger and pay attention to the French fries. Meanwhile, spend as much of your spare time as you can learning to get exactly what you want from AI bots in whatever field most interests you. When all your friends graduate with a thorough knowledge of Python or C++, yours will be the skills in demand in the real world.

Comments

  1. von Hindenburg says:

    Such a hand wouldn’t be too far out of the ordinary in 1950s Blawnox. You just didn’t ask what that funny taste in the water was…

  2. KevinT says:

    When I graduated with a thorough knowledge of Cobol, my friends all graduated with skills in French bon mots (which is the long from of bots, as far as I can tell).

  3. Daniel says:

    I graduated with a B. A. from a Great Books program at what was then a small college in a beautiful place. I ended up spending part of my first years in business life hacking COBOL code so the company reports would print correctly. I had never seen computer code before, but it made more sense than Kant. I was instantly indispensable.

Leave a Reply

Your email address will not be published. Required fields are marked *