Artificial Interns

In a recent interview in The Tim Ferris Show, Kevin Kelly shared a framing for AI that resonated with me. As a matter of fact, I used the same framing when describing my experience using Copilot to a friend1: You can think of the latest AIs as Artificial Interns.

A photograph of a robot wearing glasses and a shirt, in collaboration with DALL·E 2

To understand this, consider the process of generating the illustration for this post. I called my artificial intern, i.e. opened the DALL·E 2 web app, and told it “draw me some humanoid robots with glasses and a shirt.. I’m thinking of something like the movie I Robot.” The first results were disappointing and as close to the robots from I, Robot as I am to Chris Hemsworth.

Like any young, inexperienced intern, the AI needs precise direction and guidance. I removed “humanoid” from my prompt and got something more in line with my expectation.

Still nowhere near what I originally pictured, but I liked the goofy aspect of the last image, and I decided to stop there.

A good manager tailors their approach to suit the person they’re working with. We should approach interacting with the young artificial interns that are joining our team in the same way. DALL·E and I started working together only recently. Most of the images I can get it to produce are mediocre at best. Yet I know it’s capable of great results when stirred in the right direction.

Notice that at no time during our collaboration to generate the illustration DALL·E showed any initiative. It did what it was told and nothing more. If we were talking about a human intern, this would be quite disappointing, but I’m okay with my artificial intern not pitching its own concepts for the moment.

Initiative is what sets humans apart from generative AIs. ChatGPT can generate coherent text and format it in any kind of style, but it doesn’t know what your readers want. DALL·E can mix and match different artistic styles to produce illustrations, but it doesn’t know which is most suited to your blogpost. Copitol can suggest code that compiles, but doesn’t know which feature to build next.

Maybe these are only temporary limitation of a technology in its infancy. It could be that our artificial interns will one day mature, get a promotion (associate intelligences?) and annihilate the demand for humans in the job market. Until then, the only competition I’m concerned about is that of the humans that know how to direct their interns better than I do.

1This is the whole reason I’m writing this post, really. To humble brag about having had the same idea as Kevin Kelly.