A different take on the homework machine

Artificial Intelligence

If you have never read, or don't know about, Shel Silverstein's The Homework Machine (from the book of poems A Light in the Attic, first published by Harper & Row Junior Books in 1981), here it is. 44 years ago, Silverstein saw that machines, like humans, are bound to be fallible but in ways that are different from humans. There went the dream of a homework machine, consumed by hallucination. No machine-enabled laziness and its natural consequence, brain rot 🤯 . Until, of course, it re-emerged in the form of LLM-based chatbots.

What I see in Silverstein's fun little poem, however, is different. The key insight lies in

Three?

Oh me . . .

The protagonist student (who today would be called a user) knows that 9+4 is not 3. Therein lies a lesson that applies to LLMs and other blackboxy contraptions: you better have knowledge of the subject matter. If you know when the machine is wrong and when it is right, or if at least you can easily verify whether it is wrong or right, you are empowered to explore and learn. Otherwise, your homework won't get you a lot of love from your teachers.

As I have pointed out repeatedly, if the subject matter is you or your taste, then you are probably the world's preeminent expert (in yourself); but if the subject matter requires technical expertise, from coding to generating molecules, the ability to know what is good, or not so good, makes a world of different.

And to me that is the story of the Homework Machine.

https://www.harpercollins.com/blogs/harperkids/shel-silverstein-poems