Back in my teaching days, my colleagues and I used to assign “The Big Book,” which challenged our students to write 40 pages of original prose in a variety of genres centering on a common theme. Our students groaned the whole way through the yearlong project. But at the end, when they published their book, saw their words in print, and shared their work in a public exhibition, they beamed with pride. Their hundreds of hours of painstaking hard work had paid off.
But what’s the point of trying so hard? Soon, with the help of artificial intelligence and GPT-3, young people (and adults!) will be able to whip up perfectly cogent essays in the matter of seconds.
In this mind-boggling article, Steven Johnson explains how neural nets and large language models have combined to craft language that is becoming indistinguable from (and sometimes better than) human writing.
Mr. Johnson writes:
The machines have acquired language. The ability to express ourselves in complex prose has always been one of our defining magic tricks as a species. Until now, if you wanted a system to generate complex, syntactically coherent thoughts, you needed humans to do the work. Now, for the first time, the computers can do it, too.
Are you scared? I am. Despite the wishes of GPT-3’s founders to keep their technology open source in order to “benefit humanity as a whole,” there’s no guarantee that users won’t employ the software for nefarious ends. A current version sometimes spits out racist rhetoric – perhaps an accurate current portrayal of our species. A less-racist update is more palatable but sounds like a proponent of critical race theory. How do we teach values to a computer? Who gets to be the teacher? (45 min)