Shakespeare it's not. However, a new artificial-intelligence technology called GPT-3, which understands and generates text, could help students write essays -- or help them avoid writing essays, writes Williamena Kwapo on Education Week.
. . . if you input “write a tagline for a car company,” it will return a completion like “a car for every budget” or “the best way to get around.” Prompt it to “summarize a complex text for a 1st grader” and it generates a summarized text in language that a 1st grader can understand.
. . . GPT-3 does not just understand words; it also understands context, sentence structure, and dual meanings in multiple languages, allowing it to summarize text, write outlines, essays, reports, and recommendation letters, all in human-like prose. And beyond writing, the model is trained in producing content related to math and science.
Another AI tool, Outline, creates a detailed outline for students but doesn't write the entire assignment. CEO Sydney Montgomery said it can be used as a tool rather than a substitute for original writing.
Open AI, which developed GPT-3, has posted a free trial, writes Kwapo. She asked it to “write an essay for a 5th grade science class on the Webb telescope.” The only flaw was that the bot didn't realize that 2018 is in the past.
The Webb telescope is a very powerful telescope. It can help us see things that we can’t see with other telescopes. For example, the Webb telescope can help us see stars that are very far away. It can also help us see things that are very small, like planets. The Webb telescope is also very good at seeing things in the infrared. This means that it can see things that are very cold, like the planets in our solar system. The Webb telescope is going to be launched in 2018, and it will be a very exciting time for science!
She also tried it on Romeo and Juliet, generating a ho-hum essay on the play. The AI seems to think R and J's "love is so strong that it overcomes the obstacles in their way" because they are "united in death." I guess a 10th-grader might read the play that way.
When Kwapo asked teachers to evaluate GPT-3's writing, an engineering teacher said his students would find it useful in writing a pitch letter for a product they're developing. They're not skilled writers and he's not skilled at teaching writing. The AI pitch -- very repetitive -- would be a starting point, he said.
English teachers aren't impressed, though they concede the AI's work would earn a passing grade. GPT-3 isn't good at reasoning and lacks "depth, voice and authenticity," the teachers say.
Lauralyn Taylor, who teaches students to write research papers at Detroit's Cass Tech, asked the bot to write an essay on "a time when you faced adversity." GPT-3 wrote about a fictional bout with cancer with left it "grateful for the experience,""stronger and more resilient" and "more appreciative of the little things in life." Taylor thought the essay was "grandiose" -- and phony.
Could students use the technology to cheat? Yes, said teachers. It would be hard to distinguish the tech's flat writing from real students' flat writing.
One old-fashioned way to combat this would be to require students to first hand in their outline and rough draft, and making those be hand-written.
I've seen some of the writing coming out of these latest tools and it is remarkable. I wouldn't class it as great literature, but it shows a real grasp of the given subject matter. Until someone develops an AI that can identify the work of other AI's (is such even theoretically possible?), teachers are going to have to find creative ways to stop cheating.
Perhaps the fact that AI's can do this means that teachers need to assign more creative tests of student ability than the typical essay. For example, "write a Socratic dialog that explains the most recent thinking on <<Topic>> in contrast to historical thinking" or "Describe the way you learned to think productively about <<Topic>>".