What Happens When AI Can Write Better Than We Can?

What Happens When AI Can Write Better Than We Can?

AI experts believe that computers will write as well as humans within the next 15 years. This means that any student will be able to input a poorly-written essay into a software program, which will analyze the text and reconstruct it as well-written, grammatically correct text. Since we use calculators as an extension of our minds, shouldn’t we also use AI software to become better writers?

This is not a hypothetical question. Across the world, teams of computer scientists are racing at a breakneck speed to construct advanced artificial intelligence that can automate thinking and writing. Last month, AlphaGo, the artificial intelligence program created by Google, beat the world-champion Lee Sodel in Go, a game that is so complex that there are more choices available in a single game than there are atoms in the entire universe. Until only a couple of months ago, researchers thought that Go was so complicated that it would be another ten years before an AI could defeat a world champion. However, AlphaGo exists now, and it’s able to assess problems and create complex solutions. In the future, AI software will be able to analyze student work and construct better text.

As a community, we now face a critical question: How should artificial intelligence be used to enhance education?

I believe that artificial intelligence should be used to enable students to ask questions and articulate thinking, but it should not be used to do the thinking for students, much like how calculators are used today. We should determine the educational value of an artificial intelligence program by the complexity of thinking it enables in learners.

In writing education, my area of focus, the first wave of AI writing tools is appearing. These products, such as WriteLab, flag poorly constructed sentences and offer suggestions for better versions of a text. Once a student inputs the suggestion, the suggestion disappears and the student moves on. These products don’t necessarily teach students how to write, they simply tell students how to fix bad writing.

Matthew Ramirez, the CEO of WriteLab, states, “[WriteLab] will give you your own thoughts back to you, but it will give you a clearer, more concise version of your own thoughts... We are taking your prose and re-constructing it to make it more effective.” By automatically fixing the student’s writing, it will appear as if the student is a good writer. However, the artificial intelligence technology is masking the student’s poor writing skills.

If we look at how spellcheck impacts students' writing, we can see how software can hamper learning. Multiple studies have found that students make more writing errors when they rely on spellcheck. Studies conducted in 2003 and 2005 asked students to proofread a passage either with or without spellcheck. Students found twice as many errors when spellcheck was turned off, as the students were required to pay more careful attention to their work. With spell check we no longer have to think about the spelling of words.

Students’ stunt their own abilities when they do not consider the spelling of words. Catherine Snow et al. write, “Spelling and reading build and rely on the same mental representation of a word. Knowing the spelling of a word makes the representation of it sturdy and accessible for fluent reading." Knowledge to Support the Teaching of Reading: Preparing Teachers for a Changing World. San Francisco: Jossey-Bass). Imagine a 3rd grader typing a paragraph on his laptop—every time he spells a word wrong, he uses auto-correct to fix the word. While it seems as if the student knows how to spell words correctly, the student is not internalizing the correct spelling. Likewise, if we automatically fix student writing, students will no longer have to think about their work, limiting the development of their writing skills.

I’m not suggesting that spell check should be abandoned. Instead, imagine spelling a word incorrectly and then being taught how to spell that word. Instead of seeing the red squiggly line, you could listen to a pronunciation of it, say it out loud yourself, and use it properly in a sentence. By creating these types of learning experiences, we are creating thinking rather than automating thinking.

Another way in which we can create more thinking is by inviting students to ask questions. In these exercises, students break down complex problems into questions, and AI assess the quality of these questions. Timothy Shanahan, a previous Director of Reading at the Chicago Public Schools, presents on his blog an example of how this thinking process works. Students are provided with a complex, 44 word sentence and have to unpack the meaning of the sentence by splitting the sentence into a series of questions:

“The women of Montgomery, both young and older, would come in with their fancy holiday dresses that needed adjustments or their Sunday suits and blouses that needed just a touch—a flower or some velvet trimming or something to make the ladies look festive.”

Students can then unpack this sentence into the following questions:

Which women? The women of Montgomery.

How old? Both Young and Old.

Which dresses? Their fancy holiday dresses.

Which suits or blouses? Suits or blouses that needed just a touch-a flower or some velvet trimming.

Having students unpack complicated sentences is a cognitively demanding thinking task, and students need these thinking skills. According a 2011 Department of Education study, 76 percent of eighth grade students are not proficient writers, and these students struggle to articulate complicated thoughts in writing. We will soon enter a world in which students will be able to rely upon computer programs to synthesize their thoughts into writing. By automatically improving student writing, it will appear as if students are becoming better writers. However, by judging the final answer rather than the thinking process, we may end up depriving students of the chance to develop their own thinking.

Many thinking processes are now becoming automated. Twenty years ago, every person needed to know how to read a map. Now, only 20 percent of young people know how to read one. With Google Maps and Apple Maps leading the way, we no longer think through maps. Likewise, self-driving cars are now driving through Europe and the United States, and, according to a recent New York Times article, they may be commonplace within the next ten years. It's reasonable to expect that some children born today won't know how to drive a car. This isn’t to say this is a bad thing - 200 years ago, 90 percent lived and worked on a farm, and most of us do not think we are worse off for not knowing agriculture. However, there is a critical difference between the thinking skills we need as workers and the thinking skills we need as citizens - with writing, we express our voice as citizens, and we will always need to posses this skill. As we enter into a world where thinking becomes automated, we must define which thinking skills we should always hold on too, even if a computer could automate these skills.

Here’s a call to arms to create more writing and thinking. What’s an example of an experience where you’ve experienced or created a meaningful thinking experience? Write your examples below, or email me at whatisthinkingblog@gmail.com. I’ll post all of the stories to the public blog whatisthinking.com and share access with you so that you can post as well with your own Medium account. By articulating what thinking is, we enable students to read deeply and think critically about the world around them.

Peter Gault (@petergault3) is a cofounder and the Executive Director of Quill.org. He wishes to thank Peg Tyre and Elliot Mandel for reading drafts of this article.

GET THE LATEST HIGHER ED NEWS.Be the first to know, with our weekly newsletter.