NEXT: The Future of Reading
Reading is a very big deal in the state of Ohio…until Grade Three. State legislation says that students are required to be reading at grade level by the third grade, else there are dire consequences for schools, teachers and students. But then, after that benchmark has been reached, reading seems to disappear from everyone’s attention. There might be standardized tests for reading comprehension, but reading is ever so much more than just comprehension. Reading means not only discerning what a book is about as it is fathoming its meaning. Reading a text involves placing it in a wider context with other texts, seeing it in relation to and in conversation with other texts. As a cognitive activity, reading is much more than just comprehension of information.
The STEM disciplines get all of the attention in K-12 schooling and also in higher education. Majoring in a STEM discipline is a sure-fire path to a job. Reading, on the other hand, seems a concern for the elementary grades, not higher learning and certainly not for career preparation. Imagine the derision that would be leveled at the college student who announces that they are majoring in reading?
Reading is viewed almost entirely as a way to transmit and receive information. In so defining it in this way, it is easy to dismiss reading, for there are far more efficient (and more entertaining) ways to receive information. For generations, students have been avoiding reading by seeking out the movie versions of great literature, or searching for CliffsNotes or SparkNotes summaries. The NEA has a yearly survey on the status of reading in America, often pointing to a decline in reading for pleasure. Book sales, while not necessarily diminishing, are certainly not exploding either.
In a previous column, I talked about the idea of “conversations with Alexa,” that our interface with digital and artificial intelligence will increasingly come from the spoken word. Maybe in the future we will still “read” books, but this will not mean scanning with our eyes over a page but, rather, listening with our ears. We will be more likely to say, “Alexa, read me A Tale of Two Cities” as we will hold a physical copy in our hands and glide our eyes over the words.
In Richard Powers’ 1995 novel Galetea 2.2, a computer scientist wagers an English professor that he can develop a computer program that can interpret the Western canon of literature. Without spoiling the book, I should note that the computer scientist succeeds. And when the artificial intelligence develops the capacity to interpret literature, it becomes sentient. The moral of the story is unmistakable: when AI reaches the point at which it can read literature—not just comprehend it, but interpret its meaning—then it will have achieved true intelligence, on par with that of humans.
But that novel is science fiction: machines cannot interpret literature. Another way of saying this is that reading is a uniquely human ability that machines have yet to master, if they ever will. Another conclusion one could draw is that reading is a complex cognitive activity, more complex than playing chess or doing taxes or analyzing medical images or any of the other tasks that AI currently engages in.
That is why I am interested in work being carried out by computer scientists developing AI that purports the ability to read. Algorithms are already “reading” patterns in very large data sets like library collections or Google Books. But identifying word frequencies or keywords or topics is not the same as human reading. Some work in this area is now geared toward developing in AI the capacity for “gisting,” that is, determining the main idea or the most important points in a text.
Computer scientists at MIT have been working on an AI algorithm called Genesis. Genesis has been programed with the ability to read text, to draw inferences, and, in the estimation of the researchers, to understand a story. By “understand,” they mean more than simply summarizing it or even discerning the “gist.”
According to the researchers, Genesis is able to “identify themes and concepts that aren’t explicitly stated in the text of the story.” The instance described here is Genesis’ reading of Shakespeare’s Macbeth. When Genesis reads the text and is asked what it thinks the play is about, the AI replies “Pyrrhic victory and revenge.” Nowhere in the text is the word pyrrhic used, although the concept of a victory won at too high a cost is most certainly a theme of the play. Does this mean that Genesis is capable of reading in the same way as a human intellect?
When we look deeper into what Genesis is actually doing, however, we might want to hold off on claiming that the AI is reading like a human. In the first place, Genesis was not reading Macbeth, but a simplified version of Shakespeare’s play. The researchers fed Genesis a version of the text that was written in modern rather than Elizabethan English. Quotations and metaphors were excised from this version. Indeed, the story of Macbeth read by Genesis was only about 100 sentences in length “and included only the character types and the sequence of events.” Genesis was not so much reading Macbeth as it was a SparkNotes version of the play.
Even when considering this simplified version, arriving at the conclusion that the play is about revenge is a notable—if disquieting—achievement. At the same time, knowing what we know about the “head start” the algorithm had, the accomplishment does not seem quite as breathtaking. But the shortcomings of Genesis are nonetheless significant, if for no other reason in that it shows just how cognitively complex an activity reading actually is that the AI needs to have so many guardrails in place.
There are three scenarios for the future of reading that follow: 1). Owing to greater oral communication and the fact that machines have now learned to engage in the practice, reading declines as a cognitive practice that humans carry out. 2). AI and humans both engage in reading together. AI is invited into our humanities seminar rooms and in our monthly book clubs as a reading partner. 3). Human reading flourishes because it is understood as a complex cognitive activity on par with science and mathematics that no AI is able to mimic.