In recent years we have come to know, and even use, artificial intelligence systems that write texts, create images and even songs based on the more or less precise instructions that we give them. But what would happen if AI started reading and interpreting truly complex texts for us? What impact would this have on our own lives? To begin with, we would face problems related to text comprehensionthe human work overloadhe recognition or loss of the author’s voice and even the idea formation process.
Thus, even an innocent and fun game (such as having an artificial intelligence agent summarize a classic literary work to make a good impression at a dinner party) runs the risk of becoming one of the main areas of debate around the new technologies. Naomi Susan Baron, professor emeritus of linguistics at the American University of Washington, deals with precisely this. Last November he gave a conference titled Who has read it? The impact of artificial intelligence that reads at the Learning More Festival in Modena.
WIRED met with her a few days before the talk and asked her to help us interpret the present and imagine future scenarios.
Federico Vergari: Professor Baron, let’s start with the topic of your talk at the festival: what does it mean to trust an artificial intelligence capable of understanding and interpreting complex texts?
Naomi Susan Baron: Although the brain is not a muscle, we have to train it so that it remains efficient. Reading complex texts or texts with linguistic nuances is a form of mental exercise. Abandoning AI means losing the opportunity to train the brain and giving up the possibility of interpreting a text by connecting it with other readings or experiences.
Can you give us a positive and a negative example of the use of AI?
On the positive side, AI is extremely efficient at sifting through mountains of more or less complex written information. So if we want a quick approximation, you can give us an excellent preliminary draft. On the downside, as all companies developing language models warn, the results produced by AI are not always accurate. If we want to trust it, we have to verify it.
Other negative repercussions?
If we constantly turn to AI to understand difficult texts, we lose the motivation and ability to do it ourselves. I used to ask my students: ‘What can you do when the internet goes down?’ Now my question is: ‘What are you able to read, write or research if you don’t have access to an LLM (large language model)?’.
We asked ChatGPT if humans will stop reading with AI. They answered no and that they will continue reading to experiment, form opinions and develop critical thinking skills. What do you think?
ChatGPT is formed to support human cognitive and creative abilities. I’m not surprised by this answer at all. What worries me more is the assumption (on the part of ChatGPT or anyone else) that people will continue reading for precisely those reasons. In many countries there is a steady decline in spontaneous reading among both children and adults. The 2018 PISA (Programme for International Student Assessment) test revealed that 49% read only when forced, compared to 36% ten years ago. And this data is prior to the arrival of AI. I can only imagine this statistic today.
In a quarter of a century we have abandoned paper to read on screens. Now we inform ourselves by doing scrolling and not browsing, we move through multiple levels of reading with hypertexts and we prefer summaries to long formats. It was a Copernican revolution, but we barely realized it. How do we read today?
Although digital reading has skyrocketed, we still read a lot on paper. A recent study by the American Library Association revealed that Generation Z reads more on paper than millennials. Perhaps the BookTok phenomenon is helping to spark interest in paper among younger generations, but, at least in the United States, print book sales continue to resist digital ones. Then there is the question of scrolling facing navigation; especially in a laptopwhere you have more visual space than on a cell phone, you can read an entire page at once and it is the best option. Research tells us that we learn more when we read digitally by turning pages and not by doing scrolling. When we see a whole page we read more slowly, we go back and use that greater space to fix the notions. Hypertexts are useful and often lead us to new ideas, but they distract us. Regarding the preference for summaries you are right, it is a growing trend, now amplified by the AI summary tools that are everywhere. Enter a long text in Google NotebookLM and a 10-minute summary or a podcast; Acrobat summarizes the PDF you wanted to save; Zoom summarizes both the meeting and the side chat.
#artificial #intelligence #read