From Page Turning to Quick Scanning: The Impact of AI Summarization on Society’s Approach to Information

To go back to this article, head over to My Profile, and then click on View saved stories.

The Remarkable Contribution of AI: Succinct Summaries?

A single expression perfectly captures the investigative approach of renowned nonfiction writer Robert Caro: Leave No Page Unturned. This expression has become so synonymous with Caro that it was chosen as the title for both a recent documentary about his life and an exhibition showcasing his work at the New York Historical Society. For Caro, it's absolutely essential to scrutinize every word on every page of documentation related to his research topic, regardless of how tedious or inconvenient it may seem. He has discovered that even the most seemingly insignificant detail can lead to a breakthrough in understanding an event, direct him to previously undiscovered resources, or solve the puzzle of who was behind a particular crisis or success. Throughout his illustrious career, Caro has examined millions of pages of various documents: studies, transcripts, essays, legal documents, and correspondences (with a staggering 45 million found in the LBJ Presidential Library alone!). Many of these documents appeared to be monotonous, repetitious, or unrelated at first glance. Yet, Caro would diligently sift through them, giving each his undivided attention. This thorough and meticulous approach has cemented Caro's status as a legendary figure in his field.

In today's era dominated by artificial intelligence, a fresh mantra has emerged: Forget about flipping through pages! This includes not bothering with the transcripts from your interviews. Additionally, there's no longer a need to focus during meetings, or even show up for them. You can also skip going through your emails or the memos from your co-workers. Simply input the unprocessed information into a sophisticated language model, and you'll quickly receive a condensed version to review. Thanks to advancements from OpenAI's ChatGPT, Google's Gemini, and Anthropic's Claude, skimming through summaries now passes for being well-prepared.

Large Language Models (LLMs) are programmed to condense information, a task their developers have eagerly assigned to them. Google has introduced an "auto-summarize" feature for documents, enabling users to swiftly sift through important details and decide where to direct their attention next. This summarization capability extends to Google Chat, where it can provide overviews of unread messages. Microsoft Copilot takes things a step further by offering instant summaries for various documents, including Excel sheets, PDFs, Word documents, and PowerPoint slides, simply by hovering your mouse over them. Even the already brief bullet points in presentations can be distilled further into their core messages. Similarly, Meta has begun to aggregate the comments under popular posts for quick insights. Zoom now offers meeting summaries, producing a handy reference on the fly. Additionally, transcription services like Otter emphasize summaries, relegating the full transcription to a secondary position.

What explains the surge in summarization? Just as we're starting to understand how to derive benefits from Large Language Models (LLMs), summaries emerge as one of the most direct and practically beneficial functionalities they offer. Indeed, these summaries might include inaccuracies or overlook crucial aspects. Acknowledged. However, a greater concern is the potential for excessive reliance on summaries to diminish our intelligence.

In essence, summaries serve as rough outlines rather than a comprehensive exploration of the topic. This brings to mind a humorous anecdote from Woody Allen, who claimed to have breezed through "War and Peace" in 20 minutes, summarizing it simply as “It’s about Russia.” This isn't to suggest that AI-generated summaries are overly simplistic. On the contrary, their effectiveness lies in their ability to provide a convincing facade of knowledge, enabling one to navigate a topic superficially without truly grasping its complexities.

For instance, consider how Otter creates summaries of voice recordings using AI technology. From a journalist's perspective, relying solely on these automated transcriptions means missing out on important nuances. Transcribing interviews oneself is undoubtedly a laborious task. However, it's during this detailed process that you truly grasp the essence of your interviewee's message, including what's left unsaid. Often, it's this meticulous examination that reveals overlooked insights. While a thorough review of a transcript can partially compensate for this, relying on summarized content may lead you to focus only on the most relevant sections, potentially overlooking valuable information hidden within the dialogue.

Effective leaders have always been aware of the pitfalls of taking easy routes. This is evident in Jeff Bezos' approach during his tenure as Amazon's CEO, where he prohibited the use of PowerPoint presentations in meetings. Instead, he insisted that his team members draft detailed memos, famously known as “6-pagers.” This requirement made it necessary for managers to thoroughly consider their proposals, understanding that each word played a crucial role in the success or failure of their presentation. Meetings under Bezos would commence with a period of silence, during which everyone would carefully read through all six pages of the memo, with no room for brief summaries.

Admittedly, there's a valid opposing perspective to my unease regarding summaries. Effortlessly, an LLM manages to review each page in detail. Thus, if one desires to delve deeper than just the summary and provides the right prompts, an LLM is capable of swiftly finding even the most hidden information. It's conceivable that in the future, these systems could advance to the point where they're adept at uncovering and bringing to light specific insights tailored to our queries. However, this would likely lead to increased dependence on these models, potentially causing our own skills to diminish.

Over time, the obsession with summaries could potentially undermine the very essence of writing. If it becomes common knowledge that the full content of your emails, documents, or reports won't be read, what's the incentive to invest effort into uncovering engaging details or crafting sentences that showcase your cleverness? It might seem more practical to delegate your writing tasks to artificial intelligence, which won't object to producing lengthy reports. After all, no one is likely to raise concerns, as they'll probably be employing their own AI tools to distill the essence of these reports into a few key points. Should this trend continue, the collective output of our society might end up resembling the degraded quality of a document that's been photocopied multiple times.

Regarding Robert Caro, he is significantly overdue on delivering the fifth installment of his monumental series on LBJ. Had legal research tools been available when he started chronicling the president's narrative nearly half a century ago—and had he opted to utilize them instead of conducting extensive manual research—the series might have been finished by now. However, it likely wouldn't have reached the same level of excellence.

Time Travel

At the beginning of this year, I engaged in a dialogue with Sam Liang, the head of Otter. Initially focused on direct transcription services, the firm has expanded its offerings to include a variety of AI-powered tools for meetings, which notably encompasses summary generation and more innovative functionalities such as AI avatars capable of participating in and leading discussions. In my editorial, "Plain View," I explored the idea of whether these advancements could undermine the fundamental objectives of holding meetings.

I inquire with Liang if the increasing use of AI in meetings could potentially discourage people from participating in person. The availability of a meeting summary might reduce the motivation to attend in the first place. Liang admits that he only attends a small portion of the meetings to which he's invited. "Being the CEO of a startup, I receive a plethora of meeting invitations—often finding myself double or even triple booked," he explains. "With Otter, I can evaluate my invitations and prioritize them. I sort them by their content, how urgent and important they are, and whether my being there would contribute any additional value." As the CEO, he likely has more flexibility in choosing which meetings to skip. However, when the boss is present at a meeting, it can be more meaningful for those looking to understand his perspective or hoping to get immediate approval on a proposal.

Naturally, the underlying belief of organizing meetings is that the presence of each participant contributes added value. It completely misses the mark if, when it's time for the one individual with the necessary insight to speak up, they're absent. However, Liang proposes an artificial intelligence-based solution to this issue. He mentions, "Our team is in the process of developing a tool named Otter Avatar, designed to create a personalized model for any team member who cannot attend a meeting due to illness, vacation, or personal choice. This model will be trained using the individual's historical data, such as previous meeting contributions or Slack conversations. Consequently, if there's a need to ask that person a question, their digital counterpart could provide the answer on their behalf."

I emphasize the potential for a competition in artificial intelligence development. "I plan to have my digital representative attend all gatherings, as will others," I say. Thus, meetings will merely consist of digital personas conversing, with individuals later reviewing the recap to understand the discussions held by the AIs.

"Liang acknowledges, “It’s possible. Naturally, there will be circumstances where having a direct personal connection is preferred.”

Inquiry of the Day

Judith inquires, "A lot of autobiographies from individuals in STEM fields recount early memories of tinkering with transistors, radios, and various science kits. Is the current generation deprived of such hands-on learning opportunities?

Thank you for your inquiry, Judith. It's likely accurate to say that the golden era of child-focused science experiments has somewhat dwindled. A contributing factor is our enhanced focus on safety and legal precautions. Historically, children's chemistry sets contained hazardous materials, such as lethal sodium cyanide and radioactive uranium ore. As a result, there's been a shift towards safer, albeit less captivating, alternatives.

However, it's not primarily safety concerns that deter children from engaging in practical science experiments. In an era where computers are ubiquitous, many young learners opt for virtual simulations to satisfy their scientific curiosity. Yet, I believe that children with an insatiable curiosity about how things work manage to find their way to rockets, telescopes, and even intriguing chemicals. The Richard Feynmans of the future will likely turn to the internet, local scrapyards, or perhaps clubs and supportive science educators to acquire the equipment they desire. For those whose interest isn't as instinctive, immersive games like Minecraft and Civilization, along with programming, serve as excellent introductions to the scientific realm. In essence, those with a passion for science will always find a way to indulge it.

Send your inquiries to mail@wired.com. Ensure you include ASK LEVY in the email subject.

Apocalyptic Gazette

Previously: The peaks of North Carolina were untouched by the tempests tormenting the state's east. Now: The western region of North Carolina has become the latest hotspot for hurricanes.

Final Thoughts

This is an overview of the impact AI has on global electoral processes.

Bobbi Althoff rose to prominence and wealth through her journey as a mommy blogger, a story she shares in our exclusive interview.

With the decline in individuals registering for genetic testing, the market value of 23andMe is nearing a low point.

License plate scanners are capturing information from bumper stickers, impacting privacy significantly.

Ensure you don't miss out on exclusive editions of this column reserved for subscribers. Take advantage of a special 50% discount for Plaintext readers and subscribe to WIRED today.

Additional Content from WIRED

Evaluations and Tutorials

© 2024 Condé Nast. All rights reserved. Purchases made through our site on products linked to our affiliate retail partners may generate revenue for WIRED. Replicating, sharing, broadcasting, storing, or using the content found on this site in any form is strictly prohibited without prior written consent from Condé Nast. Advertisement Options.

Choose a global website