What We Are Giving Away to AI

When AI becomes the starting point for writing and decision-making, it can reshape how we understand ideas, trust our judgment and value our own work.

Illustration by The Epoch Times, Shutterstock

Before generative AI, Amazon built deep thinking directly into how work got done.

In 2017, Jeff Bezos described a meeting culture built around slow, written reasoning: no PowerPoint presentations, no bullet points, just a six-page narrative that often took a week to write, laying out a product or project from beginning to end.

For 20 to 30 minutes, everyone in the room—from junior staff to senior leaders—sat and read before anyone spoke.

The goal, Bezos said, was to force clarity through writing and reading. Weak reasoning surfaced immediately. Gaps could not be hidden in bullets or polished slides. Thinking took time, and that was the point.

The danger in generative AI is that it flips that logic with a single prompt—and in a way that makes your own thinking feel unnecessary.

A Growing Deference to AI

Today, however, across much of corporate America—including some departments at Amazon—employees increasingly face pressure to use AI tools to draft memos, proposals, reports, and even code. The incentives are clear: higher productivity, lower costs, faster output.

But with speed come tradeoffs.

Work that once sharpened thinking in the process is now being offloaded to AI, and with that offloading comes a loss of agency.

Related Stories

The Cognitive Debt We Accumulate Every Time We Use AI

When Smartphones Get Smarter, Do We Get Dumber?

Prompting can feel like you’re the author. However, much of the structure—the framing, the options, even the logic—has already been set by the AI system. The user can steer, but within boundaries they didn’t define, which gradually shifts agency from the user to the AI model.

“You think you have agency with prompting,” Zhivar Sourati, a computer scientist at the University of Southern California who studies how large language models (LLMs) influence reasoning, told The Epoch Times, “but compared to two years ago, you have way less.”

Two years ago, AI was just something you prompted; today, it has become something that prompts you back, and that shift can leave people less confident in their own ideas.

“When AI does the work, we feel less ownership over the output,” psychologist Michael Inzlicht, a professor at the University of Toronto who investigates how technology changes motivation, self-control, and effort, told The Epoch Times in an email. “The work isn’t really ours, and we know it.”

A growing deference to AI is occurring in the workplace. Research by Carnegie Mellon University and Microsoft has found that knowledge workers who trusted AI outputs often failed to scrutinize them closely—a tendency known as automation bias.

This kind of deference becomes risky because, although AI often produces convincing answers, those answers can be wrong, incomplete, or stripped of important context.

Some lawyers learned this lesson the hard way after filing court briefs that cited AI-fabricated cases that did not actually exist.

The Illusion of Consensus

Reliance on AI tools becomes harder to question when AI outputs themselves begin to look alike.

A dozen colleagues might prompt the same AI system, and receive similarly structured drafts with language clear, neutral, and well-formed. Ideas may begin to converge—often before independent judgment fully forms. Agreement begins to feel natural, even when it is not.

“Now people just talk to the AI, and they get ideas,” Sourati said.

In practice, using AI compresses the early stages of thinking. A manager drafting a strategy memo may start with a similar set of recommendations as others using the same tools. A doctoral student who once mapped research gaps manually can now ask a model to generate them instantly—and so can everyone else in their field.

The effect extends beyond technical work. AI tools are increasingly used in personal writing—breakup notes, wedding vows, even autobiographies—forms of writing once tightly tied to individual experience and voice. At the scale of a billion weekly users, even small shifts in how ideas get started can quickly homogenize, Sourati said.

Pressures to conform in language and thinking long predate AI. What AI changes is where convergence enters the process—not at the end of thinking, but increasingly at the beginning.

That matters, Sourati said in an article published in Trends in Cognitive Sciences, because people bring diverse ways of writing and reasoning into their work. “When those differences are filtered through the same LLMs,” he told The Epoch Times, “their distinct linguistic styles, perspectives, and reasoning strategies become homogenized.”

AI does more than simply make us sound similar; the concern is that it can narrow the boundaries of acceptable debate. When language is smoothed out and choices are repeatedly framed the same way, disagreement or dissent can feel less supported than it really is.

Over time, Sourati suggested, repeated framing can produce an “illusion of consensus”: a sense that people are independently arriving at similar conclusions or solutions to problems when, in fact, the same machine-generated patterns are smoothing distinct ideas, perspectives, and even critical stances into something that feels more unified than it is.

Even in his own writing, he notices the effect. AI smoothing makes everything more conventional—and less distinctly his. “I suddenly see that … this is not me anymore.”

When agreement emerges from shared starting points, it feels more organic than it is. In that sense, AI reverses what systems such as Amazon’s memo process were designed to do—where writing and careful reading helped people develop their own perspective before encountering others’, so agreement had to be earned.

What changes is not just what people think, but how AI can subtly shape the thoughts they arrive at.

Less Friction, Less Understanding

Growing reliance on AI—and the sense of agreement that comes with it—doesn’t emerge in isolation. It is reinforced by something harder to see: People often think they understand more than they actually do.

When people use AI, they often mistake the tool’s fast and fluent answers for their own.

An experiment by Anthropic, an AI safety and research company, helps explain why. Participants who relied on AI assistance scored about 17 percent lower on a comprehension quiz covering material they had worked with only minutes earlier. The task was finished, but understanding lagged.

“Deep understanding might require friction—some degree of struggle to work things through,” Inzlicht said. “AI bypasses [the struggle] entirely, so you end up with a polished product and a person who doesn’t really grasp what’s in it.”

For Inzlicht, the problem begins when friction disappears. Thinking involves false starts, uncertainty, and the slow construction of ideas. In an article published in Communications Psychology, he and his co-authors suggest that the mental effort involved in that process helps people learn and develop skills.

“When we struggle to understand something, we’re forced to connect it to what we already know; that connection is what makes it stick,” Inzlicht said. “Think back to high school math. Working through a problem yourself, however painful, taught you far more than seeing the solution first and reverse-engineering it. The wrestling is the point.”

AI, depending on how it is used, can short-circuit that process, encouraging cognitive offloading. Over time, that can weaken persistence and make people more likely to give up on hard problems once the AI is taken away.

Research has found that even 15 minutes of AI use can make people feel less focused and unable to push through difficult tasks on their own.

In that way, AI can shift the relationship between effort and knowledge: What starts as a convenience becomes a necessity.

Why Effort Matters

Effort does more than build understanding—it also builds meaning. When people struggle through a task, they tend to value the result more.

“We value our work less when AI helped produce it,” Inzlicht said. “It feels less meaningful, less ours, less worthwhile.”

“We even demand less money for it,” he said, pointing to AI-generated art and videos. “That’s telling. At some level we recognize that what AI made is not quite what we made.”

The benefits of friction are not unlimited; too much can be overwhelming.

The risk is not AI itself, but losing the kinds of effort that shape not just what we accomplish, but who we become.