AI can help us work faster, but it may weaken persistence

Generative AI can speed up research, writing, coding, and everyday decisions. But a growing debate asks whether that convenience comes with a human cost: less patience with difficult work, weaker subject knowledge, and less confidence in independent judgment.

A recent study involving 1,222 online participants adds experimental evidence to those concerns. Participants who used ChatGPT for just a short period performed worse and gave up more often after researchers removed the tool. The result appeared across basic fraction problems and an SAT-style reading task. People who had worked without AI remained more likely to persist and solve the later questions.

The study does not show that every use of AI damages cognitive ability. It does, however, point to a risk of depending on instant answers before a person has built enough understanding to assess them. Researchers describe the missing element as “productive struggle”: the effort involved in working through uncertainty, testing ideas, and learning from mistakes.

That concern matters well beyond classrooms. Content professionals increasingly use AI to summarize reports, draft copy, analyze documents, generate concepts, and automate repetitive tasks. These uses can save time. Yet the quality of the outcome still depends on whether the person using the system understands the task, recognizes gaps, and can challenge an unreliable answer.

Assistance is not the same as understanding

The discussion is not simply about whether AI is good or bad. It is about the difference between using a system as support and handing over decisions that require judgment.

Huahai, writing on yyhh.org, argues that an LLM can still be a tool, even when it automates substantial parts of a task. A compiler and a spreadsheet also perform work automatically, he notes. What makes them useful is the human operator’s ability to provide inputs, inspect results, spot errors, and decide what happens next.

In that view, AI use can be active rather than passive. A worker might read an original document, ask an AI system to identify counterarguments, verify the output against sources, and revise a conclusion. That process can strengthen thinking rather than replace it. Simply accepting an AI-generated summary in place of reading, by contrast, may remove much of the learning.

The key question is therefore not whether a task involves AI. It is whether the user remains engaged with the reasoning behind the result.

Technical work shows why oversight still matters

A detailed analysis of AI-assisted software rewrites offers a practical example. Piotr Sarnacki examines versions of an application rewritten in Rust, Elixir, and Go with AI assistance. He finds that comparisons based on a single benchmark can hide important trade-offs involving compatibility, reliability, memory use, response times, and data delivery.

One version processed more requests but lost many user notifications under heavy load. Another delivered more notifications but experienced request timeouts and high memory consumption. A small configuration change improved one result while creating much longer delays. None of those choices has a universally correct answer. They depend on what a team considers acceptable failure behavior.

AI can generate code, but it cannot remove the need to define priorities. Someone still needs to ask whether speed matters more than completeness, whether a delay is preferable to a dropped message, and which risks users can tolerate.

Diomidis Spinellis takes the longer view. He argues that AI’s improving performance, falling cost, and constant availability will reduce the economic need for many forms of cognitive labor. Human thinking, he suggests, could increasingly become a voluntary activity, much as some physical skills became hobbies after industrialization.

That future remains contested. But the immediate lesson is clearer: AI can extend human capability when people maintain the skills to direct, evaluate, and question it. When it becomes a substitute for attention and understanding, the time saved may come at the expense of capabilities that are difficult to rebuild.

Sources

• Using AI for Just 10 Minutes Erodes Your Ability to Persist at Hard Things – Research UC Berkeley
• AI is a tool and using it well takes skill – yyhh.org
• Thinking will become a hobby – Diomidis Spinellis blog
• I’m sorry, but you still have to think – It’s all about the bit