AI Is Quietly Rewriting How We Think, Work, and Do Science — and 1,222 People Just Showed Why It Matters

A growing body of research is raising uncomfortable questions about what happens when we hand more of our thinking to artificial intelligence (AI). From the way we solve problems to the way scientific studies get published, AI is changing not just what we do — but how we do it.

A study presented at the 2026 Conference on Language Modeling in San Francisco found that people who used AI chatbots became less willing to keep trying once the help was taken away. The research involved 1,222 participants across three randomized controlled experiments testing math problems and reading comprehension [3]. In the first experiment, 354 people were split into two groups — one could use a chatbot for 12 math exercises, the other worked alone. When researchers removed the AI and gave everyone three more problems, the group that had used the chatbot completed more tasks at first, but their performance dropped when working alone. They left more questions blank [3].

The team repeated the test with 667 more participants and found the same results. A third experiment with 201 people showed the same pattern in reading comprehension tasks, suggesting the loss of perseverance isn't limited to numbers [3]. The drop was most noticeable among users who asked for direct answers rather than guided hints. Researchers propose developing chatbots that act more like tutors — offering hints and guidance instead of direct solutions — so users complete the reasoning themselves [3].

The findings come as AI takes on a bigger role in how science itself is evaluated. Many journals now use AI tools to help judge whether a study is good enough to publish. Supporters say this saves time and reduces human bias, but critics worry that AI models learn from past research and may favor familiar ideas, rejecting new or unusual work [4]. That could push scientists to play it safe, avoiding risky topics and choosing projects that look like earlier studies. Over time, science could become less creative and slower to produce breakthroughs [4].

The stakes extend beyond the lab. Scientists are also testing whether they can reliably detect if a novel was written by AI, and early results suggest the answer is not simple. AI tools can now produce long stretches of text that read like human writing, making it harder for publishers, teachers, and readers to know who — or what — wrote a book [2]. Researchers look for patterns like repeated phrases, flat emotional tone, and small errors in logic, but such signals are inconsistent. Human writers can show the same features, and AI can be trained to avoid them [2]. Publishing deals, academic integrity, and copyright all depend on knowing the source of a work, yet current detection methods remain imperfect and easy to challenge [2].

Tech companies are trying another approach: adding hidden watermarks to AI-generated text and images. But researchers warn that the public cannot yet check these marks independently. Companies control both the pattern and the detector, creating a verification gap — the public must trust the company's own test [8]. Experts say watermarking should support, not replace, media literacy and open reporting, and that verification must be made public for the mark to mean anything [8].

Meanwhile, the pressure on human skills is growing. Biologist Paul Knoepfler has raised a critical question: could relying too heavily on AI weaken researchers' own abilities? As AI tools become more common in biomedical science, Knoepfler asks whether they help scientists or make them less capable [7]. And as AI increasingly handles tasks on its own — shopping, booking vacations, managing paperwork — the shift raises important questions for industries, companies, and jobs [6].

For individuals, the temptation to let AI finish our thoughts is real. But as one associate professor of public administration put it, thinking does not work that way. When we put words on paper ourselves, we see things differently — the act of writing forces us to examine our ideas more carefully. It was definitely harder before, and that is exactly why it was more rewarding [9].