Does money lead to happiness?<\/p>\n
Researchers from a coalition of West Coast universities were curious how 100 human participants would respond to the age-old question, but not because of their own pursuit of happiness. Instead, the researchers wanted to know how participants\u2019 use of AI systems might sway their (written) answers. <\/p>\n
The research team found that users who heavily relied on large language models (LLMs) produced responses that diverged significantly in meaning from the answers of participants who only partially relied on LLMs or avoided their use altogether, suggesting heavy AI use alters the substance of humans\u2019 arguments in addition to changing writing style.<\/p>\n
\u201cThe LLMs are pushing the essays away from anything that a human would have ever written,\u201d said Natasha Jaques, one of the lead authors of the study<\/a> and a computer science professor at the University of Washington, highlighting the “blandification” of writing that relies on AI systems. \u201cThey just change human writing in a way that\u2019s very large and very unlike what humans would have done otherwise.\u201d<\/p>\n The new research, which has been peer-reviewed and accepted to an upcoming workshop<\/a> at a leading AI conference<\/a>, found that people who heavily relied on LLMs produced essays that answered the happiness question with a neutral response 69% more often than participants who did not use AI or only used AI for light edits. The study participants who used AI less often or avoided AI entirely submitted essays that were much more passionate, either positively or negatively, about the link between money and happiness. <\/p>\n In addition to AI\u2019s impact on the meaning of the essays, the researchers also found that heavy reliance on AI systems altered the overall style of users\u2019 outputs, causing their language to become less personal and more formal. <\/p>\n After the experiment, participants who heavily relied on AI reported that their essays were significantly less creative and less in their own voice. At the same time, these participants reported similar satisfaction rates with their final outputs compared to participants who used AI less, raising concerns from the authors and outside experts about the long-term impacts of humanity\u2019s increased use of AI systems.<\/p>\n \u201cThis research highlights that LLMs are not able to adhere to peoples\u2019 preferences and personalize how the human would have written the essay,\u201d said Jaques, who is also a senior research scientist at Google DeepMind, one of the world\u2019s leading AI companies. \u201cAn ideal LLM should write the essay that you would have written and just save you time.\u201d <\/p>\n \u201cIt\u2019s not doing that at all. It\u2019s writing a very different essay.\u201d <\/p>\n The study evaluated the impacts of three leading AI systems widely used in 2025: Claude 3.5 Haiku from Anthropic, GPT-5 Mini from OpenAI, and Gemini 2.5 Flash. In initial testing, the researchers found that half of the participants refused to use an LLM at all or only used it to find information rather than generate new content. To better categorize the larger batch of participants, the researchers defined heavy AI users as the participants who said they generated more than 40% of their text written for the experiment with an LLM.<\/p>\n The authors found that users who heavily relied on LLMs submitted essays with 50% fewer pronouns, which was representative of the larger shift toward impersonal language that included fewer anecdotes and references to human experiences. <\/p>\n In addition to the experiment regarding the impact of money on happiness, the new paper analyzes differences in how LLMs edited another set of essays compared to humans and examines how the use of AI affects the criteria scientists employ to judge whether papers should be accepted to leading AI conferences.<\/p>\n To compare how LLMs edit existing writing compared to humans, Jaques and her collaborators relied on a database of human-written essays from 2021 to evaluate writing that was published before the widespread adoption of LLMs. <\/p>\n Asking the LLMs to revise the human essays based on human feedback from the original human-written dataset, the study authors found that the three leading AI systems made much larger edits than human editors in the same situation, and that the AI-powered edits also changed the meaning of the underlying essays.<\/p>\n While human editors often made changes that substituted individual words and left most of the original vocabulary untouched, the LLMs \u201creplace a much larger fraction of the original writing than humans do when revising their own work,\u201d according to the paper.<\/p>\n \u201cThis substitution of words contributes to the loss of individual voice, style, and meaning, as the unique lexical fingerprint of each writer is overwritten by the given model\u2019s preferred vocabulary,\u201d the authors wrote.<\/p>\n