The Rise of Generative AI in Scientific Research and the Risks to Institutions

Introduction

The use of large language models (LLMs) in academic work has caused a tension between the desire for faster work and the need to maintain scientific honesty.

Main Body

There is currently a big difference in how researchers use generative AI. For example, an Elsevier survey showed that usage rose from 37% to 58% in one year. However, a Nature survey suggests that most researchers only use AI to improve their language, while very few use it to write the main text. Some scholars avoid these tools entirely because they want to develop their own thinking skills and avoid ethical problems regarding where the data comes from. Institutional stability is also threatened by 'hallucinations,' which are factual errors made by AI. In chemistry and conservation science, AI often creates fake molecular structures and wrong citations. Consequently, humans must spend a lot of time checking the work, which removes the time-saving benefits of the technology. Furthermore, these tools have a high environmental cost; it is estimated that by 2025, they will produce millions of tonnes of CO2 and use vast amounts of water, which contradicts the goals of climate research. Finally, there is a growing conflict over how to detect AI-written content. Submissions to 'Organization Science' increased by 42% after ChatGPT was released, and many of these papers contained over 70% AI-generated text. Similarly, AI-generated reviews in computer science rose from 7% in 2023 to 43% in 2025. Because current detection tools cannot always tell the difference between AI editing and full AI generation, fake data may enter the official scientific record.

Conclusion

The scientific community remains divided as it tries to balance faster research production with the need for high quality and ethical standards.

Learning

The 'Cause and Effect' Upgrade

An A2 student usually says: "AI makes mistakes. So, humans check the work."

A B2 student uses connecting adverbs to show a logical flow. This makes your writing sound academic and professional.

The Power Word: Consequently In the text, we see: "AI often creates fake molecular structures... Consequently, humans must spend a lot of time checking the work."

Why this is a B2 move: Instead of using "so" (which is very basic), Consequently signals that the second sentence is a direct, logical result of the first. It transforms a simple observation into a formal argument.


Expanding Your Logical Toolkit

To reach B2, you need to vary how you connect ideas. Look at these shifts based on the article's themes:

A2 Style (Basic)B2 Style (Advanced)Transition Word
AI is fast, but it is risky.AI offers speed; however, it introduces significant risks.However
AI uses water. It also produces CO2.AI uses vast amounts of water; furthermore, it produces millions of tonnes of CO2.Furthermore
AI creates fake data. This is why it's dangerous.AI creates fake data; therefore, the scientific record is threatened.Therefore

Quick Guide to Usage

  1. However: Use this when you want to pivot to an opposite idea.
  2. Furthermore: Use this when you are adding a "second layer" of evidence to your point.
  3. Consequently / Therefore: Use these when the second fact is a result of the first.

Pro Tip: Notice that these words are often followed by a comma (,) when they start a new sentence. This is a key marker of B2-level punctuation.

Vocabulary Learning

tension (n.)
the state of being under pressure or stress
Example:The tension in the lab grew as the deadline approached.
desire (n.)
a strong feeling of wanting something
Example:Her desire to publish quickly led her to skip some checks.
maintain (v.)
to keep something in a particular state
Example:Researchers must maintain rigorous standards to ensure validity.
difference (n.)
the way in which two or more things are not the same
Example:The difference between the two models was subtle but significant.
survey (n.)
a systematic investigation to gather information
Example:The survey revealed a rise in AI usage among scholars.
usage (n.)
the act of using something
Example:The usage of large language models is increasing rapidly.
improve (v.)
to make something better
Example:We need to improve the accuracy of the data before publishing.
ethical (adj.)
conforming to moral principles
Example:Ethical guidelines help prevent misuse of AI tools.
hallucinations (n.)
false perceptions or outputs produced by a machine
Example:The hallucinations produced by the model included nonsensical equations.
factual (adj.)
based on real facts, not imagined or false
Example:The report was praised for its factual accuracy.
molecular (adj.)
relating to molecules or the smallest units of a substance
Example:Molecular structures were incorrectly depicted in the simulation.
citations (n.)
references to sources of information
Example:Proper citations give credit to original authors.
environmental (adj.)
relating to the environment or surroundings
Example:The environmental impact of data centers is a growing concern.
conflict (n.)
a serious disagreement or argument
Example:A conflict over data ownership delayed the project.
balance (v.)
to keep something in a stable or equal state
Example:Scientists must balance speed with thoroughness in their work.