Key facts
- AI could undermine faith in the scientific process, raising questions about novelty and accuracy.
- AI chatbots can hallucinate, and this is increasingly occurring in scientific research.
- Scientific journals are struggling to deal with AI-generated distortions.
- Google DeepMind's AlphaFold accurately predicted 200 million protein folds in one year.
- A DeepMind model claimed to have generated over two million new crystal structures.
- Nearly 90% of over a million biomedical research papers showed AI fingerprints by December 2023.
Concerns are mounting that artificial intelligence could undermine the scientific process, with recent incidents highlighting potential issues with data privacy, plagiarism, and the accuracy of AI-generated research. The integrity of scientific journals is also being challenged by a surge in AI-generated papers and citations. Dr. Oliver Vince of Basecamp Research believes AI will supercharge scientific research, but acknowledges the controversies surrounding its use. OpenAI's claim of solving a complex math problem and the subsequent concerns raised by mathematicians Tristan Buckmaster and Levent Alpöge about their research potentially feeding into OpenAI's models exemplify these issues. While OpenAI denies any influence, the incident underscores the growing problem of AI's role in scientific discovery. Google DeepMind's success with AlphaFold in predicting protein structures is hailed as a breakthrough, but other AI applications, like DeepMind's material discovery model, have faced scrutiny for potentially exaggerating findings. A study by Prof. Dmitry Kobak found that a significant portion of biomedical research papers show AI fingerprints, raising fears of widespread AI-generated 'slop' polluting scientific literature.