A London facility called Basecamp Research is collecting billions of genes to build an artificial intelligence that understands biology, joining a global push to supercharge scientific research with AI. However, this technological wave is increasingly sparking controversies, ranging from intellectual property disputes to fears that AI could undermine trust in the scientific process itself.
Questions Over Discovery and Claims
Last month, OpenAI announced that its AI agents solved one of mathematics’ major unsolved problems regarding Navier–Stokes equations in just 88 hours, with the prize-running body noting the matter had “apparently been settled.” Hours prior to the announcement, NYU mathematician Tristan Buckmaster raised concerns after working on the same goal with Levent Alpöge via OpenAI’s coding tool, Codex, asking whether his stored research could have fed into OpenAI’s work. OpenAI stated that nobody at the company viewed the other team’s work beforehand, and later clarified that Buckmaster’s Codex prompts could not have influenced the system.
Beyond disputes, scientists warn that AI models can hallucinate or exaggerate findings. Google DeepMind famously predicted 200 million protein folds using AlphaFold, a breakthrough hailed by DeepMind’s Pushmeet Kohli as a “watershed moment” that earned the firm a Nobel Prize. However, when DeepMind later predicted millions of new crystal structures for materials like batteries and solar cells, two University of California chemists analyzed a random sample and concluded the results lacked novel compounds and were trivial adaptations of known materials, emphasizing the need for rigorous human validation. DeepMind rejected the criticism, arguing that swapping single atoms is how many useful materials are created.
The Rise of AI ‘Slop’ in Journals
More critically, scientific journals face pollution from AI-generated “slop,” where computers write research papers featuring fake citations and phantom studies. Ghent University’s Prof Dmitry Kobak analyzed over a million biomedical papers and found that nearly 90% bore AI fingerprints by late last year, risking the spread of model biases through scientific literature. Peer review is also affected, as Stanford University researchers found up to 17% of text in referee reports across four major computer science conferences was substantially modified by language models.
Furthermore, a Cornell University study revealed that researchers used AI to compile citations, leading to roughly 147,000 pieces of phantom research cited last year. Publishers are responding with integrity checks: Elsevier developed around 100 signals to flag suspicious papers, Nature banned AI-generated images and barred referees from uploading manuscripts into AI systems, and arXiv introduced year-long bans for submitting invented references. Elsevier vice-president Prof Sarah Main advocates for “research-grade AI” to keep human researchers firmly in control.
Despite these challenges, researchers stress that AI remains a vital tool when used responsibly with human accountability. Prof Andrea Pauli of the Vienna BioCenter used AI responsibly alongside a decade of lab work to understand how sperm enters zebrafish eggs. Meanwhile, Basecamp co-founder Oliver Vince hopes his firm’s biological AI will chart a technically feasible path toward curing cancers, leaving the ultimate question whether AI will advance science or quietly erode its foundational trust.
