The business of A.I. in medicine
We probably are biased in feeling that artificial intelligence (A.I.) in medicine is being oversold; but it seems to be everywhere, whether in general medicine or infertility, and it likely is everywhere in everything, and not only in medicine. It suddenly seems—and it may turn out to be true—that everything can be done through A.I. better, with less effort, quicker, and maybe also less costly.
Those are at least the claims, and here are some with medical relevance—and not necessarily in order of importance, as we currently view the claims:
Diagnosing autism
A large language model, for example in Cell, is alleged to deconstruct the clinical intuition behind diagnosing autism (1). And if you didn’t right away understand what this sentence means, neither did we on first try. So, here are some explanations: Genome-wide assays and brain scans to diagnose autism have not been very successful. Clinical intuition of healthcare professionals remains the gold standard for diagnosis. So, investigators leveraged deep learning to deconstruct and interrogate the logic of clinician intuition. After pre-training on hundreds of millions of general sentences, they finessed large language models (LLMs) on >4,000 free-form health records from healthcare professionals to distinguish confirmed versus suspected autism cases. Their extended language model architecture then could pin down the most salient single sentences in what drives clinical thinking toward correct diagnoses, flagging the most autism-critical criteria which were stereotyped repetitive behaviors, special interests, and perception-based behaviors. These findings challenged today’s focus on deficits in social interplay, offering better potential diagnostic criteria.
Adverse effect on quality of biomedical papers
Unsurprisingly—and often avoiding disclosure requirements of medical and science journals—A.I. found wide use in manuscript preparation. Miryam Naddaf in a News article in Nature magazine (2) now, based on a paper in PLoS Biology (3), notes that hundreds of studies appeared to follow template reporting correlations using public data sets. Associations then, however, often did not hold up to statistical scrutiny. Other studies appeared to have “cherry-picked” data.
A News article in Science magazine by Cathleen O’Grady, in turn, reports that as a consequence of public data use and A.I., low-quality papers produced by paper mills (mostly in China) “drive the industrialization of shoddy research” (4). After 2021, 92% of papers using publicly available the National Health and Nutrition Examination Survey data came from China; before 2021, the percentage was only 8% (4).
Related, a letter in Nature raised the suspicion that some peer reviewers are asking large language models to write “critical” reviews to be able “to reject” a paper (5). The writer suggested that journals must strengthen their policies regarding A.I.-generated texts and scrutinize texts for signs of unauthorized A.I. use.
A.I. designed at Google to help scientists and mathematicians to create new algorithms
Science magazine also reported that Google’s DeepMind released an A.I. agent called AlphaEvolve, meant to help with the development of new algorithms (6).
And if we already are talking about Google, The Wall Street Journal recently noticed in an article how threatening A.I. is to Google’s core business of searching (7).
Brain-computer interface in pre-clinical testing may boost human intelligence
This is at least the hope for new A.I.-powered brain-computer interfaces, according to a News article by Paul Webster in Nature Medicine (8), which can send and receive signals. Brains of study participants are already being trained.
Where do we call?
References
1. Stanley et al., Cell 2025;188(8):P2235-2248.E10
2. Nadaff M. Nature 2025;641:1080-1081
3. Suchak et al., PloS Biol 2025; 23(5): e3003152
4. O’Grady C. Science 2025; 388(6749)”807-808
5. Xames MD. Nature 2025;641:39.
6. Service RF. Science 2025; 388(6749):805
7. Gallagher D. Wall Street Journal. May 8, 2025. https://www.wsj.com/tech/ai/ais-threat-to-google-just-got-real-8280b4ee?gaa_at=eafs&gaa_n=ASWzDAjwrcuJ_q56VKJFuPC1e222xwbeD6wYuPGMZxJhy7b4iuFm3CwC5mgqhqe-B6c%3D&gaa_ts=68587b7e&gaa_sig=y8yJuTEbtDjitIcw0iDmfPuqly3PqRVAxceqa04ZJ_kPAfttvL2GcRIWNFRppmqIxX1XjiVAPm-3bDrtaT4B-g%3D%3D
8. Webster P. Nat Med 2025;31:1045-1047