A Nobel prizewinning neuroscientist who lost two research papers over irregular images has rejected Donald Trump¡¯s claim that there is a ¡°reproducibility crisis¡± in US research, arguing the use of ¡°AI-powered software¡± on old images is unfairly tarnishing the reputation of scientists by erroneously flagging research misconduct.

In a passionate defence of his research, Thomas S¨¹dhof told an audience of young scientists at the annual Lindau Nobel Laureate Meeting that there is a grave danger of overstating the extent of research fraud within scientific research by accepting that any images flagged as dubious using new AI tools are always the result of research fakery.

His speech on 30 June comes after the US president issued an??on 23 May in which he stressed the need for ¡°gold standard research¡± in light of a ¡°reproducibility crisis¡±, flagging how the ¡°falsification of data by leading researchers has led to high-profile retractions of federally-funded research¡±.

Trump¡¯s science adviser Michael Kratsios has specifically highlighted retractions from Stanford University, where S¨¹dhof is based, as justification for?massive science cuts, citing the case of retraction of a 2009?Nature?paper that led to the resignation of the university¡¯s president Marc Tessier-Lavigne.

ÌÇÐÄVlog

S¨¹dhof, who won the Nobel Prize in Physiology or Medicine in 2013, said the presidential executive order has had ¡°a tremendous, largely negative effect on science in the US¡± but, on the question of the so-called reproducibility crisis, it is worth asking: ¡°Is this true?¡±

¡°While it is undoubtedly the case that science is facing a crisis ¨C and we, as scientists, are partly responsible for this crisis ¨C I would argue this crisis is not rampant fraud,¡± said S¨¹dhof, adding: ¡°It is not a reproducibility crisis, it is more of a structural crisis which is caused by the digital transformation of our lives including science.¡±

ÌÇÐÄVlog

¡°I don¡¯t believe there is a major issue with reproducibility [and] fraud,¡± he continued, stating some of the perception of the crisis related to a ¡°misunderstanding of the scientific process¡±.

S¨¹dhof explained this was ¡°particularly personal¡± because a?Nature?article last year had reported how his lab had been required to correct numerous mistakes flagged by then-independent research fraud sleuth Elisabeth Bik, now a faculty member at Stanford.

Namechecking also??an independent science blogger, among the ¡°profession¡± of research fraud sleuths, S¨¹dhof said: ¡°There is a group of people with the best intentions are following our work intensely, scrutinising tens of thousands of images published over the decades and identifying mistakes using AI-powered software that we overlooked.

¡°There were quite a few mistakes committed by many trainees, postdocs, who were obviously very disconcerted by this discovery ¨C in addition, the PubPeer pursuit, guided by these people, has identified many errors in the labs of our collaborators who became, more or less, collateral damage in the pursuit of our lab,¡± he continued, asking whether this example should be considered a case of ¡°widespread fraud¡±.

ÌÇÐÄVlog

¡°Or are we looking at things which are errors, artefacts or maybe non-existent that we have to deal with because we are in a new stage of research where AI-powered tools can identify anything, or make up anything,¡± he said.

In one case that led to a retraction of a S¨¹dhof paper, the error highlighted was a trivial one and made no difference to the evidence presented, he said.

After tracking the incidence of mistakes in his lab, S¨¹dhof found mistakes only began to arise when he began digitally processing images, suggesting some of these errors might be ¡°artefacts of digital processing and copy and paste errors¡± in papers that were ¡°10 or 20 years old¡±.

Source:?

Thomas Sudhof: Lindau Nobel Laureate Meeting

¡°The cost on junior scientists has been enormous and¡­this should be recognised,¡± he said.

ÌÇÐÄVlog

In another of his papers that was recently challenged for alleged ¡°false positives¡±, S¨¹dhof said there was a ¡°misunderstanding of our methods and approaches, and the approach we had taken was correct¡±.

Addressing the point that there were ¡°20,000 retractions in 2023 alone, including two from my own lab¡±, this did not necessarily mean there was a reproducibility crisis, he said.

ÌÇÐÄVlog

¡°The fact that people get different results doesn¡¯t mean the [original] results are unreproducible ¨C they just get different results,¡± said S¨¹dhof, who added that the scale of global scientific endeavour meant ¡°it is not possible to create science which is 100 per cent correct¡±.

jack.grove@timeshighereducation.com