Springer Nature retracts ~40 publications that trained ANNs on 'bonkers' dataset (www.thetransmitter.org)

Springer Nature has initiated a retraction of nearly 40 publications that used a controversial dataset aimed at training neural networks to differentiate between autistic and non-autistic children. The dataset, created by a retired engineer using publicly available images, raises serious ethical and reliability concerns. It lacks proper consent from the subjects, with images collected from various online platforms, making it nearly impossible to validate the claimed diagnoses. Following an internal review sparked by inquiries into related papers, Springer concluded that the dataset's methodological flaws fundamentally undermine the validity of the associated research. This incident is significant for the AI/ML community as it highlights pressing issues surrounding data ethics and the implications of using unverified datasets in machine learning applications. Experts like Dorothy Bishop and Gail Alvares stress that the use of such unreliable datasets not only jeopardizes the integrity of AI research but also sets a dangerous precedent for future studies, especially in sensitive areas like autism diagnosis, which should rely on clinically approved methods. The retraction serves as a stark reminder of the necessity for stringent ethical standards and rigorous validation processes in AI/ML research, particularly when human subjects are involved.