To make accurate predictions, artificial intelligence needs a lot of data feed Learning. “Artificial intelligence (AI) is like a small child,” says medical physicist Thomas Beyer of MedUni Vienna and Spinoff Dedicaid. Until it can distinguish blue from white – or in this case a healthy one from a diseased prostate or lung – it is a long and complex process. Large amount of data And to do so, it needs an enormous amount of data from which it can learn similarities and differences. “It’s not really an artificial intelligence, but a copied human one that can only repeat banal things faster,” says the Viennese researcher. At the moment, such intelligences are only as clever as human beings are. Quality essential The goal is for artificial intelligence to be able to classify a tumor more accurately and in more detail than a human. “You have to have a high-quality and sufficient database for this,” says the doctor. But there are many hurdles. The quality of the examination images of the patients varies depending on the equipment used at the different facilities. “Data from our cohorts from Vienna doesn’t necessarily have to match the data from other centers, because other devices with different performance are used there,” says Beyer. Reproducibility Artificial intelligence must therefore be able to correctly evaluate the different data of the institutions. In order to achieve reliable and repeatable results, the forecast model will need to be thoroughly tested in the coming months.