Researchers led by a team from Emory University recently announced that they had used artificial intelligence to predict patients’ self-reported racial identity from medical images. It is an unexpected, unsettling result.
As chest X-rays of Covid-19 patients began to be published in radiology journals, AI researchers put together an online database of the images and started experimenting with algorithms that could distinguish between them and other X-rays. Early results were astonishingly successful, but disappointment soon followed. The algorithms were responding not to signs of the disease, but to minor technical differences between the two sets of images, which were sourced from different hospitals: such things as the way the images were labelled, or how the patient was positioned in the scanner. It’s a common problem in AI. We often refer to ‘deep’ machine learning because we think of the calculations as being organised in layers and we now use many more layers than we used to, but what is learned is nevertheless superficial.
The Metropolitan Police has announced it is going to use Live Facial Recognition (LFR) in London. The controversial technique involves officers sitting in a public place and filming the people who walk past. Their faces are automatically compared to pictures in a database of wanted criminals and the police are alerted if there is a match. A few days earlier, the New York Times reported that a company called ClearView AI has developed a facial recognition tool that allows law enforcement agencies in the US to match images or video footage with photos from the internet.
Earlier this year, researchers announced a new Artificial Intelligence system, GPT-2, that can finish people’s sentences. The resulting text is relatively coherent but, as the researchers note, far from perfect. Word repetition is one problem; describing the impossible (such as a fire underwater) is another; and sentences are prone to strange topic changes.
In 1990 Hugh Loebner inaugurated the Loebner Prize for Artificial Intelligence, awarded annually to the computer chat program, or ‘chatbot’, that can most convincingly mimic the typed conversation of a human being. Loebner says it’s ‘the first formal instantiation’ of a Turing test; Marvin Minsky, one of the pioneers of AI, has called it ‘obnoxious and stupid’.