More blue on the street
It should be clear: machine learning offers a solution to all kinds of problems. At the same time, many people have reservations about the development of artificial intelligence, of which machine learning is a part. Canadian Nobel Prize winner Geoffrey Hinton, nicknamed ‘the Godfather of AI’, has even there is a 10 to 20 percent chance that artificial intelligence will ‘wipe out’ humanity in thirty years' time. He previously stated that the chance was 10 percent.
Loog lists a number of snags with the much-lauded technology. ‘First of all, machine learning can form and reinforce prejudices’, he explains. ‘The models are fed by data and if that data is coloured, those prejudices are adopted in the programme.’ Loog cites a well-known example of a programme that is supposed to predict where police deployment will be needed. ‘If certain neighbourhoods stand out based on the data entered and more officers are posted there, they will encounter more crimes there, which further reinforces the bias about those neighbourhoods. It raises the question of what we consider a reasonable bias and how we should conduct the discussion about it.’
Paul McCartney
Another concern is the protection of intellectual property. The Beatle Paul McCartney recently the use of texts by writers and musicians as input for AI programmes. Loog chuckled: ‘And when even Paul McCartney starts to worry, you know something is up.’ Then seriously: ‘With the gigantic amounts of data that systems are fed, it will become increasingly difficult to recognise whether or not a text was written by AI.’ By extension, it will also become increasingly difficult to recognise whether videos and images are real or fake.
Energy consumption is a third concern. ‘Of course, not every task is equally taxing, but our questions to programs like ChatGPT or systems that generate images cost a lot of energy.’ Loog doubts whether it is justifiable that we use energy-guzzling AI for applications that are not useful. ‘Do we really need all those image generators?’
What do we actually know?
In addition to the social concerns about machine learning, there are also fundamental questions about the new technology. Loog is addressing these questions in his research. ‘We assume that programmes perform better the more data they are fed. In my research, I ask to what extent that is true. Although we often see that more input helps, there are all kinds of other learning curves possible, where more data is not necessarily better. It shows that despite the ever-increasing role of machine learning in our lives, we still have much to discover about the learning behaviour of systems.’
Stay up to date with our research news?
Follow us on Instagram:
