Covid-19: curse or blessing for Artificial Intelligence?
The corona crisis disrupts artificial intelligence because it uses data that completely changed during the crisis. The sudden changes in our consumption and mobility behavior make it difficult for AI to make well-founded predictions.
Automated inventory management systems, marketing algorithms, fraud detection systems… they were all struggling.
The mobility landscape has also been shaken up drastically. Traffic data predictions no longer match reality at all, and irrational decisions often influence people’s mobility.
Waze and Google maps are having a hard time making estimates now that a lot of people are staying at home, and public transport has to deal with changing user habits. Data News tried to find out what the crisis means for AI.
Avoid blind trust
According to Jonathan Berte, CEO, and founder of AI company Robovision, algorithms are not designed for such a ‘black swan event’. “You could mine data from black swans from the past, but such a swan is always fundamentally different,” says Berte in Data News.
“In itself, machine-learning models are built to respond to those changes. But if the input differs too much from the data on which they were trained, the system loses its accuracy.”
Jan Van de Poel, founder of the AI marketplace SeeMe.ai, advocates the intelligent use of AI. “Every AI model requires a continuous cycle of improvement. It is not a matter of fire and forgets, but of timely intervention where necessary. The worst thing you can do is blindly trust artificial intelligence.”
AI continues to be used, but how hard should we trust its accuracy and soundness? “Nobody was prepared for a crisis of this magnitude, but a man of flesh and blood will always be more agile than an algorithm,” says Berte. “Either way, a self-learning system will still need human control to some extent.”
For Berte, it is a must that organizations that are serious about AI have a dedicated team that consistently tests the data against reality. “In addition, this is the perfect opportunity for data scientists to re-activate the buttons, and explore the possibilities of other machine-learning techniques that may be somewhat unknown to some.”
These techniques can also guarantee reliability in crisis situations. Berte talks, among other things, about ‘few-shot learning’. This technique enables systems to become smarter without a lot of input. Today, systems can already be trained to make accurate predictions with a limited amount of data.
Transfer learning is similar, reducing the long training time of systems, and the number of data points required. “Actually, it’s a pity that such a crisis had to take place to realize that artificial intelligence is not ‘one size fits all’ data,” says Berte.
“What we sometimes underestimate in the speed and effectiveness of smart algorithms is the explainability,” adds Van de Poel. “In many models, including image recognition, a system can explain why it makes a certain prediction. This is important in order to be able to build better and more robust models more quickly.
AI to fight Covid-19
But the importance of speed also indicates the AI-controlled applications to combat the virus itself. In Belgium alone, several pieces of innovation have shot up in various fields in recent weeks. For example, Robovision uses its AI software for business applications to monitor lung scans for Covid-19 in the future.
West-Flemish company, Aptus, developed its ‘Walkthrough display’. It is able to contactless measure body temperature in seconds, for example, at the entrance of a company.
“We, too, are aware that this tool requires permanent monitoring,” says founder Alexander Vanwynsberghe, managing partner at Aptus. “We don’t have to be a medical company to keep our finger on the pulse of the evolution of infections.”
Only time will tell if the corona crisis will grow into an opportunity to let technology support us even better as a society.