Covid-19 is making AI sick
Algorithms have fallen victim to the pandemic too. Amazon didn’t anticipate the panic buying of bottled water and toilet paper, which upset normal models and forecasts – and disrupted the supply chain for some time.
AI extracts correlations from enormous quantities of disparate data and learns from experience. But it struggles in the face of major, unexpected events. During the Covid-19 crisis, AI faced dramatic shifts in customer habits. For example, in addition to panic–buying household essentials, customers switched from paying cash to using bank cards. Fraud–detection systems were thrown by this sudden rise in mobile payments, so changes were necessary to avoid blocking legitimate transactions. Another example: mask wearing makes technology based on facial recognition problematic, even completely useless.
Humans don’t need previous experience to make causal links. We know that a car that falls into water will eventually sink. AI isn’t able to do this yet, so the challenge is to devise “neurosymbolic” AI that combines a capacity for processing data with an ability to make causal links. Without that, it’s up to humans to play the role of watchdog.
To go further
by Ben Dickson (Gizmodo, 29 July 2020).
© Copyright Business Digest - All rights reserved
