AI training is more human than artificial.

In a technology report devoted to China, The Economist includes an article on the human input to training AI. It is not just the vast amounts of data that China possesses that gives it a head start in machine learning. It is also the access to cheap labour. Once the data has been harvested, then more often than not it needs to be labelled. Without the labelling, most of the algorithms would not be able to be optimised. The infrastructure sitting behind the firms designing and building the algorithms includes other firms that collect and label the data. These so-called data factories are rather like car parts suppliers to the big car manufacturers. To give you an idea of the scale, one of the biggest data factories in China employs some 300,000 data labellers. These are human beings who attach labels to each piece of data. It’s not just about having the best algorithm, it’s also about employing enough humans to train them.

Link to article:

You may also like to browse other AI articles: