AI training involves significant human input.

This is a timely article on a downside to the training aspect of machine learning – the more complex the challenge, the more intensive and costly the training. In that connection, perhaps it should be no surprise that the driverless car industry is spending considerable time and money on fleets of labellers (humans), who review video footage shot by car-mounted cameras and mark it up to identify all the objects perceived in the car’s environment. How else does the driverless car learn? Some hope that deep learning will reduce the intensity of such training, or at least the human element. Some hope that in the long run unsupervised learning is the answer. There is also talk of a master algorithm that drives the human brain; maybe that becomes the answer if science can discover it. It might be, however, that the answer will lie in a combination of both architectures and learning techniques. Human behaviour is, after all, not solely generated by a single pattern-recognition capability in the neocortex.

Link to article:

You may also like to browse other AI articles: