Does creating and training AI systems require more human effort than it used to? Is there a trend?
Since AI is changing in leaps and bounds, I’m not sure comparing the past with the present in this way is very useful. I think it’s safe to say the trend in AI systems is toward more complexity, which requires more human effort. We have not yet arrived at the promised technological Singularity, where AI could build itself.
It’s important to look at the trend of automation in general, and in that context it is conceivable that AI may eventually be involved in the automation of everything that humans do. Collective intelligence as seen today can be a very powerful thing, and theoretically once AI starts to learn from collective intelligence then the logical conclusion is that it would outstrip individual humans.
It is not inconceivable that AI could eventually predict what individual humans might do before they themselves know. Be aware that training in machine learning, and training an AI system may not necessarily be the same thing. Training in machine learning can involve extensive pre-processing of data into metadata often manually by people, for instance via Amazon Mechanical Turk.