Artificial Intelligence will not make everyone unemployed Journalists say that artificial intelligence will soon replace most workers and therefore... bad things will happen. I found an empirical refutation that can be inspected right now.
In previous industrial revolutions, new technology created new
low-paying professions offering employment to
low-paid manual workers who lost their jobs to machines. (Highly paid workers typically did not lose jobs.) Exactly the same thing is happening today; here is how to see that.
Machine learning (now mistakenly called AI) requires a lot of clean, labeled data for model training. For example, if we want to build a machine that recognizes faces on photos, we first need to have a lot of photos where faces are already recognized. Somebody must manually scan thousands of photos and manually annotate them, e.g. drawing a rectangle or an oval around the face, writing down the name of the person and so on. Another example is labeling photographs as different types of landscapes, house interiors, technology scenes, people scenes etc. in order to build an automatic recognition of such photos. For speech recognition training data, one must manually label the words spoken, identify the speaker, etc. This kind of job can only be done by humans, and it's a
low-paying job. Amazon in its
Amazon Mechanical Turk service pioneered a marketplace that organizes people who want to work on these jobs as well as people who require labeled data.
This is precisely an example of a novel job created by new technology. In some cases, this job can be outsourced to other countries, but in many cases it can't be outsourced because successful performance requires cultural information - knowledge of language idioms, understanding of local customs and so on. Neural networks (the basis of new methods for speech and image recognition) require much more training data than simpler machine learning methods. Therefore, it appears that the neo-luddites are being wrong again.