Machines and computers are advancing at a very high pace in recent years, particularly with the evolution of artificial intelligence. The job market is rapidly changing as a result of the evolving technologies. More and more jobs that were previously purely for humans are currently being managed by machines. On the other hand, there are new jobs that are emerging which have never existed before. Other jobs have continued while the details of their job descriptions and the working style are changing.
In earlier times of telephone networks, humans were responsible for directing the calls according to the caller. This process was afterwards completely automated and the job disappeared. Another example was the control of the railroads; humans were responsible for configuring the rails manually to direct the trains. The process was inefficient leading to many accidents due to human errors. Accordingly, the process was automated with human supervision to avoid errors as much as possible.
These are very basic examples for jobs that disappeared based on the technology available. However, it seems that machines -with machine learning and artificial intelligence- are doing jobs that no one had thought of before. The University of Oxford published a study in 2013 suggesting that about half of the current jobs may be taken over by machines. Surely, this is far more than what commonly the public would expect.
Machine learning started doing simple jobs in 1990s including mail sorting by reading human handwritten zip codes and assisting in financial planning. Machines have already started playing the role of personal assistants using programs as Siri from Apple and Google Assistant from Google. These programs can understand human voice orders and perform tasks that were normally done by humans as making reservations at restaurants and hotels, booking tickets, sending messages and even calling someone. The programs can interact in a manner very similar to humans and they are increasingly more accurate and more responsive as their database grows larger.
Unlike human personal assistants who are normally available for senior managers and wealthy people only, virtual personal assistants are available at almost completely no cost. They are just an integrated part of the mobile or device you would buy. The German automotive company, BMW, has recently announced that its new cars would be supported by a virtual assistant that can understand not only the voice commands but also the gestures and facial expressions. The car will also be able to interact with its surroundings. This functionality of human interactions is not limited to personal assistants but it is currently being used in customer service call centers replacing many human jobs.
This is not yet the interesting part about what machines can do. Machine learning seems to be able to do a lot more. In a competition organized by Kaggle, a company specialised in machine learning, participants were able to develop a program that can be used to grade essays written by high school students. The winning program was able to give similar grades to that given by human teachers. In another version of the same competition, programers were able to develop algorithms that can diagnose diabetic retinopathy using a simple scan of the eye or even a photo. Again, the program was able to give diagnosis similar to physicians.
The advantage of these programs in such tasks is how massively scalable the work can be. A computer can thoroughly read and grade millions of essays in just few minutes which would normally take ages from humans. Limitations for this lie in the basics of machine learning. Machines learn by comparing a huge number of databases, and then concluding their own method or algorithm based on the similarities and differences. They apply this while doing their job. The problem happens when the database used is not normally distributed or not representative enough.
There have been even some trials to use computer with artificial intelligence to produce artistic works. This application adopts a hypothesis that even though art should be completely a creative task, it should also fulfill some sort of underlying pattern perceived by humans as artistic. Based on this hypothesis, computers can analyze huge numbers of artistic works, deduce their own rules and produce work that is perceived by humans as a piece of art. In December 2018, a print produced by a computer using Generative Adversarial Network was sold at a major auction house for $432,000. Critics said that this incident was not really significant as it seems. They say that the value for the work was actually for the fact that it was generated by artificial intelligence not because it was in itself original or interesting.
The current significant limitation that set the current boundaries for artificial intelligence is the need for a database of similar previous cases. Machines learn by following patterns in large databases and then deduce the rules that should be followed in similar cases. Accordingly, machines are not yet able to deal with completely new situations. For example, if we used a program for diagnosis of a variety of diseases, it would be unexpected for the machine to recognise efficiently newly emerging diseases although this happens regularly in nature.
However, the previous case does not also seem strict and the real boundaries for machine learning may be unclear. In a recent experiment by Facebook artificial intelligence team, two chatbots were trained to interact with each other. The experiment was terminated when the chatbots decided independently to develop a new language of communication that is not understandable by humans. This step was clearly beyond the current boundaries plotted for artificial intelligence. In fact, independent decisions like this one are the reason why artificial intelligence is still not used for military applications.
Powered by Froala Editor