Will Artificial Intelligence and Machine Learning allow robots to steal your work? Discover eight signs that indicate that your job can be easily automated.
In the years to come, many professions will disappear because of automation. Artificial intelligence, and more precisely Machine Learning algorithms, will make it possible to automate many tasks to such an extent that people in the corresponding professions will become useless.
In order to more accurately predict which occupations are at risk of disappearing because of AI, Dr. Tom Mitchell of Carnegie Mellon University has published a study on the likely impacts that Machine Learning will have on different types of occupations.
To this end, this specialist has established a list of 8 particularly machine-learnable tasks. . Occupations in which these tasks play an important role are most likely to disappear in the coming years.
At the end of this study, the researcher concludes that the vast majority of jobs are likely to be affected by the growth of Machine Learning. However, very few jobs are going to disappear altogether.
In most cases, jobs will be transformed by automation in the sense that thehe tasks that will fall to the humans who perform them will no longer be the same.. For example, doctors will benefit from the help of the Machine Learning to make diagnoses. However, they will continue to provide care themselves.
Discover now what are the eight tasks that the Machine Learning can easily automate. If your job is based on one or more of these tasks, consider that you are likely to be unemployed in the near future .
Learn a function that maps well-defined inputs with well-defined outputs
This type of task includes in particular the classificationFor example, labelling images of dog breeds or medical records according to the probability of cancer.
It also includes the predictionThe analysis of a loan application, for example, to predict the likelihood of a future default. If this is what you do on a daily basis, an AI may replace you.
Large datasets containing input-output pairs exist or can be created
If he such data sets exist that are related to your workyou are potentially threatened by the Learning Machine.
Yes, it is, the more data an AI has at its disposal, the easier it is to learn automatically.
The task provides clear feedback with clearly defined objectives and metrics
When a task aims to describe objectives preciselythe Learning Machine is proving to be a powerful tool.
This is the case even when it is not necessarily possible to define the best procedures to follow to achieve these objectives.
The task is not based on reasoning based on common sense.
Machine Learning systems are very good at associating empirical data, but much less for tasks that require long chains of reasoningor common sense or knowledge that the computer does not possess.
For example, a Machine Learning algorithm can excel in a video game that requires fast reactions and provides instant feedback. On the other hand, it will be much less effective in a game where the choice of an optimal action depends on the ability to remember previous events separated in time and knowledge about the world from unknown sources.
The task does not require an explanation of how a decision is made by artificial intelligence.
Neural networks learn how to make decisions by subtly adjusting several million digital weights that are used to interconnect artificial neurons. It is very difficult to explain the reasoning behind such decisionsbecause neural networks do not use the same intermediate abstractions as we do to make choices.
If your business requires detailed explanations of how decisions are madeYou are therefore protected from the Learning Machine for the moment. For example, computers can now diagnose certain types of cancer with the same efficiency as human experts, but they are unable to explain how they arrived at this diagnosis.
Errors are tolerated and there is no need to provide optimal solutions.
The majority of Machine Learning algorithms derive their solutions from statistics or probabilities. It is therefore generally impossible to train them to achieve 100% accuracy.
Even the best object or speech recognition systems can make mistakes. This is the reason why jobs where errors are not tolerated cannot yet be automated.
The learned function or phenomenon should not change rapidly over time.
As a general rule, Machine Learning algorithms do not work well that when the distribution of future test examples is similar to the distribution of test examples.
For example, mailbox spam filters work well because the acquisition rate of new sample emails is high compared to the speed at which spam changes. If the functions or phenomena that the algorithm learns change quickly, the Learning Machine quickly becomes helpless.
The job requires neither mobility nor physical ability
Fortunately for humans, Robots still lack agility compared to humans. when performing physical manipulations in unstructured environments.
Therefore, jobs that require dexterity or mobility cannot yet be automated by the Learning Machine.