Self-teaching and cognitive systems
Machine learning is primarily based on two methods: in unsupervised learning, systems search for certain patterns in large databases and filter them out. This is how recommended purchases on Amazon (customers who bought A also bought B) or music recommendations work, for example.
Analyses of the stock market and recognition of credit card fraud also work using that method. The composition of the data is very important for unsupervised learning: if it is incorrect or outdated, it influences the effectiveness and the quality of the artificial intelligence and cognitive systems.
How algorithms are trained
In supervised learning, algorithms are fed with training data. At the same time, the algorithms receive categorising feedback from people accompanying the training. In this context, people almost function as algorithm trainers for the self-teaching systems, for example in the classification of images and news or the weighting of data.
If, for example, a system is to learn how to correctly file publicly accessible information in a company database, it initially needs feedback, monitoring and supervision from a human.
Does every SME need an algorithm trainer?
Most algorithm trainers work behind the scenes. The job is only interesting to an SME if its own AI system is to be specially developed for a specific task. The program for AI or a cognitive system is written by programmers.
However, no specific training is required for the algorithm feedback and training jobs, which are precisely defined in the program in advance. It is often a case of simple sorting tasks, which are carried out by algorithm trainers. The sorting tasks are usually divided into masses of small jobs, carried out by crowdworkers. Amazon Mechanical Turk is a platform that organises this training feedback worldwide