AI in recruitment. Birth of artificial intelligence in Element ATS

As I mentioned, the development of artificial intelligence is like teaching a child. In both cases, teaching is a long-term process. In the case of artificial intelligence, the purpose of which is to parse CVs, teaching consists of providing a large number (counted in thousands) of specially prepared CVs. These specially prepared CVs are documents, created in a separate module of Element ATS, in which we tag individual parts of the content, and then assign appropriate meaning to these parts. This process of preparing CV goes like this: we read the content of the CV. When we come across the word Maciej, I tag this word as a name. If I find a date, then I tag this date as, e.g., starting a specific position in a particular company. This way, we go through the entire CV and tag each phrase in it. The process is very similar to completing a Linkedin profile.

It may seem simple at first sight. In fact, it is full of traps, ambiguities, and the need to make compromises. One of the permanent elements of the development of artificial intelligence in Element ATS are regular sessions during which we analyze the progress of work. During these sessions, we also try to solve problematic situations related to data tagging. Here are some examples:

• Does the phrase “2017 – Coca Cola” found in the CV mean that the candidate worked for this company only in 2017?
• If the candidate wrote that he works as “Freelancer for Coca Cola”, should our artificial intelligence assume that the place of employment is Freelancer or Coca Cola?
• If the candidate has given the “Chemistry Technician” in his education, then what was his major?
• What if the candidate’s skill level is described with stars? Should AI count and interpret the stars?
• What is the level of language skills according to the CEFR standard (from A1 to C2) if the resume states “English – understanding very good, writing good”?

The more CVs, the more such questions we need to answer. Human creativity is limitless, and it is impossible to create a single algorithm that will correctly parse every type of CV. This is why artificial intelligence is used to solve such difficult problems. But to teach AI to cope with the limitless diversity of human creativity, we need to provide it with many thousands of carefully prepared examples. These examples are meant to show AI how human intelligence solves such problems.

How does artificial intelligence learn from these examples? Well, for AI this is just a matter of statistics. If the algorithm receives 10,000 examples, in which two words in the first two lines of the document will be tagged as first and last name, then in document number 10,001 it will find a similar pair of words on its own and correctly indicate which of them is the first and which is the last name. It is important to understand that this is not a matter of simply “remembering” the first and last names. Artificial intelligence does not remember the meanings of words, but each time it decides on the meaning of a word, assessing the context of its use and its location throughout the document.