ARTIFICIAL INTELLIGENCE: THE BEST ALLY AGAINST "HUMAN ERROR" WITH OUR DATA SCIENTIST ANITA SANCHO
Share it in
For this interview, we decided to speak with Ana María Sancho, Chief Artificial Intelligence Officer of Kriptos. Anita loves math since childhood, which is why she decided to study a university degree in Economics, but discovered that her passion was elsewhere. Consequently, she decided to study Big Data and Business Analytics at IE in Spain.
We talked about the importance of AI in the prevention of human errors in companies in order to reduce the leakage of information caused by employees or internal collaborators.
Before starting, it is important to identify the type of 'human errors' that can cause information leakage and damage the operations, reputation and image of the companies.
There are two types of human errors; intentional and unintentional. If you want to know more about this topic, we invite you to review this blog "Employees: Are yours secure?”
According to Sancho, in order to add more value to administrative controls, there are certain AI algorithms that have the capacity to facilitate the recognition of patterns. Recognizing patterns such as unusual behaviors can be very valuable to companies as it enables them to determine the probability of information leaking out.
Here are some of the most important highlights of our interview:
Are there several types of AI?
Yes; the artificial intelligence that executes a certain action based on the inference of a model is called Narrow AI.
There are two more types:
-Board AI: A system in which several actions are executed by different inferences of different models. For example: self-driving cars.
-General AI: An intelligence that is not only capable of learning, but also of reasoning. So far, it only exists in concept.
If that is so Anita, how can Artificial Intelligence enhance cybersecurity tools?
Finding patterns of anomalies, that is, detecting irregular behavior of users and documents and raising an alert of possible information leakage. In addition, you could find patterns that can locate confidential documents. Through algorithms you can detect which characteristics make the information confidential, with great precision.
Anita says that the best way to find patterns is by applying Machine Learningalgorithms, thus optimizing time and costs, and ensuring greater accuracy.
What are machine learning supervised algorithms?
They are systems that simulate human learning in order to obtain the causality of an event.
If we take for example the case of a person that wants to understand what the features that make a document confidential are, he or she will have to read and reread several documents, previously known to be confidential. This will allow the person to find the sufficient insights and patterns that will allow him or her to recognize a confidential document.
When a computer does this analysis process, we call it machine learning, that is, iteration systems that are processed until they find a pattern. This in turn translates into a logical mathematical formula that will lead us to a result or inference. Machine learning takes less time than human learning and through it you will find every single possible pattern. Because of this last machine learning is precise.
Machine learning algorithms are composed of decision trees, linear regressions, logistic regressions, support vector classifiers, perceptions and neural networks.
What makes these found patterns become executable?
Once the algorithm is trained, we obtain a formula, that is, a set of parameters that indicate the level of confidentiality of a document: a model. Depending on the data consumed by this model as an input, it will make an inference, that is, it will give a verdict with regards to its kind, class or type.
Connecting a precise and trained model to a software that is capable of executing actions based on its decisions is an artificial intelligence method that helps minimize human errors, time and costs.
How do you determine success or an acceptable level of assertiveness of these algorithms?
The level of accuracy depends on the quality of the data with which the model was trained. It will also depend on the type of data you are using. Imagine for a moment that your company belongs to the banking industry. If you try to classify information with a model or algorithm that was trained using data from another industry then the precision will suffer. Remember that machine learning is about recognizing patterns.
An interesting way in which you can increase the precision of an algorithm is by feeding it a dictionary of related words. Using the example mentioned before, this would be very helpful given that the algorithm was trained but not in the banking industry.
Do you have any notes or final clarification that you would like to mention?
The only thing would be that I deeply believe that the AI will serve as a prevention mechanism of human errors. I see it as a tool that IT Managers can use to anticipate risks. It will allow them to take measures or suggested controls on time.
The rate with which cybersecurity attacks is growing is frightening. It is of the utmost importance to use artificial intelligence as an ally. It is a great defense tool, an anomaly detector and a confidential information locator.