
Christian Chavanel, UIC Railway System Director
According to you, what means Artificial Intelligence?
The term ‘Artificial Intelligence’ (AI) is a suitcase term and is not easy to define. There is nothing artificial about it. A.I. is made by humans, intended to behave by humans and, ultimately, to impact humans lives and human society. Massive computing power, which only performs requested routine tasks and is controlled at every step by software programmers using a classical analytical approach, is not AI. A classical algorithm, however complicated, which does not deviate from the problem-solving method programmed by a software programmer, is not capable of learning. It is not AI. A large part of artificial intelligence today is based around automatic learning.
Today there are three important subtypes for IA : Machine Learning, Natural Language Processing, and Robotics.
For machine Learning, what are the key points to be focused on?
The typical Machine Learning process requires both data and algorithms. A three-step process maximises the chances of learning success (Towards Data Science, 2018):
• Training: “A subset of real data is provided to the data scientist. The data includes a sufficient number of positive and negative examples to allow any potential algorithm to learn. The data scientist experiments with a number of algorithms before deciding on those which best fit the training data.”
• Validation set: “The data scientist will run the chosen algorithms on the validation set and measure the error. The algorithm that produces the least error is considered the best. It is possible that even the best algorithm can overfit or underfit the data, producing a level of error which is unacceptable.”
• Testing: “To obtain an accurate and reliable measure of error, a third set of data should be used, known as the test set. The algorithm is run on the test set and the error is calculated.” During this learning process, data scientists must choose the best algorithm from a considerable number of them.
The quality of the data chosen is also crucial.
Eight biases must be avoided : propagating the current state, training on the wrong thing, under-representing populations, faulty interpretation, cognitive biases, analytics bias, confirmation bias, and outlier bias (Search Business Analytics, 2020).