Here are the slides of my SASE lecture on July, 20th 2020 on AI.
SUMMARY: The rising visibility of artificial intelligence (AI) has triggered as many hopes as criticisms. In science only, the machine learning techniques are praised for their flexibility, and recent success in translation, prediction or pattern recognition (image or sound) are often hailed. But their limits have also, as often, named: predictive rather than explanatory, lacking strong mathematical foundation, or dependent on a high volume of data, AI is described by some researchers as largely inferior to standard quantitative methods.
The goal of this course is reflect on what AI can do for, but also to, the (social) sciences. Through a comparison with the now « classic methods » and empirical examples, we will present the gist of this approach. Largely non-technical, the presentation defends the idea that due to a radically different approach to quantification, AI is unlikely to root out the now termed « classic » quantitative methods. But the use of AI could serve our disciplines in several ways. One is the (welcome) addition of complexity to our quantitative approaches. Another, more important one, is the production of new data from heterogeneous sources (images, sounds, digital traces).