Copyright Law’s Impact on Machine Intelligence in the United States and The European Union
Matthew Sag · FIU Law Review · 2020
potential divergences between the law in the United States and the European data, and text data mining provides some of the most important to do so.Text data mining turns texts (or images or audio-video) into data.It is essential for any kind of search engine; it is essential for any kind of statistical analysis of large bodies of content; and it underpins a vast array of machine learning applications.Machi about have already been rendered into data, and that involves the practice of text data mining.Machine learning is great for optimizing to a known objective.Machine learning is great for taking subjective human judgments and replicating them efficiently and at scale.So, one example that I like is that, people at MIT showed a computer system a bunch of photos and how the photos had been retouched by professional photographers.The algorithm is now really good at retouching photos, right?It replicates subjective human judgment, yet in a new circumstance.It is not so great for producing nonfunctional objects.Maybe an AI can make a Rembrandt.It probably can take a photo and then give it a Rembrandt filter, effectively, but that software is not going to be radically creative.It is not going to jump paradigms and go to Greenwich Village and become a jazz musician.What are the perils of AI?Well, we have a black box problem.The problem is that, although we can look under the hood, we will just not understand what we are seeing there.We also have a replication problem in that we cannot really necessarily replicate the algorithm if we do not have This talk was given at the 2019 FIU Law Review