Lecture notes, notebooks for a course on the interaction between NLP, Ethics and Culture.
Recent technological developments have changed the way culture is consumed by most people: the Internet opened masses of data in searchable and immediately accessible multimedia form, smart mobile devices made this data available everywhere at any time, and social media platforms have created new ways to connect to others, to publish and to consume content within increasingly complex networks of acquaintances. Most recently, a new wave of technology based on Artificial Intelligence and Natural Language Processing have made possible the generation of new content, both text and images, in context, that are difficult to distinguish from human-generated artifacts.
These new text and image generators are trained on the vast amounts of content made available online by social media, mobile and internet products using machine learning techniques: summarizing, memorizing, compressing the masses of online content into generative devices.
In this course, we critically study the impact and challenges that these new text and image generator devices can have on society, culture, and human behavior. The course is intended for non-technical graduate students in the humanities and sciences.
We build an operational understanding of the technological capabilities through a historical review of the techniques used to build these generators and a non-technical presentation of the computational operations upon which they rely. We organize hands-on experimentation using recent text and image generators in different domains to understand what the strengths and weaknesses of these models are, and how to control their generative power. As users, we review computational techniques that can help adapt these generators to specific domains or corpora, and how this process modifies their output. We compare how well these generators behave in different languages (specifically English vs. Hebrew).
Building on the understanding of the capabilities of these emerging AI generators, we will address their ethical implications and the risks they create. Do these generators promise increased accessibility of technical expertise - like a new era of the Gutenberg printing press developing a new active, dynamic, evolving cultural memory? Do they threaten to propagate biases, stereotypes, uniformize cultural artifacts, reduce diversity? We will rely on readings in computational ethics that develop mathematical notions of bias, fairness, equity, diversity and compare them with humanistic views of the same concepts.
We will finally explore how responsible citizens can contribute to mitigate the risks created by this emerging technology, considering the economic incentives represented by large corporations developing and controlling the technology, the role of ill-intended actors, and discuss the balance between creativity, regulation, and society welfare.
The work will involve readings of technical articles with adaptation to non-technical readers, hands-on labs experimentation with online tools, and writing essays about specific ethical challenges posed by the emergence of new AI generators.