This final short performance-lecture is on the notion of the “uncanny” and dreamlike, the most frequently used term to describe images made through machine learning and AI, along with references to the surreal. In a provocative piece from 2018, Hal Foster asked, “How does machine vision affect our usual ideas about representation, meaning, and critique?” In my new book AI Art, Machine Learning, and the Stakes for Art Criticism, I propose we look at machine learning-made images more closely and with a different set of evaluative tools. We might use them as testing grounds for more evocative language as to how it feels to live in an algorithmic landscape. We might also critically re-examine this notion of a machine-made image as dreamlike, or like a hallucination, find an opportunity for broader language that reflects what they are, and find the obvious gaps in using art historical frameworks around the moving image, digital art, and media art. What do we mean when we call the visual output of algorithms “uncanny” or “surreal”? What counts, as Hal Foster writes, as “requisite competence in this strange world,” where cultural critics have not sufficiently tackled the “central fact of the algorithmic scripting of information”? What more might we say about the “third nature,” the shadowland they inhabit? And what might we understand as surrealism in an algorithmic age, and how might we mine, delve into, and close-read uncanny images to understand what futures are at hand? This will be read alongside visual research and work I am producing in collaboration with the artist Casey Reas, a co-founder of Processing, professor at UCLA Design Media Arts (DMA), and author of a book I edited, Making Pictures with Generative Adversarial Networks (Anteism Books, 2019). I look carefully at the last three years of neural network imagery and datasets he’s gathered with his research team, with an eye on answering the questions above.