Friday, March 4, 2022

Brian Cantwell Smith, Effing the ineffable: What AI teaches us about what can and cannot be said

Abstract:

A classical story takes the world to consist of objects exemplifying properties and standing in relations. Machine learning and other recent developments in AI support a different view: the world is stupefyingly rich and detailed, far more than can be captured in any finite representation. Representing the world in terms of objects, properties, and relations results from coarse-graining or abstracting over much richer underlying representations which carry more information than can readily be expressed in words. Or at least: more information than can be expressed in words according to classical theories of what words can mean and refer to.

Questions arise. What is out there, and how can we characterize it? What can words mean and refer to? Is the content of human language limited in the ways that classical theories assume? If I report that I laughed, and you grin in response, what has been communicated—and how?

Registration:

Smith starts discussing his notion of registration at about 20:54. He frames it in terms of the long-standing philosophical debate between realism – the world is out there independent of us – and constructivism – "objects and properties are human constructs. Each of those captures something true of the world. See the discussion of Miriam Yevick's work in, Showdown at the AI Corral, or: What kinds of mental structures are constructible by current ML/neural-net methods? [& Miriam Yevick 1975].

About Brian Cantwell Smith:

Brian Cantwell Smith came to Toronto as dean of the Faculty of Information in 2003, after positions at Xerox PARC, Stanford, University of Indiana, and Duke University. He was a founder of the Stanford Center for the Study of Language and Information, first president of Computer Professionals for Social Responsibility, and president of the Society for Philosophy and Psychology.

Smith’s research focuses on the foundations of computation and artificial intelligence. In the 1980s he developed the world’s first reflective programming language (3Lisp). He is the author of On the Origin of Objects and The Promise of Artificial Intelligence: Reckoning and Judgment (MIT Press, 1996 and 2019).

No comments:

Post a Comment