Special thanks to Leslie Valiant for answering 3 questions about his recently featured book – Probably Approximately Correct: Nature’s Algorithms for Learning and Prospering in a Complex World
Leslie Valiant is the T. Jefferson Coolidge Professor of Computer Science and Applied Mathematics at Harvard University. He is a Fellow of the Royal Society and a member of the National Academy of Sciences. He is a winner of the Nevanlinna Prize from the International Mathematical Union, and the Turing Award, known as the Nobel of computing. – From Leslie’s Homepage
Leslie’s Homepage: http://www.probablyapproximatelycorrect.com
#1 – What was the impetus for Probably Approximately Correct?
I thought there was a good story to tell. The theory of PAC learning is well known within certain academic communities. But I believe that this theory has some almost “self-evident” relevance to broader aspects of human affairs. Hence the impetus behind the book is the story that there has emerged from computer science a vantage point from which issues of much interest to humans might be studied scientifically, perhaps for the first time.
#2 – Can you briefly talk ‘ecorithms’ at the centre of your book?
The basic idea is that there is much commonality among the various means of information acquisition, whether by a biological species in the course of evolution, or by an individual organism in the course of one life. By information acquisition I include essentially everything that determines a species or individual – hence studying this commonality could be a key to understanding life at some generality. The idea of ecorithms seeks to capture this commonality as a general means of learning from a world that is neither malicious nor benevolent. Learning algorithms in the sense of machine learning offer prime examples. But I allow a broader range of phenomena not previously regarded as machine learning. Darwinian evolution is one. Algorithms that combine learning and reasoning is another.
#3 – Is your model different to ideas such as fuzzy logic?
The ecorithmic view derives from looking at natural phenomena. The argument is that since individuals and species are determined by information they derive from outside themselves, the information acquisition process is the key to understanding their nature. Hence the book is about the possibilities and limitations of this information acquisition process. Much can be said about these possibilities and limitations that is informative. It is my impression that the many alternative approaches to formalizing the processes of intelligence are not so grounded in natural phenomena. In ecorithms a primary requirement is that no process be posited for which a plausible method of learning cannot be described. In the field of artificial intelligence this has not been a traditional requirement.
[Image Credit: Larry Bercow – http://www.probablyapproximatelycorrect.com/?p=3 ]