It gives me tremendous pleasure (again) in introducing my long-time friend and colleague, Paul Bassett. Paul has written a blog contribution below, which I know you will find extremely thought-provoking. Your responses are of course, solicited.
Paul Bassett is a retired software engineer, author, entrepreneur, and inventor. His invention of Frame Technology (used around the world to automate software development) won him CIP’s Technology Innovation Award. He’s published numerous papers and a book Framing Software Reuse. Paul was a member of IEEE’s Distinguished Visitor Program, and has given keynote addresses, taught computer science at York University, and co-founded several businesses, including two successful software engineering companies. His MSc in artificial intelligence (U. of Toronto) imbued him with a life-long passion for divining the role and future life in the universe.
What is the Name of Our Universe?
“Our universe” means different things in cultures with different creation myths. In my culture, “our universe” usually means the observable universe, which is a sphere with the Earth at its centre; it is the largest volume of matter that can ever affect us. Its radius is 46.6 billion light-years (1 light-year = 9.46 billion km.) and growing at one light-year per year. But the universe created at the “Big Bang” (13.8 billion years ago) surrounds “our universe”, and is unimaginably larger still. Virtually all the matter in the “Big Bang universe” is moving away from us faster than the speed of light, so can never affect us.
In “our universe”, we can see galaxies that can never see each other because any pair of galaxies that are more than 13.8 billion light-years apart have not had enough time since the Big Bang for light to travel from one to the other. So one could say that those galaxies are outside each other’s universes.
Finally, there is the notion of a ‘multiverse’, a universe some cosmologists speculate is spawning universes all the time, just as it spawned our “Big Bang universe”. With so many universes, there is no name for any of them! That said, “our universe” is the de facto name for the one and only universe that matters to us.
Is artificial intelligence intelligent? or is it just machine learning?
There are many ways to define intelligence. Almost all of them involve problem solving proficiency. Problem-solving in turn, is deeply connected to the notion of algorithm, a method for converting inputs to outputs, or in mathematics, computing a function. Every computable function* has a countably infinite number of algorithms that can compute it, each varying greatly in its proficiency – the time and memory it requires to compute its outputs.
All brains and computers work by performing algorithms*. Brains have algorithms whose outputs are algorithms. Normally, brains invent/improve algorithms that computers use, as is. But ever since computers were invented, a goal has been to enable computers to invent/improve their own algorithms, what is commonly referred to as machine learning.
Human intelligence correlates with how quickly one can learn, with the vastness of one’s knowledge, expertise, wisdom, creativity,…This somewhat vague list of attributes all boil down, as I said, to the proficiency of various algorithms. After decades of frustratingly small advances, algorithms have recently been devised that allow simulated, multi-layered neural networks to learn to become much better than any human at quite a few impressive problem domains: from playing games such as checkers, chess, backgammon, poker and go, to medical diagnoses, to language translation, to facial recognition, to driving cars, to big-data pattern recognition, and so on. These machines are said to employ deep learning (“deep” means many layers of simulated neurons, each learning a different aspect of how to solve an overall problem).
Are these machines intelligent? In their domains of expertise, YES. Do they exhibit general intelligence? NO, because they still lack many key algorithms. In particular, no deep learning system today can give reasons for its choices (e.g., why it makes particular chess moves); nor do we know how to enable a machine to be an expert in multiple domains (e.g., chess and medicine). Billions of dollars are being spent on achieving general-purpose AI. And recent rapid progress leaves less and less room for skepticism*.
What is clear now is this: Like humans do, AIs will acquire their intelligence, not from human programmers, but by learning from experience, aided and unaided by teachers. Programmers may give them their initial learning algorithms, but what they learn, including learning to learn better, will emerge from an AI’s interactions with its environments.
*For those who still believe brains can think in ways that machines never can: Almost a century ago computer science pioneer Alan Turing and mathematician Alonzo Church, conjectured that a certain well-defined set contained all and only the functions that matter and energy can ever compute. (This countably infinite set is infinitesimal compared to the uncountably infinite set of all functions.) Since then, many have tried to refute it and failed. More recently, physicist David Deutsch finally proved the conjecture, assuming only that matter and energy obey the laws of quantum mechanics. Thus both brains and (quantum) computers are confined to thinking using algorithms in that set.