7/11/2019

Quantum Supremacy and the Threat of AI Supremacy

Filed under: — Bravus @ 10:51 am

OK, it’s not the zingiest title in the Hogwarts world… 😉

Google recently claimed (a claim disputed by others, including IBM) to have achieved ‘quantum supremacy’. This is a term, popularised by John Preskill, for the ability of a quantum computer to solve a computing problem that would be impossible for a conventional computer within its lifetime.

Google claimed that their ‘Sycamore’ quantum computer, with 54 ‘qubits’ (quantum bits) solved a problem in 3 minutes 20 seconds that would take a conventional computer 10,000 years.

The details are contested, but a working quantum computer (IBM also has one, which for some reason has 53 qubits) is a huge step forward.

Conventional computers use bits that can take values represented by 0 or 1. Like Schrodinger’s Cat, though, a qubit doesn’t have to choose until it’s ‘observed’. It allows quantum computers to ‘explore a problem space’ more efficiently. (This is a very lay explanation of quantum computing in a paragraph that has everyone who actually knows anything about it cringing!)

Dr Stuart Russell has recently published a book called ‘Human Compatible: Artificial Intelligence and the Problem of Control’. The book is well worth reading, but the Probably Science podcast interviewed him a week or so ago, and that’s how I learned about the book and his arguments.

I tend to be an optimist in these kinds of issues, but Dr Russell poses a simple but hard question: “So, you’re spending hundreds of millions of dollars a year to create entities much smarter than you. How do you plan to keep control of the world once you succeed?

He has some great examples of the unintended consequences of even the dumb AI we currently have – helping corporations make profits while making society worse – but the extra ‘horsepower’ offered by quantum computing just makes his key question a lot more urgent.