Quantum Algorithms and Their Discontents

I recently read Seth Lloyd’s A Turing Test for Free Will — conveniently related to the subject of the blog’s last piece, and absolutely engrossing. It’s short, yet it makes a wonderful nuance in the debate over determinism, arguing that predictable functions can still have unpredictable outcomes, known as “free will functions.”

I had thought that the world only needed more funding, organized effort, and goodwill to solve its biggest threats concerning all of humanity, from molecular interactions in fatal diseases to accessible, accurate weather prediction for farmers. But therein lies the rub: to be able to tackle large-scale problems, we must be able to analyze all the data points associated to find meaningful recourses in our efforts. Call it Silicon Valley marketing, but data analysis is important, and fast ways of understanding that data could be the key to faster solution implementation.

Classical computers can’t solve almost all of these complex problems in a reasonable amount of time — the time it takes for algorithms to finish increases exponentially with the size of the dataset, and approximations can run amok.

Quantum computers, using qubits instead of bits as the basic units of data storage, could theoretically cut the time needed to process algorithms by leaps and bounds. A problem that could take years could be reduced to seconds, and policymakers can more quickly integrate data into their proposals, figuring out more efficiently how to allocate resources or support community welfare. My idealism powers my studies in quantum computing.

But idealism eventually has to run into reality, and the truth can be disheartening. We are years away from any semblance of commercially viable quantum computers, much less those that can readily implement machine learning and neural networks. Heavy skepticism, albeit necessary for scientific rigor, accompanies every new announced breakthrough, and —> Read More

facebooktwittergoogle_plusredditpinterestlinkedinmail