The most popular and comprehensive Open Source ECM platform
Let us know more
Artificial Intelligence: Black Magic or Alchemy — Do We Need to Worry if it Works?
There’s a crisis in Artificial Intelligence research. Despite the headlines and astonishing successes reported by AI, there is an underlying problem with its methods. Research results often cannot be reproduced a second time. Reproducibility has always been a front-and-center requirement before scientific results could be accepted.
The focus of AI research lately has been to create algorithms that produce good results, even when it isn’t at all clear exactly what the underlying algorithms are doing. Ali Rahimi, an AI researcher at Google, criticized work in AI research as relying too heavily of rules of thumb, trial and error and even superstition.
Nicolas Rougier, a computational neuroscientist, said that “I think people outside the field might assume that because we have code, reproducibility is kind of guaranteed. Far from it.”
Rahimi said that “we are building systems that govern healthcare and mediate our civic dialogue. We would influence elections. I would like to live in a society whose systems are built on top of verifiable, rigorous, thorough knowledge, and not on alchemy.”
